Fixes#5: Add compatibility shim for find_pruneable_heads_and_indices
which was removed in transformers 4.40.0. All 12 GPT2 model files now
use a try/except fallback.
Fixes#6: Add ultra low memory mode for 12GB VRAM systems.
- New memory_mode dropdown: auto, normal, low, ultra
- Auto mode detects available VRAM and selects best mode
- Ultra mode uses ~6GB peak VRAM with aggressive cleanup
- Added VRAM monitoring and logging during generation
Bump version to 1.3.0
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Replace block-style ASCII art with compact braille-style "FL SongGen" banner for consistent branding across FL node packs.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
New features:
- Add songgeneration_v1_5_beta model variant (20.5GB, multilingual)
- Supports Chinese, English, Spanish, Japanese languages
- Requires 24GB+ VRAM (experimental)
Bug fixes:
- Fix progress bar not matching console output (fixes token count mismatch)
- Progress bar now uses actual total from model instead of pre-calculated estimate
- Applied fix to generate, style_transfer, and auto_style nodes
Bump version to 1.2.0
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add _resize_embeddings_for_checkpoint() helper method
- Apply embedding resize logic to low-mem mode (fixes#4)
- Handles tokenizer version differences between checkpoint and model
- Bump version to 1.1.2
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add patch for systems without torch.distributed support (fixes#3)
- Creates dummy ReduceOp class before dac/audiotools imports
- Allows AMD GPU users to load the node without AttributeError
- Bump version to 1.1.1
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add `packaging` to requirements.txt (fixes#1 - pkg_resources import error)
- Add `x-transformers` to requirements.txt (required by stable_audio_tools)
- Bump version to 1.1.0
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add _download_demucs_model() function to download htdemucs.pth on demand
- Auto-download from HuggingFace when load_separator() is called
- Required for style transfer functionality
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Filter out GenerationMixin inheritance warnings (not needed for inference)
- Filter out deprecated checkpointing format warnings (training-only)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Version bump to 1.0.2 for compatibility fixes release
- Remove transformers <4.50.0 constraint (now compatible with newer versions)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Remove unused flashy imports from codeclm/utils/utils.py
- Add SequenceSummary fallback stub for transformers 4.50+ compatibility
- Remove transformers version assertion in levo.py (bundled llama works with newer versions)
- Add conf/vocab.yaml with pinyin tokens for OmegaConf resolver
- Fix OmegaConf load_yaml resolver to use package-relative paths
- Add embedding layer resizing to handle tokenizer vocabulary mismatches
- Fix relative path resolution for model checkpoints and third_party paths
- Remove flashy from requirements.txt
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
