Skip to content

  • Projects
  • Groups
  • Snippets
  • Help
    • Loading...
    • Help
    • Contribute to GitLab
  • Sign in
C
coderai
  • Project
    • Project
    • Details
    • Activity
    • Cycle Analytics
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
    • Charts
  • Issues 0
    • Issues 0
    • List
    • Board
    • Labels
    • Milestones
  • Merge Requests 0
    • Merge Requests 0
  • CI / CD
    • CI / CD
    • Pipelines
    • Jobs
    • Schedules
    • Charts
  • Wiki
    • Wiki
  • Snippets
    • Snippets
  • Members
    • Members
  • Collapse sidebar
  • Activity
  • Graph
  • Charts
  • Create a new issue
  • Jobs
  • Commits
  • Issue Boards
  • nexlab
  • coderai
  • Repository

Switch branch/tag
  • coderai
  • codai
  • models
  • quant.py
Find file
BlameHistoryPermalink
  • Stefy Lanza (nextime / spora )'s avatar
    quant: reject checkpoints whose weights weren't actually quantized · c741ff5b
    Stefy Lanza (nextime / spora ) authored Jun 18, 2026
    GPTQModel silently leaves layers it can't map (e.g. gemma-4's fused batched MoE
    experts) in bf16, producing a near-full-size "checkpoint" that the loader would
    redirect to and then offload. The worker now scans the saved safetensors and, if
    <50% of large weight bytes are int-packed, deletes the output and marks the job
    failed (so it falls back to bitsandbytes) instead of reporting "done".
    Co-Authored-By: 's avatarClaude Opus 4.8 <noreply@anthropic.com>
    c741ff5b
quant.py 14 KB
EditWeb IDE

Replace quant.py

Attach a file by drag & drop or click to upload


Cancel
A new branch will be created in your fork and a new merge request will be started.