Skip to content
View Hayabusa-YL's full-sized avatar
  • Joined Oct 28, 2026

Block or report Hayabusa-YL

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Hayabusa-YL/README.md

Hi, I'm Hayabusa

专注 AI Agent 的工程落地与成本优化。

作品

  • longbench-harness —— LongBench v1 900 题 Agent 评测全量跑完。官方判分口径（qa_f1 / rouge-l / retrieval）：passage_retrieval 100%、hotpotqa 71.04、qasper 49.40、gov_report 23.14；1.02 亿 token 按样例口径折算约 8.6 元，全程零崩溃。
  • agent-cost-lens —— 会话成本审计工具：逐轮成本归因、缓存命中率、按天明细、压缩仿真。样例轨迹单步 API 成本 0.026 → 0.0043 元（降 84%），缓存命中率 99%。

我在找

AI Agent 方向远程实习。大三在读，时间窗口完整；擅长把 Agent 的 API 账单算清楚、压下去，并把系统跑稳。

如果你们团队在招 agent 工程方向，或者想把 Agent 费用砍下来，欢迎直接聊。

联系

Popular repositories Loading

  1. agent-cost-lens agent-cost-lens Public

    把 Agent 会话流水算成钱：花费、缓存命中、每轮均价

    JavaScript

  2. longbench-harness longbench-harness Public

    LongBench 900-item full run: self-built retrieval/compression harness on deepseek-flash. Scripts, logs and results.

    JavaScript

  3. Hayabusa-YL Hayabusa-YL Public

    Personal profile — AI Agent engineering & cost optimization

  4. click click Public

    Forked from pallets/click

    Python composable command line interface toolkit

    Python

  5. vllm-8gb-whitepaper vllm-8gb-whitepaper Public

    8GB 显存跑 7B 大模型的 vLLM 实战白皮书：显存拆解、吞吐实测、三个上游 issue 复现与修复

    Python

  6. vllm vllm Public

    Forked from vllm-project/vllm

    A high-throughput and memory-efficient inference and serving engine for LLMs

    Python