diff --git a/docs/TASKS.md b/docs/TASKS.md index cfb5ae2c..e4dcd875 100644 --- a/docs/TASKS.md +++ b/docs/TASKS.md @@ -83,7 +83,7 @@ Languages are **derived in code** — there is no `languages` field to set in th YAML. A task resolves to a canonical [`lang_Script`](https://en.wikipedia.org/wiki/IETF_language_tag) code (e.g. `deu_Latn`) from, in order: -1. **`flores200:src-tgt` task names** → the non-English side(s) of the pair. +1. **`flores200:src-tgt` (and `flores200_src-tgt_bpb`) task names** → the non-English side(s) of the pair. 2. **The `{lang}` value** substituted into a `valid_langs` template (preferred for new multilingual groups — see the template expansion above). 3. **The task's `subset`** (e.g. `de`, `german`, `deu_Latn` all fold to diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_belebele_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_belebele_bpb_template_yaml new file mode 100644 index 00000000..f8494d0a --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_belebele_bpb_template_yaml @@ -0,0 +1,17 @@ +# Belebele in cloze form: the continuation is the answer text rather than the +# letter "A".."D", so BPB measures comprehension instead of format. +# Per-language tasks live in belebele/. +include: _bpb_template_yaml +tag: + - bpb + - bpb_multilingual + - belebele_bpb +dataset_path: facebook/belebele +test_split: test +fewshot_split: test +fewshot_config: + sampler: first_n +doc_to_text: "{{flores_passage}}\nQuestion: {{question.strip()}}\nAnswer:" +doc_to_target: "{{['1', '2', '3', '4'].index(correct_answer_num)}}" +doc_to_choice: "{{[mc_answer1, mc_answer2, mc_answer3, mc_answer4]}}" +process_results: !function utils.process_results_belebele diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_bpb_template_yaml new file mode 100644 index 00000000..81c14789 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_bpb_template_yaml @@ -0,0 +1,22 @@ +# Shared config for the BPB (bits-per-byte) task variants. +# +# Each task keeps its upstream prompt format and dataset so that BPB is +# measured on exactly the same documents as the accuracy-based task, but +# overrides process_results (see utils.py) so that the gold continuation's +# (loglikelihood, byte_length) pair is emitted for the bits_per_byte +# aggregation. acc/acc_norm come along for free. +tag: + - bpb +output_type: multiple_choice +metric_list: + - metric: bits_per_byte + aggregation: bits_per_byte + higher_is_better: false + - metric: acc + aggregation: mean + higher_is_better: true + - metric: acc_norm + aggregation: mean + higher_is_better: true +metadata: + version: 1.0 diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_bpb_template_yaml new file mode 100644 index 00000000..819f6b23 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_bpb_template_yaml @@ -0,0 +1,21 @@ +# FLORES-200: BPB of the reference translation given the source sentence, +# using the same "DE: EN:" prompt shape as lighteval's flores200 CF +# formulation. Included by the two direction templates below; per-pair tasks +# live in flores200/. +include: _bpb_template_yaml +tag: + - bpb + - bpb_multilingual + - flores200_bpb +dataset_path: facebook/flores +test_split: devtest +fewshot_split: dev +fewshot_config: + sampler: first_n +output_type: loglikelihood +# doc_to_target supplies the leading space itself (see utils.py). +target_delimiter: "" +metric_list: + - metric: bits_per_byte + aggregation: bits_per_byte + higher_is_better: false diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_eng_xx_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_eng_xx_bpb_template_yaml new file mode 100644 index 00000000..5845c54d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_eng_xx_bpb_template_yaml @@ -0,0 +1,3 @@ +include: _flores200_bpb_template_yaml +doc_to_target: !function utils.doc_to_target_flores_eng_xx +process_results: !function utils.process_results_flores_eng_xx diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_xx_eng_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_xx_eng_bpb_template_yaml new file mode 100644 index 00000000..a7bea989 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_flores200_xx_eng_bpb_template_yaml @@ -0,0 +1,3 @@ +include: _flores200_bpb_template_yaml +doc_to_target: !function utils.doc_to_target_flores_xx_eng +process_results: !function utils.process_results_flores_xx_eng diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_global_mmlu_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_global_mmlu_bpb_template_yaml new file mode 100644 index 00000000..5fc3ef3f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_global_mmlu_bpb_template_yaml @@ -0,0 +1,18 @@ +# Global-MMLU in cloze form, like _mmlu_stem_bpb_template_yaml. Each language +# is a single task over the whole test split (all 57 subjects), so BPB is one +# byte-weighted corpus aggregate per language. Per-language tasks live in +# global_mmlu/. +include: _bpb_template_yaml +tag: + - bpb + - bpb_multilingual + - global_mmlu_bpb +dataset_path: CohereForAI/Global-MMLU +test_split: test +fewshot_split: dev +fewshot_config: + sampler: first_n +doc_to_text: "Question: {{question.strip()}}\nAnswer:" +doc_to_target: "{{['A', 'B', 'C', 'D'].index(answer.strip())}}" +doc_to_choice: "{{[option_a, option_b, option_c, option_d]}}" +process_results: !function utils.process_results_global_mmlu diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_include_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_include_bpb_template_yaml new file mode 100644 index 00000000..43d70ec5 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_include_bpb_template_yaml @@ -0,0 +1,14 @@ +# INCLUDE in cloze form, like _mmlu_stem_bpb_template_yaml. Each language is a +# single task over the whole test split (all domains). Per-language tasks live +# in include/. +include: _bpb_template_yaml +tag: + - bpb + - bpb_multilingual + - include_bpb +dataset_path: CohereForAI/include-base-44 +test_split: test +doc_to_text: "Question: {{question.strip()}}\nAnswer:" +doc_to_target: answer +doc_to_choice: "{{[option_a, option_b, option_c, option_d]}}" +process_results: !function utils.process_results_include diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_mgsm_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_mgsm_bpb_template_yaml new file mode 100644 index 00000000..781f2533 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_mgsm_bpb_template_yaml @@ -0,0 +1,20 @@ +# MGSM: BPB of the final answer number after the upstream lm-eval +# mgsm_direct_ prompt (the test split has no worked solutions). +# Per-language tasks live in mgsm/ and carry the localized prompt. +include: _bpb_template_yaml +tag: + - bpb + - bpb_multilingual + - mgsm_bpb +dataset_path: jbross-ibm-research/mgsm +training_split: train +test_split: test +output_type: loglikelihood +# doc_to_target supplies the leading space itself (see utils.py). +target_delimiter: "" +doc_to_target: !function utils.doc_to_target_mgsm +process_results: !function utils.process_results_mgsm +metric_list: + - metric: bits_per_byte + aggregation: bits_per_byte + higher_is_better: false diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_mmlu_stem_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_mmlu_stem_bpb_template_yaml new file mode 100644 index 00000000..e6d70fa4 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_mmlu_stem_bpb_template_yaml @@ -0,0 +1,18 @@ +# MMLU in cloze/continuation form: the continuation is the answer text rather +# than the letter "A".."D", so BPB measures knowledge instead of format. +# Every subject carries the `mmlu_stem_bpb` tag, so a single +# `--tasks mmlu_stem_bpb` invocation runs all 19 STEM subjects in one process +# and writes them to one results JSON. +include: _bpb_template_yaml +tag: + - bpb + - mmlu_stem_bpb +dataset_path: cais/mmlu +test_split: test +fewshot_split: dev +fewshot_config: + sampler: first_n +doc_to_text: "Question: {{question.strip()}}\nAnswer:" +doc_to_target: "{{answer}}" +doc_to_choice: "{{choices}}" +process_results: !function utils.process_results_mmlu diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_xcopa_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_xcopa_bpb_template_yaml new file mode 100644 index 00000000..d689d5e9 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_xcopa_bpb_template_yaml @@ -0,0 +1,13 @@ +# XCOPA with the upstream lm-eval prompt; the per-language connector comes from +# each task's doc_to_text. Per-language tasks live in xcopa/. +include: _bpb_template_yaml +tag: + - bpb + - bpb_multilingual + - xcopa_bpb +dataset_path: cambridgeltl/xcopa +validation_split: validation +test_split: test +doc_to_target: label +doc_to_choice: !function utils.doc_to_choice_xcopa +process_results: !function utils.process_results_xcopa diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_xstorycloze_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_xstorycloze_bpb_template_yaml new file mode 100644 index 00000000..4fe4a804 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_xstorycloze_bpb_template_yaml @@ -0,0 +1,14 @@ +# XStoryCloze with the upstream lm-eval prompt. Per-language tasks live in +# xstorycloze/. +include: _bpb_template_yaml +tag: + - bpb + - bpb_multilingual + - xstorycloze_bpb +dataset_path: juletxara/xstory_cloze +training_split: train +validation_split: eval +doc_to_text: "{{[input_sentence_1, input_sentence_2, input_sentence_3, input_sentence_4]|join(' ')}}" +doc_to_target: "{{answer_right_ending-1}}" +doc_to_choice: "{{[sentence_quiz1, sentence_quiz2]}}" +process_results: !function utils.process_results_xstorycloze diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/_xwinograd_bpb_template_yaml b/oellm/resources/custom_lm_eval_tasks/bpb/_xwinograd_bpb_template_yaml new file mode 100644 index 00000000..e9e3f145 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/_xwinograd_bpb_template_yaml @@ -0,0 +1,14 @@ +# XWinograd shares the Winogrande schema, so it reuses winogrande_bpb's +# multiple_input functions: BPB is measured on the shared continuation. +# Per-language tasks live in xwinograd/. +include: _bpb_template_yaml +tag: + - bpb + - bpb_multilingual + - xwinograd_bpb +dataset_path: Muennighoff/xwinograd +test_split: test +doc_to_text: !function utils.doc_to_text_winogrande +doc_to_target: !function utils.doc_to_target_winogrande +doc_to_choice: !function utils.doc_to_choice_winogrande +process_results: !function utils.process_results_winogrande diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/arc_challenge_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/arc_challenge_bpb.yaml new file mode 100644 index 00000000..a2a80c99 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/arc_challenge_bpb.yaml @@ -0,0 +1,11 @@ +include: _bpb_template_yaml +task: arc_challenge_bpb +dataset_path: allenai/ai2_arc +dataset_name: ARC-Challenge +training_split: train +validation_split: validation +test_split: test +doc_to_text: "Question: {{question}}\nAnswer:" +doc_to_target: "{{choices.label.index(answerKey)}}" +doc_to_choice: "{{choices.text}}" +process_results: !function utils.process_results_arc diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/arc_easy_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/arc_easy_bpb.yaml new file mode 100644 index 00000000..d4e8f437 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/arc_easy_bpb.yaml @@ -0,0 +1,11 @@ +include: _bpb_template_yaml +task: arc_easy_bpb +dataset_path: allenai/ai2_arc +dataset_name: ARC-Easy +training_split: train +validation_split: validation +test_split: test +doc_to_text: "Question: {{question}}\nAnswer:" +doc_to_target: "{{choices.label.index(answerKey)}}" +doc_to_choice: "{{choices.text}}" +process_results: !function utils.process_results_arc diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_bul_Cyrl_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_bul_Cyrl_bpb.yaml new file mode 100644 index 00000000..bde960f5 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_bul_Cyrl_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_bul_Cyrl_bpb +dataset_name: bul_Cyrl diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ces_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ces_Latn_bpb.yaml new file mode 100644 index 00000000..91f3f984 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ces_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_ces_Latn_bpb +dataset_name: ces_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_dan_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_dan_Latn_bpb.yaml new file mode 100644 index 00000000..2d09e0cf --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_dan_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_dan_Latn_bpb +dataset_name: dan_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_deu_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_deu_Latn_bpb.yaml new file mode 100644 index 00000000..6be6e913 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_deu_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_deu_Latn_bpb +dataset_name: deu_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ell_Grek_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ell_Grek_bpb.yaml new file mode 100644 index 00000000..6467ed0b --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ell_Grek_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_ell_Grek_bpb +dataset_name: ell_Grek diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_eng_Latn_bpb.yaml new file mode 100644 index 00000000..5e418e6b --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_eng_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_eng_Latn_bpb +dataset_name: eng_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_est_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_est_Latn_bpb.yaml new file mode 100644 index 00000000..fa3e7798 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_est_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_est_Latn_bpb +dataset_name: est_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_fin_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_fin_Latn_bpb.yaml new file mode 100644 index 00000000..e0205942 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_fin_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_fin_Latn_bpb +dataset_name: fin_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_fra_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_fra_Latn_bpb.yaml new file mode 100644 index 00000000..2932d13c --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_fra_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_fra_Latn_bpb +dataset_name: fra_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_hrv_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_hrv_Latn_bpb.yaml new file mode 100644 index 00000000..988c8ba3 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_hrv_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_hrv_Latn_bpb +dataset_name: hrv_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_hun_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_hun_Latn_bpb.yaml new file mode 100644 index 00000000..c94cd8ff --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_hun_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_hun_Latn_bpb +dataset_name: hun_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ita_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ita_Latn_bpb.yaml new file mode 100644 index 00000000..981a531f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ita_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_ita_Latn_bpb +dataset_name: ita_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_lit_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_lit_Latn_bpb.yaml new file mode 100644 index 00000000..c9208cb9 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_lit_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_lit_Latn_bpb +dataset_name: lit_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_lvs_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_lvs_Latn_bpb.yaml new file mode 100644 index 00000000..e8d1e17d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_lvs_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_lvs_Latn_bpb +dataset_name: lvs_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_mlt_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_mlt_Latn_bpb.yaml new file mode 100644 index 00000000..dbaec74d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_mlt_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_mlt_Latn_bpb +dataset_name: mlt_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_nld_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_nld_Latn_bpb.yaml new file mode 100644 index 00000000..2a9227d2 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_nld_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_nld_Latn_bpb +dataset_name: nld_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_nob_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_nob_Latn_bpb.yaml new file mode 100644 index 00000000..af49436d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_nob_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_nob_Latn_bpb +dataset_name: nob_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_pol_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_pol_Latn_bpb.yaml new file mode 100644 index 00000000..d05cd88e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_pol_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_pol_Latn_bpb +dataset_name: pol_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_por_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_por_Latn_bpb.yaml new file mode 100644 index 00000000..5cc81e60 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_por_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_por_Latn_bpb +dataset_name: por_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ron_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ron_Latn_bpb.yaml new file mode 100644 index 00000000..57ef7b79 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_ron_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_ron_Latn_bpb +dataset_name: ron_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_slk_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_slk_Latn_bpb.yaml new file mode 100644 index 00000000..538513fc --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_slk_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_slk_Latn_bpb +dataset_name: slk_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_slv_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_slv_Latn_bpb.yaml new file mode 100644 index 00000000..ef601d41 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_slv_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_slv_Latn_bpb +dataset_name: slv_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_spa_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_spa_Latn_bpb.yaml new file mode 100644 index 00000000..4e145035 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_spa_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_spa_Latn_bpb +dataset_name: spa_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_swe_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_swe_Latn_bpb.yaml new file mode 100644 index 00000000..368037f2 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/belebele/belebele_swe_Latn_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_belebele_bpb_template_yaml +task: belebele_swe_Latn_bpb +dataset_name: swe_Latn diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/boolq_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/boolq_bpb.yaml new file mode 100644 index 00000000..7871226d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/boolq_bpb.yaml @@ -0,0 +1,10 @@ +include: _bpb_template_yaml +task: boolq_bpb +dataset_path: aps/super_glue +dataset_name: boolq +training_split: train +validation_split: validation +doc_to_text: "{{passage}}\nQuestion: {{question}}?\nAnswer:" +doc_to_target: label +doc_to_choice: ["no", "yes"] +process_results: !function utils.process_results_boolq diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/commonsense_qa_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/commonsense_qa_bpb.yaml new file mode 100644 index 00000000..60ba6b65 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/commonsense_qa_bpb.yaml @@ -0,0 +1,11 @@ +# Cloze form: the continuation is the answer *text*, not the letter "A".."E". +# Scoring BPB on a single letter would measure format-following, not knowledge. +include: _bpb_template_yaml +task: commonsense_qa_bpb +dataset_path: tau/commonsense_qa +training_split: train +validation_split: validation +doc_to_text: "Question: {{ question.strip() }}\nAnswer:" +doc_to_target: "{{choices.label.index(answerKey)}}" +doc_to_choice: "{{choices.text}}" +process_results: !function utils.process_results_commonsense_qa diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/copa_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/copa_bpb.yaml new file mode 100644 index 00000000..4f21fe1c --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/copa_bpb.yaml @@ -0,0 +1,10 @@ +include: _bpb_template_yaml +task: copa_bpb +dataset_path: aps/super_glue +dataset_name: copa +training_split: train +validation_split: validation +doc_to_text: !function utils.doc_to_text_copa +doc_to_target: !function utils.doc_to_target_copa +doc_to_choice: !function utils.doc_to_choice_copa +process_results: !function utils.process_results_copa diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_als_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_als_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..29c1c19c --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_als_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_als_Latn-eng_Latn_bpb +dataset_name: als_Latn-eng_Latn +doc_to_text: 'SQ: {{sentence_als_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_bos_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_bos_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..30cbebe5 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_bos_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_bos_Latn-eng_Latn_bpb +dataset_name: bos_Latn-eng_Latn +doc_to_text: 'BS: {{sentence_bos_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_bul_Cyrl-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_bul_Cyrl-eng_Latn_bpb.yaml new file mode 100644 index 00000000..0d6c9846 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_bul_Cyrl-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_bul_Cyrl-eng_Latn_bpb +dataset_name: bul_Cyrl-eng_Latn +doc_to_text: 'BG: {{sentence_bul_Cyrl}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_cat_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_cat_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..49d8781e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_cat_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_cat_Latn-eng_Latn_bpb +dataset_name: cat_Latn-eng_Latn +doc_to_text: 'CA: {{sentence_cat_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ces_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ces_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..92d87031 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ces_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_ces_Latn-eng_Latn_bpb +dataset_name: ces_Latn-eng_Latn +doc_to_text: 'CS: {{sentence_ces_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_dan_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_dan_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..4bb1f91e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_dan_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_dan_Latn-eng_Latn_bpb +dataset_name: dan_Latn-eng_Latn +doc_to_text: 'DA: {{sentence_dan_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_deu_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_deu_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..96c398bd --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_deu_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_deu_Latn-eng_Latn_bpb +dataset_name: deu_Latn-eng_Latn +doc_to_text: 'DE: {{sentence_deu_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ell_Grek-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ell_Grek-eng_Latn_bpb.yaml new file mode 100644 index 00000000..bac96b37 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ell_Grek-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_ell_Grek-eng_Latn_bpb +dataset_name: ell_Grek-eng_Latn +doc_to_text: 'EL: {{sentence_ell_Grek}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-als_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-als_Latn_bpb.yaml new file mode 100644 index 00000000..7c08a96e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-als_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-als_Latn_bpb +dataset_name: eng_Latn-als_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} SQ:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-bos_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-bos_Latn_bpb.yaml new file mode 100644 index 00000000..32229e52 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-bos_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-bos_Latn_bpb +dataset_name: eng_Latn-bos_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} BS:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-bul_Cyrl_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-bul_Cyrl_bpb.yaml new file mode 100644 index 00000000..cc45c194 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-bul_Cyrl_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-bul_Cyrl_bpb +dataset_name: eng_Latn-bul_Cyrl +doc_to_text: 'EN: {{sentence_eng_Latn}} BG:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-cat_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-cat_Latn_bpb.yaml new file mode 100644 index 00000000..afe4ac31 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-cat_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-cat_Latn_bpb +dataset_name: eng_Latn-cat_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} CA:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ces_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ces_Latn_bpb.yaml new file mode 100644 index 00000000..afc09020 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ces_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-ces_Latn_bpb +dataset_name: eng_Latn-ces_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} CS:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-dan_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-dan_Latn_bpb.yaml new file mode 100644 index 00000000..962ccb5b --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-dan_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-dan_Latn_bpb +dataset_name: eng_Latn-dan_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} DA:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-deu_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-deu_Latn_bpb.yaml new file mode 100644 index 00000000..ce10dcda --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-deu_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-deu_Latn_bpb +dataset_name: eng_Latn-deu_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} DE:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ell_Grek_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ell_Grek_bpb.yaml new file mode 100644 index 00000000..2e8c4906 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ell_Grek_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-ell_Grek_bpb +dataset_name: eng_Latn-ell_Grek +doc_to_text: 'EN: {{sentence_eng_Latn}} EL:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-est_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-est_Latn_bpb.yaml new file mode 100644 index 00000000..b44c1be4 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-est_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-est_Latn_bpb +dataset_name: eng_Latn-est_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} ET:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-eus_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-eus_Latn_bpb.yaml new file mode 100644 index 00000000..7dd4b126 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-eus_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-eus_Latn_bpb +dataset_name: eng_Latn-eus_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} EU:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-fin_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-fin_Latn_bpb.yaml new file mode 100644 index 00000000..9b8e69c8 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-fin_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-fin_Latn_bpb +dataset_name: eng_Latn-fin_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} FI:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-fra_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-fra_Latn_bpb.yaml new file mode 100644 index 00000000..07f7ed90 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-fra_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-fra_Latn_bpb +dataset_name: eng_Latn-fra_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} FR:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-gle_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-gle_Latn_bpb.yaml new file mode 100644 index 00000000..10db051e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-gle_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-gle_Latn_bpb +dataset_name: eng_Latn-gle_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} GA:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-glg_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-glg_Latn_bpb.yaml new file mode 100644 index 00000000..d0709825 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-glg_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-glg_Latn_bpb +dataset_name: eng_Latn-glg_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} GL:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-hrv_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-hrv_Latn_bpb.yaml new file mode 100644 index 00000000..79df786a --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-hrv_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-hrv_Latn_bpb +dataset_name: eng_Latn-hrv_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} HR:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-hun_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-hun_Latn_bpb.yaml new file mode 100644 index 00000000..6f69d714 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-hun_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-hun_Latn_bpb +dataset_name: eng_Latn-hun_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} HU:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-isl_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-isl_Latn_bpb.yaml new file mode 100644 index 00000000..892ca918 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-isl_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-isl_Latn_bpb +dataset_name: eng_Latn-isl_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} IS:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ita_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ita_Latn_bpb.yaml new file mode 100644 index 00000000..95cbbf0e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ita_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-ita_Latn_bpb +dataset_name: eng_Latn-ita_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} IT:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-kat_Geor_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-kat_Geor_bpb.yaml new file mode 100644 index 00000000..2a787cc0 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-kat_Geor_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-kat_Geor_bpb +dataset_name: eng_Latn-kat_Geor +doc_to_text: 'EN: {{sentence_eng_Latn}} KA:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-lit_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-lit_Latn_bpb.yaml new file mode 100644 index 00000000..ae27ca7a --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-lit_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-lit_Latn_bpb +dataset_name: eng_Latn-lit_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} LT:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-lvs_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-lvs_Latn_bpb.yaml new file mode 100644 index 00000000..c33180d3 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-lvs_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-lvs_Latn_bpb +dataset_name: eng_Latn-lvs_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} LV:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-mkd_Cyrl_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-mkd_Cyrl_bpb.yaml new file mode 100644 index 00000000..07d59193 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-mkd_Cyrl_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-mkd_Cyrl_bpb +dataset_name: eng_Latn-mkd_Cyrl +doc_to_text: 'EN: {{sentence_eng_Latn}} MK:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-mlt_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-mlt_Latn_bpb.yaml new file mode 100644 index 00000000..e22a1dc9 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-mlt_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-mlt_Latn_bpb +dataset_name: eng_Latn-mlt_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} MT:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-nld_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-nld_Latn_bpb.yaml new file mode 100644 index 00000000..765f534b --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-nld_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-nld_Latn_bpb +dataset_name: eng_Latn-nld_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} NL:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-nob_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-nob_Latn_bpb.yaml new file mode 100644 index 00000000..d5ca5a8d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-nob_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-nob_Latn_bpb +dataset_name: eng_Latn-nob_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} NB:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-pol_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-pol_Latn_bpb.yaml new file mode 100644 index 00000000..85960d32 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-pol_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-pol_Latn_bpb +dataset_name: eng_Latn-pol_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} PL:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-por_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-por_Latn_bpb.yaml new file mode 100644 index 00000000..2dad20f0 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-por_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-por_Latn_bpb +dataset_name: eng_Latn-por_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} PT:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ron_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ron_Latn_bpb.yaml new file mode 100644 index 00000000..c1a62802 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ron_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-ron_Latn_bpb +dataset_name: eng_Latn-ron_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} RO:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-slk_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-slk_Latn_bpb.yaml new file mode 100644 index 00000000..6f9806af --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-slk_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-slk_Latn_bpb +dataset_name: eng_Latn-slk_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} SK:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-slv_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-slv_Latn_bpb.yaml new file mode 100644 index 00000000..dbc8f328 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-slv_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-slv_Latn_bpb +dataset_name: eng_Latn-slv_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} SL:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-spa_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-spa_Latn_bpb.yaml new file mode 100644 index 00000000..908aa144 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-spa_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-spa_Latn_bpb +dataset_name: eng_Latn-spa_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} ES:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-srp_Cyrl_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-srp_Cyrl_bpb.yaml new file mode 100644 index 00000000..29a46df2 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-srp_Cyrl_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-srp_Cyrl_bpb +dataset_name: eng_Latn-srp_Cyrl +doc_to_text: 'EN: {{sentence_eng_Latn}} SR:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-swe_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-swe_Latn_bpb.yaml new file mode 100644 index 00000000..83610761 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-swe_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-swe_Latn_bpb +dataset_name: eng_Latn-swe_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} SV:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-tur_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-tur_Latn_bpb.yaml new file mode 100644 index 00000000..9a5c80b6 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-tur_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-tur_Latn_bpb +dataset_name: eng_Latn-tur_Latn +doc_to_text: 'EN: {{sentence_eng_Latn}} TR:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ukr_Cyrl_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ukr_Cyrl_bpb.yaml new file mode 100644 index 00000000..8e8e0a7f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eng_Latn-ukr_Cyrl_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_eng_xx_bpb_template_yaml +task: flores200_eng_Latn-ukr_Cyrl_bpb +dataset_name: eng_Latn-ukr_Cyrl +doc_to_text: 'EN: {{sentence_eng_Latn}} UK:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_est_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_est_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..7b7fd709 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_est_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_est_Latn-eng_Latn_bpb +dataset_name: est_Latn-eng_Latn +doc_to_text: 'ET: {{sentence_est_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eus_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eus_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..40a70799 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_eus_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_eus_Latn-eng_Latn_bpb +dataset_name: eus_Latn-eng_Latn +doc_to_text: 'EU: {{sentence_eus_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_fin_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_fin_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..8d99b504 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_fin_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_fin_Latn-eng_Latn_bpb +dataset_name: fin_Latn-eng_Latn +doc_to_text: 'FI: {{sentence_fin_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_fra_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_fra_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..bfbd8ccd --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_fra_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_fra_Latn-eng_Latn_bpb +dataset_name: fra_Latn-eng_Latn +doc_to_text: 'FR: {{sentence_fra_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_gle_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_gle_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..5f10723d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_gle_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_gle_Latn-eng_Latn_bpb +dataset_name: gle_Latn-eng_Latn +doc_to_text: 'GA: {{sentence_gle_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_glg_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_glg_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..37b5d9b3 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_glg_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_glg_Latn-eng_Latn_bpb +dataset_name: glg_Latn-eng_Latn +doc_to_text: 'GL: {{sentence_glg_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_hrv_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_hrv_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..f8123d2c --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_hrv_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_hrv_Latn-eng_Latn_bpb +dataset_name: hrv_Latn-eng_Latn +doc_to_text: 'HR: {{sentence_hrv_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_hun_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_hun_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..fb4ad0cc --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_hun_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_hun_Latn-eng_Latn_bpb +dataset_name: hun_Latn-eng_Latn +doc_to_text: 'HU: {{sentence_hun_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_isl_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_isl_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..9bf5e2f7 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_isl_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_isl_Latn-eng_Latn_bpb +dataset_name: isl_Latn-eng_Latn +doc_to_text: 'IS: {{sentence_isl_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ita_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ita_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..4c092131 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ita_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_ita_Latn-eng_Latn_bpb +dataset_name: ita_Latn-eng_Latn +doc_to_text: 'IT: {{sentence_ita_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_kat_Geor-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_kat_Geor-eng_Latn_bpb.yaml new file mode 100644 index 00000000..638c1bc8 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_kat_Geor-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_kat_Geor-eng_Latn_bpb +dataset_name: kat_Geor-eng_Latn +doc_to_text: 'KA: {{sentence_kat_Geor}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_lit_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_lit_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..c86d9e5b --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_lit_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_lit_Latn-eng_Latn_bpb +dataset_name: lit_Latn-eng_Latn +doc_to_text: 'LT: {{sentence_lit_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_lvs_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_lvs_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..7f8d87a4 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_lvs_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_lvs_Latn-eng_Latn_bpb +dataset_name: lvs_Latn-eng_Latn +doc_to_text: 'LV: {{sentence_lvs_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_mkd_Cyrl-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_mkd_Cyrl-eng_Latn_bpb.yaml new file mode 100644 index 00000000..265e4d7a --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_mkd_Cyrl-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_mkd_Cyrl-eng_Latn_bpb +dataset_name: mkd_Cyrl-eng_Latn +doc_to_text: 'MK: {{sentence_mkd_Cyrl}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_mlt_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_mlt_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..ae4bbace --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_mlt_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_mlt_Latn-eng_Latn_bpb +dataset_name: mlt_Latn-eng_Latn +doc_to_text: 'MT: {{sentence_mlt_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_nld_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_nld_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..6031d85f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_nld_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_nld_Latn-eng_Latn_bpb +dataset_name: nld_Latn-eng_Latn +doc_to_text: 'NL: {{sentence_nld_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_nob_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_nob_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..698b2278 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_nob_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_nob_Latn-eng_Latn_bpb +dataset_name: nob_Latn-eng_Latn +doc_to_text: 'NB: {{sentence_nob_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_pol_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_pol_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..1a5f33bd --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_pol_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_pol_Latn-eng_Latn_bpb +dataset_name: pol_Latn-eng_Latn +doc_to_text: 'PL: {{sentence_pol_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_por_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_por_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..fb73abc7 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_por_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_por_Latn-eng_Latn_bpb +dataset_name: por_Latn-eng_Latn +doc_to_text: 'PT: {{sentence_por_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ron_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ron_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..48abd04f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ron_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_ron_Latn-eng_Latn_bpb +dataset_name: ron_Latn-eng_Latn +doc_to_text: 'RO: {{sentence_ron_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_slk_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_slk_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..91af5e6c --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_slk_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_slk_Latn-eng_Latn_bpb +dataset_name: slk_Latn-eng_Latn +doc_to_text: 'SK: {{sentence_slk_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_slv_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_slv_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..75bc0fd5 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_slv_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_slv_Latn-eng_Latn_bpb +dataset_name: slv_Latn-eng_Latn +doc_to_text: 'SL: {{sentence_slv_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_spa_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_spa_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..38cd3494 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_spa_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_spa_Latn-eng_Latn_bpb +dataset_name: spa_Latn-eng_Latn +doc_to_text: 'ES: {{sentence_spa_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_srp_Cyrl-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_srp_Cyrl-eng_Latn_bpb.yaml new file mode 100644 index 00000000..71577b93 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_srp_Cyrl-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_srp_Cyrl-eng_Latn_bpb +dataset_name: srp_Cyrl-eng_Latn +doc_to_text: 'SR: {{sentence_srp_Cyrl}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_swe_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_swe_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..9e791957 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_swe_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_swe_Latn-eng_Latn_bpb +dataset_name: swe_Latn-eng_Latn +doc_to_text: 'SV: {{sentence_swe_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_tur_Latn-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_tur_Latn-eng_Latn_bpb.yaml new file mode 100644 index 00000000..75f01837 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_tur_Latn-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_tur_Latn-eng_Latn_bpb +dataset_name: tur_Latn-eng_Latn +doc_to_text: 'TR: {{sentence_tur_Latn}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ukr_Cyrl-eng_Latn_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ukr_Cyrl-eng_Latn_bpb.yaml new file mode 100644 index 00000000..8484fb44 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/flores200/flores200_ukr_Cyrl-eng_Latn_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_flores200_xx_eng_bpb_template_yaml +task: flores200_ukr_Cyrl-eng_Latn_bpb +dataset_name: ukr_Cyrl-eng_Latn +doc_to_text: 'UK: {{sentence_ukr_Cyrl}} EN:' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_cs_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_cs_bpb.yaml new file mode 100644 index 00000000..8a135368 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_cs_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_cs_bpb +dataset_name: cs diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_de_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_de_bpb.yaml new file mode 100644 index 00000000..4e3d2fdd --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_de_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_de_bpb +dataset_name: de diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_el_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_el_bpb.yaml new file mode 100644 index 00000000..867466ef --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_el_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_el_bpb +dataset_name: el diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_en_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_en_bpb.yaml new file mode 100644 index 00000000..182d317c --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_en_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_en_bpb +dataset_name: en diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_es_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_es_bpb.yaml new file mode 100644 index 00000000..c3eae381 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_es_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_es_bpb +dataset_name: es diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_fr_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_fr_bpb.yaml new file mode 100644 index 00000000..bb4f741a --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_fr_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_fr_bpb +dataset_name: fr diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_he_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_he_bpb.yaml new file mode 100644 index 00000000..39eb4d3a --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_he_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_he_bpb +dataset_name: he diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_it_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_it_bpb.yaml new file mode 100644 index 00000000..55ad795d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_it_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_it_bpb +dataset_name: it diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_lt_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_lt_bpb.yaml new file mode 100644 index 00000000..a350fdd2 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_lt_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_lt_bpb +dataset_name: lt diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_nl_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_nl_bpb.yaml new file mode 100644 index 00000000..4b2d2f97 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_nl_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_nl_bpb +dataset_name: nl diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_pl_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_pl_bpb.yaml new file mode 100644 index 00000000..4cac8955 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_pl_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_pl_bpb +dataset_name: pl diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_pt_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_pt_bpb.yaml new file mode 100644 index 00000000..7f9accae --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_pt_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_pt_bpb +dataset_name: pt diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_ro_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_ro_bpb.yaml new file mode 100644 index 00000000..90fb1143 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_ro_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_ro_bpb +dataset_name: ro diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_ru_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_ru_bpb.yaml new file mode 100644 index 00000000..993bb565 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_ru_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_ru_bpb +dataset_name: ru diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_sr_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_sr_bpb.yaml new file mode 100644 index 00000000..8adcea91 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_sr_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_sr_bpb +dataset_name: sr diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_sv_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_sv_bpb.yaml new file mode 100644 index 00000000..eb1c1f89 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_sv_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_sv_bpb +dataset_name: sv diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_tr_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_tr_bpb.yaml new file mode 100644 index 00000000..fe0ce6d6 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_tr_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_tr_bpb +dataset_name: tr diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_uk_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_uk_bpb.yaml new file mode 100644 index 00000000..ee29b9e9 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/global_mmlu/global_mmlu_uk_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_global_mmlu_bpb_template_yaml +task: global_mmlu_uk_bpb +dataset_name: uk diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/hellaswag_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/hellaswag_bpb.yaml new file mode 100644 index 00000000..00cb7866 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/hellaswag_bpb.yaml @@ -0,0 +1,12 @@ +include: _bpb_template_yaml +task: hellaswag_bpb +dataset_path: Rowan/hellaswag +dataset_name: null +training_split: train +validation_split: validation +test_split: null +process_docs: !function utils.process_docs_hellaswag +doc_to_text: "{{query}}" +doc_to_target: "{{gold}}" +doc_to_choice: "choices" +process_results: !function utils.process_results_hellaswag diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_albanian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_albanian_bpb.yaml new file mode 100644 index 00000000..457f76f8 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_albanian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_albanian_bpb +dataset_name: Albanian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_armenian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_armenian_bpb.yaml new file mode 100644 index 00000000..e60a051e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_armenian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_armenian_bpb +dataset_name: Armenian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_azerbaijani_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_azerbaijani_bpb.yaml new file mode 100644 index 00000000..f42aa978 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_azerbaijani_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_azerbaijani_bpb +dataset_name: Azerbaijani diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_basque_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_basque_bpb.yaml new file mode 100644 index 00000000..102ef8a6 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_basque_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_basque_bpb +dataset_name: Basque diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_belarusian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_belarusian_bpb.yaml new file mode 100644 index 00000000..55173014 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_belarusian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_belarusian_bpb +dataset_name: Belarusian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_bulgarian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_bulgarian_bpb.yaml new file mode 100644 index 00000000..40c65e68 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_bulgarian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_bulgarian_bpb +dataset_name: Bulgarian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_croatian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_croatian_bpb.yaml new file mode 100644 index 00000000..3ebd7109 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_croatian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_croatian_bpb +dataset_name: Croatian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_dutch_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_dutch_bpb.yaml new file mode 100644 index 00000000..35d37e05 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_dutch_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_dutch_bpb +dataset_name: Dutch diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_estonian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_estonian_bpb.yaml new file mode 100644 index 00000000..ce298876 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_estonian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_estonian_bpb +dataset_name: Estonian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_finnish_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_finnish_bpb.yaml new file mode 100644 index 00000000..669e950e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_finnish_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_finnish_bpb +dataset_name: Finnish diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_french_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_french_bpb.yaml new file mode 100644 index 00000000..7cee7615 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_french_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_french_bpb +dataset_name: French diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_georgian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_georgian_bpb.yaml new file mode 100644 index 00000000..e8c3a0cd --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_georgian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_georgian_bpb +dataset_name: Georgian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_german_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_german_bpb.yaml new file mode 100644 index 00000000..56439977 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_german_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_german_bpb +dataset_name: German diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_greek_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_greek_bpb.yaml new file mode 100644 index 00000000..44d4a3bb --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_greek_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_greek_bpb +dataset_name: Greek diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_hungarian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_hungarian_bpb.yaml new file mode 100644 index 00000000..5c478d6f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_hungarian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_hungarian_bpb +dataset_name: Hungarian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_italian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_italian_bpb.yaml new file mode 100644 index 00000000..893d68fd --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_italian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_italian_bpb +dataset_name: Italian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_lithuanian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_lithuanian_bpb.yaml new file mode 100644 index 00000000..c3748d40 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_lithuanian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_lithuanian_bpb +dataset_name: Lithuanian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_north_macedonian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_north_macedonian_bpb.yaml new file mode 100644 index 00000000..d9eda43a --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_north_macedonian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_north_macedonian_bpb +dataset_name: North Macedonian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_polish_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_polish_bpb.yaml new file mode 100644 index 00000000..9d327ca8 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_polish_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_polish_bpb +dataset_name: Polish diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_portuguese_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_portuguese_bpb.yaml new file mode 100644 index 00000000..489047e2 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_portuguese_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_portuguese_bpb +dataset_name: Portuguese diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_russian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_russian_bpb.yaml new file mode 100644 index 00000000..5c7b30d6 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_russian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_russian_bpb +dataset_name: Russian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_serbian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_serbian_bpb.yaml new file mode 100644 index 00000000..8e399b3d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_serbian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_serbian_bpb +dataset_name: Serbian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_spanish_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_spanish_bpb.yaml new file mode 100644 index 00000000..61f3e531 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_spanish_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_spanish_bpb +dataset_name: Spanish diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_turkish_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_turkish_bpb.yaml new file mode 100644 index 00000000..2e8d02ba --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_turkish_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_turkish_bpb +dataset_name: Turkish diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_ukrainian_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_ukrainian_bpb.yaml new file mode 100644 index 00000000..d4a63944 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/include/include_base_44_ukrainian_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_include_bpb_template_yaml +task: include_base_44_ukrainian_bpb +dataset_name: Ukrainian diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_de_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_de_bpb.yaml new file mode 100644 index 00000000..6acc1354 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_de_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_mgsm_bpb_template_yaml +task: mgsm_direct_de_bpb +dataset_name: de +doc_to_text: '{% if answer is not none %}{{question+"\nAntwort:"}}{% else %}{{"Frage: "+question+"\nAntwort:"}}{% endif %}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_en_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_en_bpb.yaml new file mode 100644 index 00000000..a3d17ac7 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_en_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_mgsm_bpb_template_yaml +task: mgsm_direct_en_bpb +dataset_name: en +doc_to_text: '{% if answer is not none %}{{question+"\nAnswer:"}}{% else %}{{"Question: "+question+"\nAnswer:"}}{% endif %}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_es_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_es_bpb.yaml new file mode 100644 index 00000000..72b385e6 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_es_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_mgsm_bpb_template_yaml +task: mgsm_direct_es_bpb +dataset_name: es +doc_to_text: '{% if answer is not none %}{{question+"\nRespuesta:"}}{% else %}{{"Pregunta: "+question+"\nRespuesta:"}}{% endif %}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_fr_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_fr_bpb.yaml new file mode 100644 index 00000000..a64bc4da --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mgsm/mgsm_direct_fr_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_mgsm_bpb_template_yaml +task: mgsm_direct_fr_bpb +dataset_name: fr +doc_to_text: '{% if answer is not none %}{{question+"\nRéponse :"}}{% else %}{{"Question : "+question+"\nRéponse :"}}{% endif %}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_abstract_algebra_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_abstract_algebra_bpb.yaml new file mode 100644 index 00000000..2e8f24ac --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_abstract_algebra_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_abstract_algebra_bpb +dataset_name: abstract_algebra diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_anatomy_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_anatomy_bpb.yaml new file mode 100644 index 00000000..6b038236 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_anatomy_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_anatomy_bpb +dataset_name: anatomy diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_astronomy_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_astronomy_bpb.yaml new file mode 100644 index 00000000..e8abb094 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_astronomy_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_astronomy_bpb +dataset_name: astronomy diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_biology_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_biology_bpb.yaml new file mode 100644 index 00000000..90f06223 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_biology_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_college_biology_bpb +dataset_name: college_biology diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_chemistry_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_chemistry_bpb.yaml new file mode 100644 index 00000000..9839d22f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_chemistry_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_college_chemistry_bpb +dataset_name: college_chemistry diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_computer_science_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_computer_science_bpb.yaml new file mode 100644 index 00000000..f95ab073 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_computer_science_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_college_computer_science_bpb +dataset_name: college_computer_science diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_mathematics_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_mathematics_bpb.yaml new file mode 100644 index 00000000..20b76800 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_mathematics_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_college_mathematics_bpb +dataset_name: college_mathematics diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_physics_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_physics_bpb.yaml new file mode 100644 index 00000000..e749fbb6 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_college_physics_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_college_physics_bpb +dataset_name: college_physics diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_computer_security_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_computer_security_bpb.yaml new file mode 100644 index 00000000..a47d1a88 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_computer_security_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_computer_security_bpb +dataset_name: computer_security diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_conceptual_physics_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_conceptual_physics_bpb.yaml new file mode 100644 index 00000000..0e2693f7 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_conceptual_physics_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_conceptual_physics_bpb +dataset_name: conceptual_physics diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_electrical_engineering_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_electrical_engineering_bpb.yaml new file mode 100644 index 00000000..ae8d26e5 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_electrical_engineering_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_electrical_engineering_bpb +dataset_name: electrical_engineering diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_elementary_mathematics_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_elementary_mathematics_bpb.yaml new file mode 100644 index 00000000..219590ff --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_elementary_mathematics_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_elementary_mathematics_bpb +dataset_name: elementary_mathematics diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_biology_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_biology_bpb.yaml new file mode 100644 index 00000000..e174245f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_biology_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_high_school_biology_bpb +dataset_name: high_school_biology diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_chemistry_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_chemistry_bpb.yaml new file mode 100644 index 00000000..d395cbae --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_chemistry_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_high_school_chemistry_bpb +dataset_name: high_school_chemistry diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_computer_science_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_computer_science_bpb.yaml new file mode 100644 index 00000000..44353b47 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_computer_science_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_high_school_computer_science_bpb +dataset_name: high_school_computer_science diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_mathematics_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_mathematics_bpb.yaml new file mode 100644 index 00000000..84af7190 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_mathematics_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_high_school_mathematics_bpb +dataset_name: high_school_mathematics diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_physics_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_physics_bpb.yaml new file mode 100644 index 00000000..baff24ed --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_physics_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_high_school_physics_bpb +dataset_name: high_school_physics diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_statistics_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_statistics_bpb.yaml new file mode 100644 index 00000000..cad9b14f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_high_school_statistics_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_high_school_statistics_bpb +dataset_name: high_school_statistics diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_machine_learning_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_machine_learning_bpb.yaml new file mode 100644 index 00000000..e29b5470 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/mmlu_machine_learning_bpb.yaml @@ -0,0 +1,3 @@ +include: _mmlu_stem_bpb_template_yaml +task: mmlu_machine_learning_bpb +dataset_name: machine_learning diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/openbookqa_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/openbookqa_bpb.yaml new file mode 100644 index 00000000..b372f1d6 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/openbookqa_bpb.yaml @@ -0,0 +1,11 @@ +include: _bpb_template_yaml +task: openbookqa_bpb +dataset_path: allenai/openbookqa +dataset_name: main +training_split: train +validation_split: validation +test_split: test +doc_to_text: question_stem +doc_to_target: "{{choices.label.index(answerKey.lstrip())}}" +doc_to_choice: "{{choices.text}}" +process_results: !function utils.process_results_openbookqa diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/piqa_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/piqa_bpb.yaml new file mode 100644 index 00000000..36152c1b --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/piqa_bpb.yaml @@ -0,0 +1,11 @@ +include: _bpb_template_yaml +task: piqa_bpb +dataset_path: baber/piqa +dataset_name: null +training_split: train +validation_split: validation +test_split: null +doc_to_text: "Question: {{goal}}\nAnswer:" +doc_to_target: label +doc_to_choice: "{{[sol1, sol2]}}" +process_results: !function utils.process_results_piqa diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/social_iqa_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/social_iqa_bpb.yaml new file mode 100644 index 00000000..17b0b5ed --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/social_iqa_bpb.yaml @@ -0,0 +1,10 @@ +include: _bpb_template_yaml +task: social_iqa_bpb +dataset_path: allenai/social_i_qa +dataset_name: null +training_split: train +validation_split: validation +doc_to_text: "Question: {{context}} {{question}}\nAnswer:" +doc_to_target: "{{label|int - 1}}" +doc_to_choice: "{{[answerA, answerB, answerC]}}" +process_results: !function utils.process_results_social_iqa diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/utils.py b/oellm/resources/custom_lm_eval_tasks/bpb/utils.py new file mode 100644 index 00000000..8f520914 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/utils.py @@ -0,0 +1,411 @@ +"""Bits-per-byte (BPB) scoring for multiple-choice lm-eval-harness tasks. + +Why this file exists +-------------------- +lm-eval-harness only wires ``bits_per_byte`` into ``loglikelihood_rolling`` +tasks (wikitext-style perplexity). For ``multiple_choice`` tasks the built-in +``ConfigurableTask.process_results`` emits a fixed set of keys +(``acc``/``acc_norm``/``acc_bytes``/``f1``/``mcc``/``exact_match``/``brier_score``), +so adding ``bits_per_byte`` to a task's ``metric_list`` alone does nothing. + +The tasks in this directory therefore override ``process_results`` with the +functions below. They compute, per document: + + (loglikelihood_of_gold_continuation, byte_length_of_gold_continuation) + +which is exactly the ``(a, b)`` pair shape that lm-eval's already-registered +``bits_per_byte`` *aggregation* consumes: + + bits_per_byte = -sum(a) / sum(b) / ln(2) + +i.e. a corpus-level, byte-weighted average, not a mean of per-doc ratios. That +is the standard BPB definition used by DCLM/OLMo/Olmix, and it is what makes +BPB a smooth fitness signal for data-mixture regression at swarm scale, where +accuracy on 30M-parameter models is close to chance and very noisy. + +Byte-length convention +---------------------- +Bytes are counted on the continuation string exactly as returned by the task's +``doc_to_choice`` (UTF-8), *without* the ``target_delimiter`` (the leading +space lm-eval inserts between context and continuation). This matches lm-eval's +own ``acc_bytes`` convention. Two notes: + +* Absolute BPB values shift slightly under a different convention, but the + *ranking* of data mixtures -- the only thing the Olmix regressor consumes -- + is unaffected, as long as every run in the swarm uses the same convention. +* Do not compare these numbers against BPB from another codebase without first + checking that codebase's byte convention. + +Adding a task +------------- +Write a ``process_results_(doc, results)`` that resolves the document's +choice strings and gold index, then returns ``_score(results, choices, gold)``. +``results`` is the list of ``(loglikelihood, is_greedy)`` pairs lm-eval +produces, one per choice, in ``doc_to_choice`` order. + +Single-reference tasks (FLORES translation, MGSM answers) use +``output_type: loglikelihood`` instead and return +``_score_continuation(results, text)``. lm-eval does not insert +``target_delimiter`` for ``loglikelihood`` requests, so their ``doc_to_target`` +prepends the space itself and ``text`` is the reference *without* it, keeping +the byte convention above. +""" + +import re + +# -------------------------------------------------------------------------- +# Core scoring +# -------------------------------------------------------------------------- + + +def _loglikelihoods(results): + """Extract per-choice loglikelihoods from lm-eval's result payload.""" + lls = [] + for r in results: + # multiple_choice requests come back as (loglikelihood, is_greedy) + lls.append(float(r[0]) if isinstance(r, (tuple, list)) else float(r)) + return lls + + +def _byte_len(text): + """UTF-8 byte length, floored at 1 so an empty choice can't divide by zero.""" + return max(len(str(text).encode("utf-8")), 1) + + +def _resolve_gold(gold, choices): + """Normalise a gold label to a valid index into ``choices``, or None.""" + if isinstance(gold, (list, tuple)): + # Multi-target tasks: score against the first valid gold. + for g in gold: + resolved = _resolve_gold(g, choices) + if resolved is not None: + return resolved + return None + if isinstance(gold, str): + if gold in choices: + return choices.index(gold) + try: + gold = int(gold) + except (TypeError, ValueError): + return None + if isinstance(gold, int) and 0 <= gold < len(choices): + return gold + return None + + +def _score(results, choices, gold, continuation=None): + """Return the metric dict for one document. + + Args: + results: list of (loglikelihood, is_greedy) pairs, one per choice. + choices: the continuation strings, in ``doc_to_choice`` order. + gold: index (or label) of the correct choice. + continuation: for ``multiple_input`` tasks (e.g. winogrande) the choices + are *contexts* and the scored continuation is shared; pass it here + so bytes are counted on the continuation rather than on the context. + + Returns: + dict with ``bits_per_byte`` as an (loglikelihood, n_bytes) pair plus + ``acc``/``acc_norm`` for free, since we already have every choice's + loglikelihood. + """ + lls = _loglikelihoods(results) + choices = [str(c) for c in choices] + gold_idx = _resolve_gold(gold, choices) + + if gold_idx is None or len(lls) != len(choices): + # Malformed doc: contribute nothing rather than poisoning the corpus + # aggregate with a bogus (ll, bytes) pair. + return {"acc": 0.0, "acc_norm": 0.0, "bits_per_byte": (0.0, 1)} + + scored_text = continuation if continuation is not None else choices[gold_idx] + gold_bytes = _byte_len(scored_text) + + char_lens = [max(len(c), 1) for c in choices] + pred = max(range(len(lls)), key=lambda i: lls[i]) + pred_norm = max(range(len(lls)), key=lambda i: lls[i] / char_lens[i]) + + return { + "acc": 1.0 if pred == gold_idx else 0.0, + "acc_norm": 1.0 if pred_norm == gold_idx else 0.0, + # Consumed by lm-eval's registered "bits_per_byte" aggregation. + "bits_per_byte": (lls[gold_idx], gold_bytes), + } + + +def _score_continuation(results, continuation): + """Return the metric dict for one ``loglikelihood`` (single-reference) doc. + + Args: + results: one-element list holding the reference's (loglikelihood, is_greedy). + continuation: the reference text, without the leading space that the + task's ``doc_to_target`` prepends. + """ + return {"bits_per_byte": (_loglikelihoods(results)[0], _byte_len(continuation))} + + +# -------------------------------------------------------------------------- +# ARC (easy + challenge) and OpenBookQA share the AI2 label/answerKey schema +# -------------------------------------------------------------------------- + + +def _ai2_choices_and_gold(doc, answer_key="answerKey"): + choices = list(doc["choices"]["text"]) + labels = list(doc["choices"]["label"]) + key = str(doc[answer_key]).strip() + gold = labels.index(key) if key in labels else _resolve_gold(key, choices) + return choices, gold + + +def process_results_arc(doc, results): + choices, gold = _ai2_choices_and_gold(doc) + return _score(results, choices, gold) + + +def process_results_openbookqa(doc, results): + choices, gold = _ai2_choices_and_gold(doc) + return _score(results, choices, gold) + + +# -------------------------------------------------------------------------- +# HellaSwag +# -------------------------------------------------------------------------- + + +def _hellaswag_preprocess(text): + text = text.strip() + # Brackets are artifacts of the WikiHow portion of HellaSwag. + text = text.replace(" [title]", ". ") + text = re.sub("\\[.*?\\]", "", text) + text = text.replace(" ", " ") + return text + + +def process_docs_hellaswag(dataset): + def _process_doc(doc): + ctx = doc["ctx_a"] + " " + doc["ctx_b"].capitalize() + return { + "query": _hellaswag_preprocess(doc["activity_label"] + ": " + ctx), + "choices": [_hellaswag_preprocess(ending) for ending in doc["endings"]], + "gold": int(doc["label"]), + } + + return dataset.map(_process_doc) + + +def process_results_hellaswag(doc, results): + gold = doc["gold"] if "gold" in doc else doc["label"] + return _score(results, list(doc["choices"]), gold) + + +# -------------------------------------------------------------------------- +# PIQA / BoolQ / Social IQa +# -------------------------------------------------------------------------- + + +def process_results_piqa(doc, results): + return _score(results, [doc["sol1"], doc["sol2"]], int(doc["label"])) + + +def process_results_boolq(doc, results): + return _score(results, ["no", "yes"], int(doc["label"])) + + +def process_results_social_iqa(doc, results): + choices = [doc["answerA"], doc["answerB"], doc["answerC"]] + # social_i_qa labels are 1-indexed strings. + return _score(results, choices, int(str(doc["label"]).strip()) - 1) + + +# -------------------------------------------------------------------------- +# COPA (reimplemented here so this directory has no cross-task imports) +# -------------------------------------------------------------------------- + + +def _copa_convert_choice(choice): + return choice[0].lower() + choice[1:] + + +def doc_to_text_copa(doc): + connector = {"cause": "because", "effect": "therefore"}[doc["question"]] + # Drop the trailing period before appending the connector. + return doc["premise"].strip()[:-1] + f" {connector}" + + +def doc_to_choice_copa(doc): + return [ + " " + _copa_convert_choice(doc["choice1"]), + " " + _copa_convert_choice(doc["choice2"]), + ] + + +def doc_to_target_copa(doc): + correct = doc["choice1"] if doc["label"] == 0 else doc["choice2"] + return " " + _copa_convert_choice(correct) + + +def process_results_copa(doc, results): + return _score(results, doc_to_choice_copa(doc), int(doc["label"])) + + +# -------------------------------------------------------------------------- +# CommonsenseQA -- cloze form +# +# The stock lm-eval task scores the letters "A".."E", so its BPB would measure +# how well the model predicts a single letter, not whether it knows the answer. +# These tasks put the answer *text* in the continuation instead. +# -------------------------------------------------------------------------- + + +def process_results_commonsense_qa(doc, results): + choices, gold = _ai2_choices_and_gold(doc) + return _score(results, choices, gold) + + +# -------------------------------------------------------------------------- +# MMLU -- cloze/continuation form, for the same reason as CommonsenseQA +# -------------------------------------------------------------------------- + + +def process_results_mmlu(doc, results): + return _score(results, list(doc["choices"]), int(doc["answer"])) + + +# -------------------------------------------------------------------------- +# Winogrande -- multiple_input: the choices are contexts and the continuation +# is shared, so bytes are counted on the continuation. +# -------------------------------------------------------------------------- + + +def doc_to_text_winogrande(doc): + return {"1": 0, "2": 1}[doc["answer"]] + + +def doc_to_target_winogrande(doc): + idx = doc["sentence"].index("_") + 1 + return doc["sentence"][idx:].strip() + + +def doc_to_choice_winogrande(doc): + idx = doc["sentence"].index("_") + return [doc["sentence"][:idx] + opt for opt in [doc["option1"], doc["option2"]]] + + +def process_results_winogrande(doc, results): + return _score( + results, + doc_to_choice_winogrande(doc), + doc_to_text_winogrande(doc), + continuation=doc_to_target_winogrande(doc), + ) + + +# ========================================================================== +# Multilingual tasks (the oellm-multilingual-bpb super group) +# +# Task YAMLs live in per-benchmark subdirectories and include the +# __bpb_template_yaml files next to this module. XWinograd has the +# Winogrande schema and reuses the Winogrande functions above. +# ========================================================================== + + +# -------------------------------------------------------------------------- +# Belebele -- cloze form: the continuation is the answer text, not a letter +# -------------------------------------------------------------------------- + + +def process_results_belebele(doc, results): + choices = [doc[f"mc_answer{i}"] for i in range(1, 5)] + return _score(results, choices, int(doc["correct_answer_num"]) - 1) + + +# -------------------------------------------------------------------------- +# Global-MMLU / INCLUDE -- cloze form, for the same reason as MMLU +# -------------------------------------------------------------------------- + + +def _exam_choices(doc): + return [doc["option_a"], doc["option_b"], doc["option_c"], doc["option_d"]] + + +def process_results_global_mmlu(doc, results): + # Global-MMLU stores the gold answer as a letter. + gold = {"A": 0, "B": 1, "C": 2, "D": 3}.get(str(doc["answer"]).strip()) + return _score(results, _exam_choices(doc), gold) + + +def process_results_include(doc, results): + return _score(results, _exam_choices(doc), int(doc["answer"])) + + +# -------------------------------------------------------------------------- +# XCOPA -- same prompt as the upstream lm-eval xcopa tasks. The per-language +# cause/effect connector is in each task's doc_to_text (xcopa/*.yaml). +# -------------------------------------------------------------------------- + + +def doc_to_choice_xcopa(doc): + return [_copa_convert_choice(doc["choice1"]), _copa_convert_choice(doc["choice2"])] + + +def process_results_xcopa(doc, results): + return _score(results, doc_to_choice_xcopa(doc), int(doc["label"])) + + +# -------------------------------------------------------------------------- +# XStoryCloze +# -------------------------------------------------------------------------- + + +def process_results_xstorycloze(doc, results): + choices = [doc["sentence_quiz1"], doc["sentence_quiz2"]] + return _score(results, choices, int(doc["answer_right_ending"]) - 1) + + +# -------------------------------------------------------------------------- +# MGSM -- the test split has no worked solutions, only the final number, so +# BPB is measured on the number that follows the upstream "direct" prompt. +# -------------------------------------------------------------------------- + + +def _mgsm_answer(doc): + answer = doc["answer_number"] + if isinstance(answer, float) and answer.is_integer(): + answer = int(answer) + return str(answer) + + +def doc_to_target_mgsm(doc): + return " " + _mgsm_answer(doc) + + +def process_results_mgsm(doc, results): + return _score_continuation(results, _mgsm_answer(doc)) + + +# -------------------------------------------------------------------------- +# FLORES-200 -- BPB of the reference translation given the source sentence. +# The facebook/flores pair configs (e.g. deu_Latn-eng_Latn) carry exactly two +# sentence_ columns, one of which is English. +# -------------------------------------------------------------------------- + + +def _flores_non_english(doc): + (key,) = [k for k in doc if k.startswith("sentence_") and k != "sentence_eng_Latn"] + return doc[key] + + +def doc_to_target_flores_xx_eng(doc): + return " " + doc["sentence_eng_Latn"] + + +def process_results_flores_xx_eng(doc, results): + return _score_continuation(results, doc["sentence_eng_Latn"]) + + +def doc_to_target_flores_eng_xx(doc): + return " " + _flores_non_english(doc) + + +def process_results_flores_eng_xx(doc, results): + return _score_continuation(results, _flores_non_english(doc)) diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/winogrande_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/winogrande_bpb.yaml new file mode 100644 index 00000000..c191c3ec --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/winogrande_bpb.yaml @@ -0,0 +1,12 @@ +# multiple_input task: the two options are substituted into the *context* and +# the continuation is shared, so BPB is measured on the shared continuation. +include: _bpb_template_yaml +task: winogrande_bpb +dataset_path: allenai/winogrande +dataset_name: winogrande_xl +training_split: train +validation_split: validation +doc_to_text: !function utils.doc_to_text_winogrande +doc_to_target: !function utils.doc_to_target_winogrande +doc_to_choice: !function utils.doc_to_choice_winogrande +process_results: !function utils.process_results_winogrande diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_et_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_et_bpb.yaml new file mode 100644 index 00000000..5c730b1b --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_et_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_et_bpb +dataset_name: et +doc_to_text: '{{premise.strip()[:-1]}} {{"sest" if question == "cause" else "seetõttu"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_ht_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_ht_bpb.yaml new file mode 100644 index 00000000..d40d8c75 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_ht_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_ht_bpb +dataset_name: ht +doc_to_text: '{{premise.strip()[:-1]}} {{"poukisa" if question == "cause" else "donk sa"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_id_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_id_bpb.yaml new file mode 100644 index 00000000..cefaea6e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_id_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_id_bpb +dataset_name: id +doc_to_text: '{{premise.strip()[:-1]}} {{"karena" if question == "cause" else "maka"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_it_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_it_bpb.yaml new file mode 100644 index 00000000..78e931dc --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_it_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_it_bpb +dataset_name: it +doc_to_text: '{{premise.strip()[:-1]}} {{"perché" if question == "cause" else "quindi"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_qu_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_qu_bpb.yaml new file mode 100644 index 00000000..43200f3f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_qu_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_qu_bpb +dataset_name: qu +doc_to_text: '{{premise.strip()[:-1]}} {{"imataq" if question == "cause" else "chaymi"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_sw_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_sw_bpb.yaml new file mode 100644 index 00000000..bc4ed78a --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_sw_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_sw_bpb +dataset_name: sw +doc_to_text: '{{premise.strip()[:-1]}} {{"kwa sababu" if question == "cause" else "kwa hiyo"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_ta_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_ta_bpb.yaml new file mode 100644 index 00000000..90692a93 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_ta_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_ta_bpb +dataset_name: ta +doc_to_text: '{{premise.strip()[:-1]}} {{"காரணமாக" if question == "cause" else "எனவே"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_th_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_th_bpb.yaml new file mode 100644 index 00000000..779b920f --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_th_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_th_bpb +dataset_name: th +doc_to_text: '{{premise.strip()[:-1]}} {{"เพราะ" if question == "cause" else "ดังนั้น"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_tr_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_tr_bpb.yaml new file mode 100644 index 00000000..a029979e --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_tr_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_tr_bpb +dataset_name: tr +doc_to_text: '{{premise.strip()[:-1]}} {{"çünkü" if question == "cause" else "bu yüzden"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_vi_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_vi_bpb.yaml new file mode 100644 index 00000000..47f0a7ec --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_vi_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_vi_bpb +dataset_name: vi +doc_to_text: '{{premise.strip()[:-1]}} {{"bởi vì" if question == "cause" else "vì vậy"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_zh_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_zh_bpb.yaml new file mode 100644 index 00000000..baffa28d --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xcopa/xcopa_zh_bpb.yaml @@ -0,0 +1,4 @@ +include: ../_xcopa_bpb_template_yaml +task: xcopa_zh_bpb +dataset_name: zh +doc_to_text: '{{premise.strip()[:-1]}} {{"因为" if question == "cause" else "所以"}}' diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_ar_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_ar_bpb.yaml new file mode 100644 index 00000000..a3ba7975 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_ar_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_ar_bpb +dataset_name: ar diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_en_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_en_bpb.yaml new file mode 100644 index 00000000..53909233 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_en_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_en_bpb +dataset_name: en diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_es_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_es_bpb.yaml new file mode 100644 index 00000000..e3128bfe --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_es_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_es_bpb +dataset_name: es diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_eu_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_eu_bpb.yaml new file mode 100644 index 00000000..b7a154c1 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_eu_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_eu_bpb +dataset_name: eu diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_hi_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_hi_bpb.yaml new file mode 100644 index 00000000..a31b631c --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_hi_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_hi_bpb +dataset_name: hi diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_id_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_id_bpb.yaml new file mode 100644 index 00000000..f6eb34dd --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_id_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_id_bpb +dataset_name: id diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_my_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_my_bpb.yaml new file mode 100644 index 00000000..965136fe --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_my_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_my_bpb +dataset_name: my diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_ru_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_ru_bpb.yaml new file mode 100644 index 00000000..aa499c32 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_ru_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_ru_bpb +dataset_name: ru diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_sw_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_sw_bpb.yaml new file mode 100644 index 00000000..7f76bca0 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_sw_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_sw_bpb +dataset_name: sw diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_te_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_te_bpb.yaml new file mode 100644 index 00000000..82357523 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_te_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_te_bpb +dataset_name: te diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_zh_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_zh_bpb.yaml new file mode 100644 index 00000000..ffc610d3 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xstorycloze/xstorycloze_zh_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xstorycloze_bpb_template_yaml +task: xstorycloze_zh_bpb +dataset_name: zh diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_en_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_en_bpb.yaml new file mode 100644 index 00000000..d0140447 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_en_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xwinograd_bpb_template_yaml +task: xwinograd_en_bpb +dataset_name: en diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_fr_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_fr_bpb.yaml new file mode 100644 index 00000000..de5fff36 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_fr_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xwinograd_bpb_template_yaml +task: xwinograd_fr_bpb +dataset_name: fr diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_jp_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_jp_bpb.yaml new file mode 100644 index 00000000..f1d0df47 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_jp_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xwinograd_bpb_template_yaml +task: xwinograd_jp_bpb +dataset_name: jp diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_pt_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_pt_bpb.yaml new file mode 100644 index 00000000..5fd81236 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_pt_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xwinograd_bpb_template_yaml +task: xwinograd_pt_bpb +dataset_name: pt diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_ru_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_ru_bpb.yaml new file mode 100644 index 00000000..ba32f2fe --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_ru_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xwinograd_bpb_template_yaml +task: xwinograd_ru_bpb +dataset_name: ru diff --git a/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_zh_bpb.yaml b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_zh_bpb.yaml new file mode 100644 index 00000000..707d3385 --- /dev/null +++ b/oellm/resources/custom_lm_eval_tasks/bpb/xwinograd/xwinograd_zh_bpb.yaml @@ -0,0 +1,3 @@ +include: ../_xwinograd_bpb_template_yaml +task: xwinograd_zh_bpb +dataset_name: zh diff --git a/oellm/resources/task-groups.yaml b/oellm/resources/task-groups.yaml index 21242cc2..30ff7e9b 100644 --- a/oellm/resources/task-groups.yaml +++ b/oellm/resources/task-groups.yaml @@ -1,3 +1,560 @@ +task_metrics: + # BPB tasks: the primary metric is bits_per_byte (lower is better). + arc_easy_bpb: bits_per_byte + arc_challenge_bpb: bits_per_byte + hellaswag_bpb: bits_per_byte + piqa_bpb: bits_per_byte + openbookqa_bpb: bits_per_byte + boolq_bpb: bits_per_byte + copa_bpb: bits_per_byte + commonsense_qa_bpb: bits_per_byte + social_iqa_bpb: bits_per_byte + winogrande_bpb: bits_per_byte + mmlu_abstract_algebra_bpb: bits_per_byte + mmlu_anatomy_bpb: bits_per_byte + mmlu_astronomy_bpb: bits_per_byte + mmlu_college_biology_bpb: bits_per_byte + mmlu_college_chemistry_bpb: bits_per_byte + mmlu_college_computer_science_bpb: bits_per_byte + mmlu_college_mathematics_bpb: bits_per_byte + mmlu_college_physics_bpb: bits_per_byte + mmlu_computer_security_bpb: bits_per_byte + mmlu_conceptual_physics_bpb: bits_per_byte + mmlu_electrical_engineering_bpb: bits_per_byte + mmlu_elementary_mathematics_bpb: bits_per_byte + mmlu_high_school_biology_bpb: bits_per_byte + mmlu_high_school_chemistry_bpb: bits_per_byte + mmlu_high_school_computer_science_bpb: bits_per_byte + mmlu_high_school_mathematics_bpb: bits_per_byte + mmlu_high_school_physics_bpb: bits_per_byte + mmlu_high_school_statistics_bpb: bits_per_byte + mmlu_machine_learning_bpb: bits_per_byte + # Multilingual BPB tasks (oellm-multilingual-bpb): bits_per_byte, lower is better. + flores200_als_Latn-eng_Latn_bpb: bits_per_byte + flores200_bos_Latn-eng_Latn_bpb: bits_per_byte + flores200_bul_Cyrl-eng_Latn_bpb: bits_per_byte + flores200_cat_Latn-eng_Latn_bpb: bits_per_byte + flores200_ces_Latn-eng_Latn_bpb: bits_per_byte + flores200_dan_Latn-eng_Latn_bpb: bits_per_byte + flores200_deu_Latn-eng_Latn_bpb: bits_per_byte + flores200_ell_Grek-eng_Latn_bpb: bits_per_byte + flores200_eng_Latn-als_Latn_bpb: bits_per_byte + flores200_eng_Latn-bos_Latn_bpb: bits_per_byte + flores200_eng_Latn-bul_Cyrl_bpb: bits_per_byte + flores200_eng_Latn-cat_Latn_bpb: bits_per_byte + flores200_eng_Latn-ces_Latn_bpb: bits_per_byte + flores200_eng_Latn-dan_Latn_bpb: bits_per_byte + flores200_eng_Latn-deu_Latn_bpb: bits_per_byte + flores200_eng_Latn-ell_Grek_bpb: bits_per_byte + flores200_eng_Latn-est_Latn_bpb: bits_per_byte + flores200_eng_Latn-eus_Latn_bpb: bits_per_byte + flores200_eng_Latn-fin_Latn_bpb: bits_per_byte + flores200_eng_Latn-fra_Latn_bpb: bits_per_byte + flores200_eng_Latn-gle_Latn_bpb: bits_per_byte + flores200_eng_Latn-glg_Latn_bpb: bits_per_byte + flores200_eng_Latn-hrv_Latn_bpb: bits_per_byte + flores200_eng_Latn-hun_Latn_bpb: bits_per_byte + flores200_eng_Latn-isl_Latn_bpb: bits_per_byte + flores200_eng_Latn-ita_Latn_bpb: bits_per_byte + flores200_eng_Latn-kat_Geor_bpb: bits_per_byte + flores200_eng_Latn-lit_Latn_bpb: bits_per_byte + flores200_eng_Latn-lvs_Latn_bpb: bits_per_byte + flores200_eng_Latn-mkd_Cyrl_bpb: bits_per_byte + flores200_eng_Latn-mlt_Latn_bpb: bits_per_byte + flores200_eng_Latn-nld_Latn_bpb: bits_per_byte + flores200_eng_Latn-nob_Latn_bpb: bits_per_byte + flores200_eng_Latn-pol_Latn_bpb: bits_per_byte + flores200_eng_Latn-por_Latn_bpb: bits_per_byte + flores200_eng_Latn-ron_Latn_bpb: bits_per_byte + flores200_eng_Latn-slk_Latn_bpb: bits_per_byte + flores200_eng_Latn-slv_Latn_bpb: bits_per_byte + flores200_eng_Latn-spa_Latn_bpb: bits_per_byte + flores200_eng_Latn-srp_Cyrl_bpb: bits_per_byte + flores200_eng_Latn-swe_Latn_bpb: bits_per_byte + flores200_eng_Latn-tur_Latn_bpb: bits_per_byte + flores200_eng_Latn-ukr_Cyrl_bpb: bits_per_byte + flores200_est_Latn-eng_Latn_bpb: bits_per_byte + flores200_eus_Latn-eng_Latn_bpb: bits_per_byte + flores200_fin_Latn-eng_Latn_bpb: bits_per_byte + flores200_fra_Latn-eng_Latn_bpb: bits_per_byte + flores200_gle_Latn-eng_Latn_bpb: bits_per_byte + flores200_glg_Latn-eng_Latn_bpb: bits_per_byte + flores200_hrv_Latn-eng_Latn_bpb: bits_per_byte + flores200_hun_Latn-eng_Latn_bpb: bits_per_byte + flores200_isl_Latn-eng_Latn_bpb: bits_per_byte + flores200_ita_Latn-eng_Latn_bpb: bits_per_byte + flores200_kat_Geor-eng_Latn_bpb: bits_per_byte + flores200_lit_Latn-eng_Latn_bpb: bits_per_byte + flores200_lvs_Latn-eng_Latn_bpb: bits_per_byte + flores200_mkd_Cyrl-eng_Latn_bpb: bits_per_byte + flores200_mlt_Latn-eng_Latn_bpb: bits_per_byte + flores200_nld_Latn-eng_Latn_bpb: bits_per_byte + flores200_nob_Latn-eng_Latn_bpb: bits_per_byte + flores200_pol_Latn-eng_Latn_bpb: bits_per_byte + flores200_por_Latn-eng_Latn_bpb: bits_per_byte + flores200_ron_Latn-eng_Latn_bpb: bits_per_byte + flores200_slk_Latn-eng_Latn_bpb: bits_per_byte + flores200_slv_Latn-eng_Latn_bpb: bits_per_byte + flores200_spa_Latn-eng_Latn_bpb: bits_per_byte + flores200_srp_Cyrl-eng_Latn_bpb: bits_per_byte + flores200_swe_Latn-eng_Latn_bpb: bits_per_byte + flores200_tur_Latn-eng_Latn_bpb: bits_per_byte + flores200_ukr_Cyrl-eng_Latn_bpb: bits_per_byte + belebele_bul_Cyrl_bpb: bits_per_byte + belebele_ces_Latn_bpb: bits_per_byte + belebele_dan_Latn_bpb: bits_per_byte + belebele_deu_Latn_bpb: bits_per_byte + belebele_ell_Grek_bpb: bits_per_byte + belebele_eng_Latn_bpb: bits_per_byte + belebele_est_Latn_bpb: bits_per_byte + belebele_fin_Latn_bpb: bits_per_byte + belebele_fra_Latn_bpb: bits_per_byte + belebele_hrv_Latn_bpb: bits_per_byte + belebele_hun_Latn_bpb: bits_per_byte + belebele_ita_Latn_bpb: bits_per_byte + belebele_lit_Latn_bpb: bits_per_byte + belebele_lvs_Latn_bpb: bits_per_byte + belebele_mlt_Latn_bpb: bits_per_byte + belebele_nld_Latn_bpb: bits_per_byte + belebele_nob_Latn_bpb: bits_per_byte + belebele_pol_Latn_bpb: bits_per_byte + belebele_por_Latn_bpb: bits_per_byte + belebele_ron_Latn_bpb: bits_per_byte + belebele_slk_Latn_bpb: bits_per_byte + belebele_slv_Latn_bpb: bits_per_byte + belebele_spa_Latn_bpb: bits_per_byte + belebele_swe_Latn_bpb: bits_per_byte + global_mmlu_cs_bpb: bits_per_byte + global_mmlu_de_bpb: bits_per_byte + global_mmlu_el_bpb: bits_per_byte + global_mmlu_en_bpb: bits_per_byte + global_mmlu_es_bpb: bits_per_byte + global_mmlu_fr_bpb: bits_per_byte + global_mmlu_he_bpb: bits_per_byte + global_mmlu_it_bpb: bits_per_byte + global_mmlu_lt_bpb: bits_per_byte + global_mmlu_nl_bpb: bits_per_byte + global_mmlu_pl_bpb: bits_per_byte + global_mmlu_pt_bpb: bits_per_byte + global_mmlu_ro_bpb: bits_per_byte + global_mmlu_ru_bpb: bits_per_byte + global_mmlu_sr_bpb: bits_per_byte + global_mmlu_sv_bpb: bits_per_byte + global_mmlu_tr_bpb: bits_per_byte + global_mmlu_uk_bpb: bits_per_byte + mgsm_direct_de_bpb: bits_per_byte + mgsm_direct_en_bpb: bits_per_byte + mgsm_direct_es_bpb: bits_per_byte + mgsm_direct_fr_bpb: bits_per_byte + xwinograd_en_bpb: bits_per_byte + xwinograd_fr_bpb: bits_per_byte + xwinograd_jp_bpb: bits_per_byte + xwinograd_pt_bpb: bits_per_byte + xwinograd_ru_bpb: bits_per_byte + xwinograd_zh_bpb: bits_per_byte + xcopa_et_bpb: bits_per_byte + xcopa_ht_bpb: bits_per_byte + xcopa_id_bpb: bits_per_byte + xcopa_it_bpb: bits_per_byte + xcopa_qu_bpb: bits_per_byte + xcopa_sw_bpb: bits_per_byte + xcopa_ta_bpb: bits_per_byte + xcopa_th_bpb: bits_per_byte + xcopa_tr_bpb: bits_per_byte + xcopa_vi_bpb: bits_per_byte + xcopa_zh_bpb: bits_per_byte + xstorycloze_ar_bpb: bits_per_byte + xstorycloze_en_bpb: bits_per_byte + xstorycloze_es_bpb: bits_per_byte + xstorycloze_eu_bpb: bits_per_byte + xstorycloze_hi_bpb: bits_per_byte + xstorycloze_id_bpb: bits_per_byte + xstorycloze_my_bpb: bits_per_byte + xstorycloze_ru_bpb: bits_per_byte + xstorycloze_sw_bpb: bits_per_byte + xstorycloze_te_bpb: bits_per_byte + xstorycloze_zh_bpb: bits_per_byte + include_base_44_albanian_bpb: bits_per_byte + include_base_44_armenian_bpb: bits_per_byte + include_base_44_azerbaijani_bpb: bits_per_byte + include_base_44_basque_bpb: bits_per_byte + include_base_44_belarusian_bpb: bits_per_byte + include_base_44_bulgarian_bpb: bits_per_byte + include_base_44_croatian_bpb: bits_per_byte + include_base_44_dutch_bpb: bits_per_byte + include_base_44_estonian_bpb: bits_per_byte + include_base_44_finnish_bpb: bits_per_byte + include_base_44_french_bpb: bits_per_byte + include_base_44_georgian_bpb: bits_per_byte + include_base_44_german_bpb: bits_per_byte + include_base_44_greek_bpb: bits_per_byte + include_base_44_hungarian_bpb: bits_per_byte + include_base_44_italian_bpb: bits_per_byte + include_base_44_lithuanian_bpb: bits_per_byte + include_base_44_north_macedonian_bpb: bits_per_byte + include_base_44_polish_bpb: bits_per_byte + include_base_44_portuguese_bpb: bits_per_byte + include_base_44_russian_bpb: bits_per_byte + include_base_44_serbian_bpb: bits_per_byte + include_base_44_spanish_bpb: bits_per_byte + include_base_44_turkish_bpb: bits_per_byte + include_base_44_ukrainian_bpb: bits_per_byte + mmlu: acc + mmlu_prox_cs: exact_match + mmlu_prox_de: exact_match + mmlu_prox_en: exact_match + mmlu_prox_es: exact_match + mmlu_prox_fr: exact_match + mmlu_prox_hu: exact_match + mmlu_prox_it: exact_match + mmlu_prox_pt: exact_match + mmlu_prox_sr: exact_match + mmlu_prox_uk: exact_match + sib200_bul_Cyrl: acc_norm + sib200_hrv_Latn: acc_norm + sib200_ces_Latn: acc_norm + sib200_dan_Latn: acc_norm + sib200_nld_Latn: acc_norm + sib200_eng_Latn: acc_norm + sib200_est_Latn: acc_norm + sib200_fin_Latn: acc_norm + sib200_fra_Latn: acc_norm + sib200_deu_Latn: acc_norm + sib200_ell_Grek: acc_norm + sib200_hun_Latn: acc_norm + sib200_gle_Latn: acc_norm + sib200_ita_Latn: acc_norm + sib200_lvs_Latn: acc_norm + sib200_lit_Latn: acc_norm + sib200_mlt_Latn: acc_norm + sib200_pol_Latn: acc_norm + sib200_por_Latn: acc_norm + sib200_ron_Latn: acc_norm + sib200_slk_Latn: acc_norm + sib200_slv_Latn: acc_norm + sib200_spa_Latn: acc_norm + sib200_swe_Latn: acc_norm + sib200_cat_Latn: acc_norm + sib200_eus_Latn: acc_norm + sib200_glg_Latn: acc_norm + sib200_bos_Latn: acc_norm + sib200_kat_Geor: acc_norm + sib200_mkd_Cyrl: acc_norm + sib200_als_Latn: acc_norm + sib200_srp_Cyrl: acc_norm + sib200_tur_Latn: acc_norm + sib200_ukr_Cyrl: acc_norm + sib200_isl_Latn: acc_norm + sib200_nob_Latn: acc_norm + belebele_bul_Cyrl_cf: acc_norm + belebele_hrv_Latn_cf: acc_norm + belebele_ces_Latn_cf: acc_norm + belebele_dan_Latn_cf: acc_norm + belebele_nld_Latn_cf: acc_norm + belebele_eng_Latn_cf: acc_norm + belebele_est_Latn_cf: acc_norm + belebele_fin_Latn_cf: acc_norm + belebele_fra_Latn_cf: acc_norm + belebele_deu_Latn_cf: acc_norm + belebele_ell_Grek_cf: acc_norm + belebele_hun_Latn_cf: acc_norm + belebele_ita_Latn_cf: acc_norm + belebele_lvs_Latn_cf: acc_norm + belebele_lit_Latn_cf: acc_norm + belebele_mlt_Latn_cf: acc_norm + belebele_pol_Latn_cf: acc_norm + belebele_por_Latn_cf: acc_norm + belebele_ron_Latn_cf: acc_norm + belebele_slk_Latn_cf: acc_norm + belebele_slv_Latn_cf: acc_norm + belebele_spa_Latn_cf: acc_norm + belebele_swe_Latn_cf: acc_norm + belebele_nob_Latn_cf: acc_norm + belebele_eus_Latn_cf: acc_norm + belebele_cat_Latn_cf: acc_norm + xcsqa_deu_Latn: acc_norm + xcsqa_eng_Latn: acc_norm + xcsqa_spa_Latn: acc_norm + xcsqa_fra_Latn: acc_norm + xcsqa_ita_Latn: acc_norm + xcsqa_nld_Latn: acc_norm + xcsqa_pol_Latn: acc_norm + xcsqa_por_Latn: acc_norm + pawsx_deu_Latn: acc + pawsx_eng_Latn: acc + pawsx_spa_Latn: acc + pawsx_fra_Latn: acc + xnli_bul_Cyrl: acc + xnli_deu_Latn: acc + xnli_ell_Grek: acc + xnli_eng_Latn: acc + xnli_spa_Latn: acc + xnli_fra_Latn: acc + copa: acc + lambada_openai: acc + openbookqa: acc_norm + winogrande: acc + arc_challenge: acc_norm + arc_easy: acc_norm + arc_challenge_mt_bg: acc_norm + arc_challenge_mt_cs: acc_norm + arc_challenge_mt_da: acc_norm + arc_challenge_mt_nl: acc_norm + arc_challenge_mt_et: acc_norm + arc_challenge_mt_fi: acc_norm + arc_challenge_mt_fr: acc_norm + arc_challenge_mt_de: acc_norm + arc_challenge_mt_el: acc_norm + arc_challenge_mt_hu: acc_norm + arc_challenge_mt_it: acc_norm + arc_challenge_mt_lv: acc_norm + arc_challenge_mt_lt: acc_norm + arc_challenge_mt_pl: acc_norm + arc_challenge_mt_pt: acc_norm + arc_challenge_mt_ro: acc_norm + arc_challenge_mt_sk: acc_norm + arc_challenge_mt_sl: acc_norm + arc_challenge_mt_es: acc_norm + arc_challenge_mt_sv: acc_norm + arc_challenge_mt_is: acc_norm + arc_challenge_mt_nb: acc_norm + boolq: acc + commonsense_qa: acc + # global-mgsm (CohereLabs/global-mgsm): flexible-extract exact_match + global_mgsm_ca: exact_match + global_mgsm_cs: exact_match + global_mgsm_de: exact_match + global_mgsm_el: exact_match + global_mgsm_en: exact_match + global_mgsm_es: exact_match + global_mgsm_eu: exact_match + global_mgsm_fr: exact_match + global_mgsm_gl: exact_match + global_mgsm_hu: exact_match + global_mgsm_sr: exact_match + hellaswag: acc_norm + piqa: acc_norm + social_iqa: acc + agieval_lsat_ar: acc + wsc273: acc + bigbench_language_identification_multiple_choice: acc + squadv2: f1 + coqa: f1 + bigbench_qa_wikidata_generate_until: exact_match + bigbench_dyck_languages_generate_until: exact_match + bigbench_operators_generate_until: exact_match + bigbench_repeat_copy_logic_generate_until: exact_match + bigbench_cs_algorithms_generate_until: exact_match + opensubtitles_multi40_en_to_bg: bleu + opensubtitles_multi40_en_to_hr: bleu + opensubtitles_multi40_en_to_cs: bleu + opensubtitles_multi40_en_to_da: bleu + opensubtitles_multi40_en_to_nl: bleu + opensubtitles_multi40_en_to_et: bleu + opensubtitles_multi40_en_to_fi: bleu + opensubtitles_multi40_en_to_fr: bleu + opensubtitles_multi40_en_to_de: bleu + opensubtitles_multi40_en_to_el: bleu + opensubtitles_multi40_en_to_hu: bleu + opensubtitles_multi40_en_to_it: bleu + opensubtitles_multi40_en_to_lv: bleu + opensubtitles_multi40_en_to_lt: bleu + opensubtitles_multi40_en_to_pl: bleu + opensubtitles_multi40_en_to_pt: bleu + opensubtitles_multi40_en_to_ro: bleu + opensubtitles_multi40_en_to_sk: bleu + opensubtitles_multi40_en_to_sl: bleu + opensubtitles_multi40_en_to_es: bleu + opensubtitles_multi40_en_to_sv: bleu + opensubtitles_multi40_en_to_sr: bleu + opensubtitles_multi40_en_to_tr: bleu + opensubtitles_multi40_en_to_uk: bleu + opensubtitles_multi40_en_to_no: bleu + opensubtitles_multi40_bg_to_en: bleu + opensubtitles_multi40_hr_to_en: bleu + opensubtitles_multi40_cs_to_en: bleu + opensubtitles_multi40_da_to_en: bleu + opensubtitles_multi40_nl_to_en: bleu + opensubtitles_multi40_et_to_en: bleu + opensubtitles_multi40_fi_to_en: bleu + opensubtitles_multi40_fr_to_en: bleu + opensubtitles_multi40_de_to_en: bleu + opensubtitles_multi40_el_to_en: bleu + opensubtitles_multi40_hu_to_en: bleu + opensubtitles_multi40_it_to_en: bleu + opensubtitles_multi40_lv_to_en: bleu + opensubtitles_multi40_lt_to_en: bleu + opensubtitles_multi40_pl_to_en: bleu + opensubtitles_multi40_pt_to_en: bleu + opensubtitles_multi40_ro_to_en: bleu + opensubtitles_multi40_sk_to_en: bleu + opensubtitles_multi40_sl_to_en: bleu + opensubtitles_multi40_es_to_en: bleu + opensubtitles_multi40_sv_to_en: bleu + opensubtitles_multi40_sr_to_en: bleu + opensubtitles_multi40_tr_to_en: bleu + opensubtitles_multi40_uk_to_en: bleu + opensubtitles_multi40_no_to_en: bleu + # flores200 (lighteval) translation: chrf++ is the primary metric + flores200:bul_Cyrl-eng_Latn: chrf++ + flores200:hrv_Latn-eng_Latn: chrf++ + flores200:ces_Latn-eng_Latn: chrf++ + flores200:dan_Latn-eng_Latn: chrf++ + flores200:nld_Latn-eng_Latn: chrf++ + flores200:est_Latn-eng_Latn: chrf++ + flores200:fin_Latn-eng_Latn: chrf++ + flores200:fra_Latn-eng_Latn: chrf++ + flores200:deu_Latn-eng_Latn: chrf++ + flores200:ell_Grek-eng_Latn: chrf++ + flores200:hun_Latn-eng_Latn: chrf++ + flores200:gle_Latn-eng_Latn: chrf++ + flores200:ita_Latn-eng_Latn: chrf++ + flores200:lvs_Latn-eng_Latn: chrf++ + flores200:lit_Latn-eng_Latn: chrf++ + flores200:mlt_Latn-eng_Latn: chrf++ + flores200:pol_Latn-eng_Latn: chrf++ + flores200:por_Latn-eng_Latn: chrf++ + flores200:ron_Latn-eng_Latn: chrf++ + flores200:slk_Latn-eng_Latn: chrf++ + flores200:slv_Latn-eng_Latn: chrf++ + flores200:spa_Latn-eng_Latn: chrf++ + flores200:swe_Latn-eng_Latn: chrf++ + flores200:cat_Latn-eng_Latn: chrf++ + flores200:eus_Latn-eng_Latn: chrf++ + flores200:glg_Latn-eng_Latn: chrf++ + flores200:bos_Latn-eng_Latn: chrf++ + flores200:kat_Geor-eng_Latn: chrf++ + flores200:mkd_Cyrl-eng_Latn: chrf++ + flores200:als_Latn-eng_Latn: chrf++ + flores200:srp_Cyrl-eng_Latn: chrf++ + flores200:tur_Latn-eng_Latn: chrf++ + flores200:ukr_Cyrl-eng_Latn: chrf++ + flores200:isl_Latn-eng_Latn: chrf++ + flores200:nob_Latn-eng_Latn: chrf++ + flores200:eng_Latn-bul_Cyrl: chrf++ + flores200:eng_Latn-hrv_Latn: chrf++ + flores200:eng_Latn-ces_Latn: chrf++ + flores200:eng_Latn-dan_Latn: chrf++ + flores200:eng_Latn-nld_Latn: chrf++ + flores200:eng_Latn-est_Latn: chrf++ + flores200:eng_Latn-fin_Latn: chrf++ + flores200:eng_Latn-fra_Latn: chrf++ + flores200:eng_Latn-deu_Latn: chrf++ + flores200:eng_Latn-ell_Grek: chrf++ + flores200:eng_Latn-hun_Latn: chrf++ + flores200:eng_Latn-gle_Latn: chrf++ + flores200:eng_Latn-ita_Latn: chrf++ + flores200:eng_Latn-lvs_Latn: chrf++ + flores200:eng_Latn-lit_Latn: chrf++ + flores200:eng_Latn-mlt_Latn: chrf++ + flores200:eng_Latn-pol_Latn: chrf++ + flores200:eng_Latn-por_Latn: chrf++ + flores200:eng_Latn-ron_Latn: chrf++ + flores200:eng_Latn-slk_Latn: chrf++ + flores200:eng_Latn-slv_Latn: chrf++ + flores200:eng_Latn-spa_Latn: chrf++ + flores200:eng_Latn-swe_Latn: chrf++ + flores200:eng_Latn-cat_Latn: chrf++ + flores200:eng_Latn-eus_Latn: chrf++ + flores200:eng_Latn-glg_Latn: chrf++ + flores200:eng_Latn-bos_Latn: chrf++ + flores200:eng_Latn-kat_Geor: chrf++ + flores200:eng_Latn-mkd_Cyrl: chrf++ + flores200:eng_Latn-als_Latn: chrf++ + flores200:eng_Latn-srp_Cyrl: chrf++ + flores200:eng_Latn-tur_Latn: chrf++ + flores200:eng_Latn-ukr_Cyrl: chrf++ + flores200:eng_Latn-isl_Latn: chrf++ + flores200:eng_Latn-nob_Latn: chrf++ + global_piqa_completions_als_latn: acc_norm + global_piqa_completions_bos_latn: acc_norm + global_piqa_completions_bul_cyrl: acc_norm + global_piqa_completions_cat_latn: acc_norm + global_piqa_completions_ces_latn: acc_norm + global_piqa_completions_deu_latn: acc_norm + global_piqa_completions_ekk_latn: acc_norm + global_piqa_completions_ell_grek: acc_norm + global_piqa_completions_eng_latn: acc_norm + global_piqa_completions_fin_latn: acc_norm + global_piqa_completions_fra_latn_fran: acc_norm + global_piqa_completions_glg_latn: acc_norm + global_piqa_completions_hrv_latn: acc_norm + global_piqa_completions_hun_latn: acc_norm + global_piqa_completions_isl_latn: acc_norm + global_piqa_completions_ita_latn: acc_norm + global_piqa_completions_kat_geor: acc_norm + global_piqa_completions_lit_latn: acc_norm + global_piqa_completions_mkd_cyrl: acc_norm + global_piqa_completions_nld_latn: acc_norm + global_piqa_completions_nno_latn: acc_norm + global_piqa_completions_nob_latn: acc_norm + global_piqa_completions_pol_latn: acc_norm + global_piqa_completions_por_latn_port: acc_norm + global_piqa_completions_ron_latn: acc_norm + global_piqa_completions_slk_latn: acc_norm + global_piqa_completions_slv_latn: acc_norm + global_piqa_completions_spa_latn_spai: acc_norm + global_piqa_completions_srp_cyrl: acc_norm + global_piqa_completions_swe_latn: acc_norm + global_piqa_completions_tur_latn: acc_norm + global_piqa_completions_ukr_cyrl: acc_norm + global_piqa_prompted_als_latn: exact_match + global_piqa_prompted_bos_latn: exact_match + global_piqa_prompted_bul_cyrl: exact_match + global_piqa_prompted_cat_latn: exact_match + global_piqa_prompted_ces_latn: exact_match + global_piqa_prompted_deu_latn: exact_match + global_piqa_prompted_ekk_latn: exact_match + global_piqa_prompted_ell_grek: exact_match + global_piqa_prompted_eng_latn: exact_match + global_piqa_prompted_fin_latn: exact_match + global_piqa_prompted_fra_latn_fran: exact_match + global_piqa_prompted_glg_latn: exact_match + global_piqa_prompted_hrv_latn: exact_match + global_piqa_prompted_hun_latn: exact_match + global_piqa_prompted_isl_latn: exact_match + global_piqa_prompted_ita_latn: exact_match + global_piqa_prompted_kat_geor: exact_match + global_piqa_prompted_lit_latn: exact_match + global_piqa_prompted_mkd_cyrl: exact_match + global_piqa_prompted_nld_latn: exact_match + global_piqa_prompted_nno_latn: exact_match + global_piqa_prompted_nob_latn: exact_match + global_piqa_prompted_pol_latn: exact_match + global_piqa_prompted_por_latn_port: exact_match + global_piqa_prompted_ron_latn: exact_match + global_piqa_prompted_slk_latn: exact_match + global_piqa_prompted_slv_latn: exact_match + global_piqa_prompted_spa_latn_spai: exact_match + global_piqa_prompted_srp_cyrl: exact_match + global_piqa_prompted_swe_latn: exact_match + global_piqa_prompted_tur_latn: exact_match + global_piqa_prompted_ukr_cyrl: exact_match + polymath_de_low: exact_match + polymath_de_medium: exact_match + polymath_de_high: exact_match + polymath_de_top: exact_match + polymath_en_low: exact_match + polymath_en_medium: exact_match + polymath_en_high: exact_match + polymath_en_top: exact_match + polymath_es_low: exact_match + polymath_es_medium: exact_match + polymath_es_high: exact_match + polymath_es_top: exact_match + polymath_fr_low: exact_match + polymath_fr_medium: exact_match + polymath_fr_high: exact_match + polymath_fr_top: exact_match + polymath_it_low: exact_match + polymath_it_medium: exact_match + polymath_it_high: exact_match + polymath_it_top: exact_match + polymath_pt_low: exact_match + polymath_pt_medium: exact_match + polymath_pt_high: exact_match + polymath_pt_top: exact_match + task_groups: sib200-eu: description: "SIB-200 European language topic classification tasks (0-shot)" @@ -588,6 +1145,254 @@ task_groups: - task: "polymath_{lang}_top" subset: "{lang}" + # ------------------------------------------------------------------ + # BPB (bits-per-byte) groups. + # + # These score the *gold* continuation's loglikelihood normalised by its + # UTF-8 byte length, instead of argmax accuracy. At swarm scale (30M-150M + # params, a few B tokens) accuracy on these benchmarks sits near chance and + # is dominated by noise, which makes it a poor regression target; BPB is + # continuous, strictly decreasing with model quality, and is what + # DCLM/OLMo/Olmix fit data-mixture regressors on. + # + # Task definitions live in custom_lm_eval_tasks/bpb/ and are picked up via + # the default --include_path. + # ------------------------------------------------------------------ + bpb-core: + description: "Bits-per-byte on the DCLM/Olmix core multiple-choice benchmarks (0-shot)" + suite: lm-eval-harness + n_shots: [0] + tasks: + - task: arc_easy_bpb + dataset: allenai/ai2_arc + subset: ARC-Easy + - task: arc_challenge_bpb + dataset: allenai/ai2_arc + subset: ARC-Challenge + - task: hellaswag_bpb + dataset: Rowan/hellaswag + - task: piqa_bpb + dataset: baber/piqa + - task: openbookqa_bpb + dataset: allenai/openbookqa + subset: main + - task: boolq_bpb + dataset: aps/super_glue + subset: boolq + - task: copa_bpb + dataset: aps/super_glue + subset: copa + - task: commonsense_qa_bpb + dataset: tau/commonsense_qa + - task: social_iqa_bpb + dataset: allenai/social_i_qa + - task: winogrande_bpb + dataset: allenai/winogrande + subset: winogrande_xl + + bpb-mmlu-stem: + description: "Bits-per-byte on MMLU STEM subjects, cloze form (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: cais/mmlu + tasks: + - task: mmlu_abstract_algebra_bpb + subset: abstract_algebra + - task: mmlu_anatomy_bpb + subset: anatomy + - task: mmlu_astronomy_bpb + subset: astronomy + - task: mmlu_college_biology_bpb + subset: college_biology + - task: mmlu_college_chemistry_bpb + subset: college_chemistry + - task: mmlu_college_computer_science_bpb + subset: college_computer_science + - task: mmlu_college_mathematics_bpb + subset: college_mathematics + - task: mmlu_college_physics_bpb + subset: college_physics + - task: mmlu_computer_security_bpb + subset: computer_security + - task: mmlu_conceptual_physics_bpb + subset: conceptual_physics + - task: mmlu_electrical_engineering_bpb + subset: electrical_engineering + - task: mmlu_elementary_mathematics_bpb + subset: elementary_mathematics + - task: mmlu_high_school_biology_bpb + subset: high_school_biology + - task: mmlu_high_school_chemistry_bpb + subset: high_school_chemistry + - task: mmlu_high_school_computer_science_bpb + subset: high_school_computer_science + - task: mmlu_high_school_mathematics_bpb + subset: high_school_mathematics + - task: mmlu_high_school_physics_bpb + subset: high_school_physics + - task: mmlu_high_school_statistics_bpb + subset: high_school_statistics + - task: mmlu_machine_learning_bpb + subset: machine_learning + + # ------------------------------------------------------------------ + # Multilingual BPB groups: the BPB counterparts of the oellm-multilingual + # super group, with the same conventions as the English BPB groups above + # (0-shot; multiple-choice exams in cloze form). FLORES and MGSM score a + # single reference (translation / final answer number) instead of a choice. + # Task definitions live in custom_lm_eval_tasks/bpb//. + # ------------------------------------------------------------------ + flores-200-eu-to-eng-bpb: + description: "Bits-per-byte of FLORES-200 EU->English reference translations (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: facebook/flores + valid_langs: [bul_Cyrl, hrv_Latn, ces_Latn, dan_Latn, nld_Latn, est_Latn, + fin_Latn, fra_Latn, deu_Latn, ell_Grek, hun_Latn, gle_Latn, + ita_Latn, lvs_Latn, lit_Latn, mlt_Latn, pol_Latn, por_Latn, + ron_Latn, slk_Latn, slv_Latn, spa_Latn, swe_Latn, cat_Latn, + eus_Latn, glg_Latn, bos_Latn, kat_Geor, mkd_Cyrl, als_Latn, + srp_Cyrl, tur_Latn, ukr_Cyrl, isl_Latn, nob_Latn] + tasks: + - task: "flores200_{lang}-eng_Latn_bpb" + subset: "{lang}-eng_Latn" + + flores-200-eng-to-eu-bpb: + description: "Bits-per-byte of FLORES-200 English->EU reference translations (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: facebook/flores + valid_langs: [bul_Cyrl, hrv_Latn, ces_Latn, dan_Latn, nld_Latn, est_Latn, + fin_Latn, fra_Latn, deu_Latn, ell_Grek, hun_Latn, gle_Latn, + ita_Latn, lvs_Latn, lit_Latn, mlt_Latn, pol_Latn, por_Latn, + ron_Latn, slk_Latn, slv_Latn, spa_Latn, swe_Latn, cat_Latn, + eus_Latn, glg_Latn, bos_Latn, kat_Geor, mkd_Cyrl, als_Latn, + srp_Cyrl, tur_Latn, ukr_Cyrl, isl_Latn, nob_Latn] + tasks: + - task: "flores200_eng_Latn-{lang}_bpb" + subset: "eng_Latn-{lang}" + + belebele-eu-bpb: + description: "Bits-per-byte on Belebele European languages, cloze form (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: facebook/belebele + valid_langs: [bul_Cyrl, hrv_Latn, ces_Latn, dan_Latn, nld_Latn, eng_Latn, est_Latn, fin_Latn, + fra_Latn, deu_Latn, ell_Grek, hun_Latn, ita_Latn, lvs_Latn, lit_Latn, mlt_Latn, + pol_Latn, por_Latn, ron_Latn, slk_Latn, slv_Latn, spa_Latn, swe_Latn, nob_Latn] + tasks: + - task: "belebele_{lang}_bpb" + subset: "{lang}" + + global-mmlu-eu-bpb: + description: "Bits-per-byte on Global-MMLU (all subjects per language), cloze form (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: CohereForAI/Global-MMLU + valid_langs: [cs, de, el, en, es, fr, it, lt, nl, pl, pt, ro, ru, sr, sv, tr, uk, he] + tasks: + - task: "global_mmlu_{lang}_bpb" + subset: "{lang}" + + mgsm-eu-bpb: + description: "Bits-per-byte of the MGSM final answer after the direct prompt (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: jbross-ibm-research/mgsm + valid_langs: [en, de, es, fr] + tasks: + - task: "mgsm_direct_{lang}_bpb" + subset: "{lang}" + + # XWinograd, XCOPA and XStoryCloze together mirror generic-multilingual, split + # per benchmark so each task carries its language. + xwinograd-bpb: + description: "Bits-per-byte on XWinograd (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: Muennighoff/xwinograd + valid_langs: [en, fr, jp, pt, ru, zh] + tasks: + - task: "xwinograd_{lang}_bpb" + subset: "{lang}" + + xcopa-bpb: + description: "Bits-per-byte on XCOPA (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: cambridgeltl/xcopa + valid_langs: [et, ht, id, it, qu, sw, ta, th, tr, vi, zh] + tasks: + - task: "xcopa_{lang}_bpb" + subset: "{lang}" + + xstorycloze-bpb: + description: "Bits-per-byte on XStoryCloze (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: juletxara/xstory_cloze + valid_langs: [ar, en, es, eu, hi, id, my, ru, sw, te, zh] + tasks: + - task: "xstorycloze_{lang}_bpb" + subset: "{lang}" + + include-bpb: + description: "Bits-per-byte on INCLUDE (all domains per language), cloze form (0-shot)" + suite: lm-eval-harness + n_shots: [0] + dataset: CohereForAI/include-base-44 + tasks: + - task: include_base_44_albanian_bpb + subset: Albanian + - task: include_base_44_armenian_bpb + subset: Armenian + - task: include_base_44_azerbaijani_bpb + subset: Azerbaijani + - task: include_base_44_basque_bpb + subset: Basque + - task: include_base_44_belarusian_bpb + subset: Belarusian + - task: include_base_44_bulgarian_bpb + subset: Bulgarian + - task: include_base_44_croatian_bpb + subset: Croatian + - task: include_base_44_dutch_bpb + subset: Dutch + - task: include_base_44_estonian_bpb + subset: Estonian + - task: include_base_44_finnish_bpb + subset: Finnish + - task: include_base_44_french_bpb + subset: French + - task: include_base_44_georgian_bpb + subset: Georgian + - task: include_base_44_german_bpb + subset: German + - task: include_base_44_greek_bpb + subset: Greek + - task: include_base_44_hungarian_bpb + subset: Hungarian + - task: include_base_44_italian_bpb + subset: Italian + - task: include_base_44_lithuanian_bpb + subset: Lithuanian + - task: include_base_44_north_macedonian_bpb + subset: North Macedonian + - task: include_base_44_polish_bpb + subset: Polish + - task: include_base_44_portuguese_bpb + subset: Portuguese + - task: include_base_44_russian_bpb + subset: Russian + - task: include_base_44_serbian_bpb + subset: Serbian + - task: include_base_44_spanish_bpb + subset: Spanish + - task: include_base_44_turkish_bpb + subset: Turkish + - task: include_base_44_ukrainian_bpb + subset: Ukrainian + super_groups: oellm-multilingual: description: "Combined Belebele EU set plus multilingual benchmarks" @@ -599,6 +1404,25 @@ super_groups: - task: mgsm-eu - task: generic-multilingual - task: include + + bpb-all: + description: "All BPB benchmarks: core multiple-choice plus MMLU STEM" + task_groups: + - task: bpb-core + - task: bpb-mmlu-stem + + oellm-multilingual-bpb: + description: "Bits-per-byte counterparts of the oellm-multilingual benchmarks" + task_groups: + - task: flores-200-eu-to-eng-bpb + - task: flores-200-eng-to-eu-bpb + - task: belebele-eu-bpb + - task: global-mmlu-eu-bpb + - task: mgsm-eu-bpb + - task: xwinograd-bpb + - task: xcopa-bpb + - task: xstorycloze-bpb + - task: include-bpb - task: multiblimp-eu - task: xcopa-eu diff --git a/oellm/task_groups.py b/oellm/task_groups.py index 8e5a846b..223376c6 100644 --- a/oellm/task_groups.py +++ b/oellm/task_groups.py @@ -49,6 +49,20 @@ "ca": "cat_Latn", "eu": "eus_Latn", "gl": "glg_Latn", + # non-EU languages of xwinograd / xcopa / xstorycloze + "ar": "arb_Arab", + "hi": "hin_Deva", + "ht": "hat_Latn", + "id": "ind_Latn", + "jp": "jpn_Jpan", + "my": "mya_Mymr", + "qu": "que_Latn", + "sw": "swh_Latn", + "ta": "tam_Taml", + "te": "tel_Telu", + "th": "tha_Thai", + "vi": "vie_Latn", + "zh": "zho_Hans", # openai/MMMLU locale codes (lang_REGION) "de_de": "deu_Latn", "es_la": "spa_Latn", @@ -148,18 +162,23 @@ def _canonical_language(code: str | None) -> str | None: return None +_FLORES_PAIR = re.compile( + r"^flores200[:_]([a-z]{3}_[A-Z][a-z]{3})-([a-z]{3}_[A-Z][a-z]{3})" +) + + def _resolve_task_languages(name: str, subset: str | None) -> list[str]: """Return the canonical language code(s) a task belongs to, if any. - Translation pairs (``flores200:src-tgt`` and + Translation pairs (``flores200:src-tgt``, ``flores200_src-tgt_bpb`` and ``opensubtitles_multi40_src_to_tgt``) resolve to their non-English side; every other task resolves via its ``subset``. Tasks with no recognisable language (e.g. English-only standard benchmarks) return []. """ - if name.startswith("flores200:"): - pair = name.split(":", 1)[1] + flores = _FLORES_PAIR.match(name) + if flores: langs = [ - _canonical_language(part) for part in pair.split("-") if part != "eng_Latn" + _canonical_language(part) for part in flores.groups() if part != "eng_Latn" ] return [lang for lang in langs if lang] if name.startswith("opensubtitles_multi40_"): diff --git a/tests/test_bpb_multilingual.py b/tests/test_bpb_multilingual.py new file mode 100644 index 00000000..5bac808e --- /dev/null +++ b/tests/test_bpb_multilingual.py @@ -0,0 +1,180 @@ +"""Tests for the multilingual BPB tasks (the ``oellm-multilingual-bpb`` super group). + +The registry checks guard against drift between task-groups.yaml and the task +YAMLs in custom_lm_eval_tasks/bpb/; the scoring checks pin the byte convention +documented in bpb/utils.py on synthetic documents (no datasets or models). +""" + +import importlib.util +import math +import re +from importlib.resources import files + +import pytest +import yaml + +from oellm.task_groups import _expand_task_groups + +BPB_DIR = files("oellm.resources") / "custom_lm_eval_tasks" / "bpb" + + +def _load_utils(): + spec = importlib.util.spec_from_file_location("bpb_utils", str(BPB_DIR / "utils.py")) + module = importlib.util.module_from_spec(spec) + spec.loader.exec_module(module) + return module + + +utils = _load_utils() + + +def _defined_bpb_tasks() -> set[str]: + names = set() + for path in BPB_DIR.rglob("*.yaml"): + match = re.search(r"^task: (\S+)$", path.read_text(encoding="utf-8"), re.M) + if match: + names.add(match.group(1)) + return names + + +# --- registry ----------------------------------------------------------------- + + +def test_super_group_expands_to_every_benchmark(): + jobs = _expand_task_groups(["oellm-multilingual-bpb"]) + assert len(jobs) == 169 + assert {j.suite for j in jobs} == {"lm-eval-harness"} + assert {j.n_shot for j in jobs} == {0} + + +def test_every_scheduled_task_is_defined_and_reports_bpb(): + task_metrics = yaml.safe_load( + (files("oellm.resources") / "task-groups.yaml").read_text() + )["task_metrics"] + defined = _defined_bpb_tasks() + for job in _expand_task_groups(["oellm-multilingual-bpb"]): + assert job.task in defined, f"{job.task} has no YAML in custom_lm_eval_tasks/bpb" + assert task_metrics.get(job.task) == "bits_per_byte", job.task + + +def test_super_group_bracket_mirrors_oellm_multilingual(): + """Scoping to one language selects that language's BPB counterpart of each + oellm-multilingual benchmark (the non-BPB test expects the same set).""" + tasks = {j.task for j in _expand_task_groups(["oellm-multilingual-bpb[deu_Latn]"])} + assert tasks == { + "belebele_deu_Latn_bpb", + "flores200_deu_Latn-eng_Latn_bpb", + "flores200_eng_Latn-deu_Latn_bpb", + "global_mmlu_de_bpb", + "include_base_44_german_bpb", + "mgsm_direct_de_bpb", + } + + +def test_template_functions_exist_in_utils(): + for path in BPB_DIR.glob("_*_template_yaml"): + for fn in re.findall(r"!function utils\.(\w+)", path.read_text(encoding="utf-8")): + assert callable(getattr(utils, fn, None)), f"{path.name}: utils.{fn}" + + +def test_task_yamls_in_subdirectories_use_no_functions(): + """`!function utils.x` resolves next to the YAML that declares it, so only + the templates beside bpb/utils.py may use it.""" + for path in BPB_DIR.glob("*/*.yaml"): + assert "!function" not in path.read_text(encoding="utf-8"), path + + +# --- scoring ------------------------------------------------------------------ + + +def _mc(*lls): + return [(ll, False) for ll in lls] + + +def test_belebele_scores_gold_answer_text(): + doc = { + "mc_answer1": "a", + "mc_answer2": "é", + "mc_answer3": "猫", + "mc_answer4": "d", + "correct_answer_num": "3", + } + out = utils.process_results_belebele(doc, _mc(-4, -3, -1, -5)) + assert out == {"acc": 1.0, "acc_norm": 1.0, "bits_per_byte": (-1.0, 3)} + + +def test_global_mmlu_letter_answer(): + doc = {"option_a": "x", "option_b": "réponse", "option_c": "y", "option_d": "z"} + out = utils.process_results_global_mmlu({**doc, "answer": "B"}, _mc(-4, -1, -2, -3)) + assert out["bits_per_byte"] == (-1.0, len("réponse".encode())) + malformed = utils.process_results_global_mmlu( + {**doc, "answer": ""}, _mc(-4, -1, -2, -3) + ) + assert malformed["bits_per_byte"] == (0.0, 1) + + +def test_include_integer_answer(): + doc = {"option_a": "x", "option_b": "y", "option_c": "Antwort", "option_d": "z"} + out = utils.process_results_include({**doc, "answer": 2}, _mc(-4, -3, -1, -5)) + assert out["bits_per_byte"] == (-1.0, 7) + + +def test_xcopa_lowercases_choice_like_upstream(): + doc = {"choice1": "See oli õrn.", "choice2": "Ärge.", "label": 1} + assert utils.doc_to_choice_xcopa(doc) == ["see oli õrn.", "ärge."] + out = utils.process_results_xcopa(doc, _mc(-2, -1)) + assert out["bits_per_byte"] == (-1.0, len("ärge.".encode())) + + +def test_xstorycloze_one_indexed_ending(): + doc = {"sentence_quiz1": "a", "sentence_quiz2": "猫", "answer_right_ending": 2} + assert utils.process_results_xstorycloze(doc, _mc(-2, -1))["bits_per_byte"] == ( + -1.0, + 3, + ) + + +def test_xwinograd_counts_shared_continuation(): + doc = { + "sentence": "_ était ici", + "option1": "Elle", + "option2": "La personne", + "answer": "1", + } + out = utils.process_results_winogrande(doc, _mc(-1, -2)) + assert out["bits_per_byte"] == (-1.0, len("était ici".encode())) + + +def test_mgsm_excludes_the_space_it_prepends(): + doc = {"answer_number": 18.0} + assert utils.doc_to_target_mgsm(doc) == " 18" + assert utils.process_results_mgsm(doc, [(-2.0, False)]) == { + "bits_per_byte": (-2.0, 2) + } + + +def test_flores_scores_reference_in_both_directions(): + doc = {"sentence_deu_Latn": "Grüße", "sentence_eng_Latn": "Greetings"} + assert utils.doc_to_target_flores_xx_eng(doc) == " Greetings" + assert utils.process_results_flores_xx_eng(doc, [(-3.0, False)]) == { + "bits_per_byte": (-3.0, 9) + } + assert utils.doc_to_target_flores_eng_xx(doc) == " Grüße" + assert utils.process_results_flores_eng_xx(doc, [(-3.0, False)]) == { + "bits_per_byte": (-3.0, len("Grüße".encode())) + } + + +def test_pairs_aggregate_to_corpus_level_bpb(): + """lm-eval's bits_per_byte aggregation is byte-weighted: -sum(ll)/sum(bytes)/ln 2.""" + pairs = [ + utils.process_results_flores_xx_eng( + {"sentence_deu_Latn": "x", "sentence_eng_Latn": "a"}, [(-math.log(2), False)] + )["bits_per_byte"], + utils.process_results_flores_xx_eng( + {"sentence_deu_Latn": "x", "sentence_eng_Latn": "猫"}, + [(-6 * math.log(2), False)], + )["bits_per_byte"], + ] + lls, n_bytes = zip(*pairs, strict=True) + assert -sum(lls) / sum(n_bytes) / math.log(2) == pytest.approx(1.75) diff --git a/tests/test_language_groups.py b/tests/test_language_groups.py index 16a81cee..8ab2c457 100644 --- a/tests/test_language_groups.py +++ b/tests/test_language_groups.py @@ -30,6 +30,7 @@ "xcsqa-eu", "pawsx-eu", "xnli-eu", + "include-bpb", "mmmlu-eu", ]