Skip to content
View philipshurpik's full-sized avatar

Block or report philipshurpik

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. Markdown is supported. 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

Pinned Loading

  1. claude-code-status-line claude-code-status-line Public

    Color-coded status line for Claude Code

    JavaScript 12 4

  2. Relink Claude Team and Personal acco... Relink Claude Team and Personal accounts with old (default) account
    1
    #!/usr/bin/env bash
    2
    # Script to symlink two Claude accounts (e.g. team and personal one) to your old default "$HOME/.claude" folder
    3
    # 1) You need to first create folders, e.g.:
    4
    # mkdir -p ~/.claude-team ~/.claude-personal
    5
    # 2) Create team and personal aliases in ~/.zprofile :
  3. Resume eval lottery on multi-GPU RTX... Resume eval lottery on multi-GPU RTX PRO 6000 (sm_120): identical relaunches give different evals
    1
    """Resume eval lottery on multi-GPU RTX PRO 6000 (sm_120): identical relaunches give different evals.
    2
    
                  
    3
    Run the same command every time (first run trains 10 steps and saves; every later run resumes):
    4
        torchrun --nproc_per_node=4 repro_lora.py
    5
    
                  
  4. vLLM patch: Gemma4 MTP draft layers ... vLLM patch: Gemma4 MTP draft layers inherit target KV-cache scales — fixes silent speculative-decoding slowdown on calibrated-KV checkpoints (e.g. unsloth NVFP4). Apply: patch -p1 -d <site-packages> < gemma4-mtp-kv-scale-inherit.patch
    1
    Subject: [PATCH] Inherit target KV-cache scales in Gemma4 MTP draft layers
    2
    
                  
    3
    The Gemma4 MTP drafter shares the target's physical KV cache (draft layers
    4
    map onto target layers 58/59), but its Attention objects keep their own
    5
    _k_scale/_v_scale, which stay at 1.0 because assistant checkpoints carry no