12月1月2月3月4月5月6月7月8月9月10月11月
周一周三周五
一年内 265876 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/gguf-experimentation 分支比较 6 提交 »68d14c68c6 Get alternative GGUF implementation working... barely.776f274ed5 Initial experimentation with Tensor-like extension for GGUF.01b088712d recognize .gguf files when scanning a folder for importb1999e2b52 Run Ruff0a1a0cac28 Add unit tests for torch patcher
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/gguf-cleanup 分支比较 10 提交 »bc300bc498 Add a compute_dtype field to GGMLTensor.c95a8d2a3c Add unit tests for GGMLTensor.b31ba61d23 Fix type errors in GGMLTensor.579eb8718e Remove no longer used code paths, general cleanup of new dequantization code, update probeb4c5210902 Run ruff and update imports
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/flux-t5-bnb 分支比较 10 提交 »70def37280 Move quantization scripts to a scripts/ subdir.8af3c72de7 Update docs for T5 quantization script.6405214940 Remove all references to optimum-quanto and downgrade diffusers.544ab296e7 Update the T5 8-bit quantized starter model to use the BnB LLM.int8() variant.86e49c423c Fixes to the T5XXL quantization script.
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/flux-tg-simplified-noise-guidance 分支6a289e9f40 experiments...
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/flux-remove-einops 分支ffa89126d1 WIP experimentation with replacing einops with native torch ops for faster execution. More investigation needed.
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/flux-min-t5-seq-len 分支比较 4 提交 »437d1087a2 Dynamically select smaller t5 seq len to save inference time.a76a1244bc Avoid unnecessary dtype conversions with rope encodings.debc84cf95 Use view() instead of rearrange() for better performance.4000dd1843 Replace custom RMSNorm implementation with torch.nn.functional.rms_norm(...) for improved speed.
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/flux-lora-with-te3 分支711f47280c Add test state dict shape for FLUX kohya LoRA models with both te1 and te3 layers.
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/flux-lora-sidecar 分支比较 10 提交 »3ce02be1ea Add links to test models for loha, lokr, ia3.53da78f1ac Update all lycoris layer types to use the new torch.nn.Module base class.7246e007b7 Assume LoRA alpha=8 for FLUX diffusers PEFT LoRAs.5d081aad54 Get diffusers FLUX LoRA working as sidecar patch on quantized model.5b41d74bce WIP - Implement sidecar LoRA layers using functional API.
- frostbyte_neo 推送了仓库 github-featured/invoke-ai--invokeai 的 ryan/flux-lora-non-quantized 分支比较 10 提交 »2ed86c082a Assume LoRA alpha=8 for FLUX diffusers PEFT LoRAs.41ff0a5af6 Consolidate all LoRA patching logic in the LoRAPatcher.c2ec65d582 lora_layer_from_state_dict(...) -> any_lora_layer_from_state_dict(...)a0bede02f4 Rename peft/ -> lora/8f3c09348d Genera cleanup/documentation.