- LoadFFTNetModel / FFTNetGenerate ComfyUI nodes - FFTNet architecture (SpectralFilter, LocalWindowMixing, FFTNetMixer) - upgrade/: replace LLM self-attention with FFTNet + MPS-safe fine-tuning script - Auto-detects self_attn return count (1/2/3) for transformers version compat - MPS FFT CPU offload workaround for Apple Silicon Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
9 lines
263 B
Plaintext
9 lines
263 B
Plaintext
torch>=2.0.0
|
|
|
|
# Upgrade 파인튜닝에 필요 (standalone 추론만 할 경우 불필요)
|
|
transformers>=4.40.0
|
|
datasets>=2.18.0 # --dataset wikitext 사용 시
|
|
|
|
# Optional: tiktoken (transformers 없을 때 GPT-2 토크나이저 fallback)
|
|
tiktoken>=0.5.0
|