Skip to content

Add interactive metadata editor and LoRA scaling#2

Merged
harelc merged 1 commit into
lora_analysisfrom
metadata_and_scaling
May 3, 2026
Merged

Add interactive metadata editor and LoRA scaling#2
harelc merged 1 commit into
lora_analysisfrom
metadata_and_scaling

Conversation

@harelc
Copy link
Copy Markdown
Owner

@harelc harelc commented Mar 6, 2026

Summary

  • Interactive metadata editor (m key): replaces the read-only metadata view with a field-by-field editor — edit (e), add (a), delete (d), save (Ctrl+S). Saves in-place.
  • LoRA scaling (a key from LoRA analysis): scale all lora_A tensors by an alpha coefficient. Always writes to a new file ({name}_x{alpha}.safetensors).
  • Backend functions: rewrite_with_metadata() and scale_lora_a() in data.py.

Test plan

  • Open a safetensors file, press m, verify metadata fields are shown in a table
  • Edit a field, add a field, delete a field, save with Ctrl+S, reopen to verify persistence
  • Open LoRA analysis (l), press a, enter an alpha value, verify output file is created with correct name
  • Verify original file is never modified by scaling

🤖 Generated with Claude Code

- Replace read-only MetadataScreen with interactive editor:
  field-by-field view, edit (e), add (a), delete (d), save (Ctrl+S)
- Add ScaleScreen for scaling LoRA pairs by alpha coefficient
  (accessible via 'a' key from LoRA analysis view)
- Add scale_lora_a() and rewrite_with_metadata() to data.py
- Load sigma0 from cached JSON on startup

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@harelc harelc merged commit 64ba0e1 into lora_analysis May 3, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant