Conversation
This reverts commit 2bec5b3.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Load diffusers pipelines component-by-component from the local modl store instead of falling back to HuggingFace downloads. Supports fp8 quantized models via layerwise casting. Tested working for Flux-dev (fp8), Z-Image Turbo, and SDXL. - Add assemble_pipeline() with per-component loading from store - Bundle diffusers/transformers config files for offline pipeline init - Rich gen_components in arch_config (model_class, config_dir, model_id) - resolve_gen_assembly() maps model IDs to local store paths - detect_model_format() via safetensors header inspection - fp8 inference: load bf16 then enable_layerwise_casting(fp8 storage) - Text encoders: init_empty_weights + load_state_dict for zero-copy load - Fix preflight to accept optional_variant dependencies (e.g. t5-xxl-fp8) - Add pipeline-loading-strategy spec doc Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix Chroma: pipeline_class → ChromaPipeline, text encoder → T5-XXL (was incorrectly z-image-text-encoder/Qwen3), add ChromaTransformer2DModel - Add Qwen-Image gen_components: QwenImagePipeline, QwenImageTransformer2DModel, Qwen2_5_VLForConditionalGeneration, AutoencoderKLQwenImage - Bundle config files for chroma-transformer, chroma-scheduler, qwen-image-transformer, qwen-image-vae, qwen-image-text-encoder, qwen-image-tokenizer, qwen-image-scheduler - Add _detect_weight_dtype() — reads safetensors header to show fp8/bf16/fp16 - Log each component with filename + dtype during pipeline loading - Embed model_files (file + dtype per component) in PNG metadata - Include base_model_path in Rust DB artifact metadata Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Flux 2 is BFL's new 32B model with a completely different architecture: - Single text encoder (Mistral Small 3.1 24B) instead of CLIP-L + T5-XXL - New VAE (AutoencoderKLFlux2) with patch-based encoding - Flux2Transformer2DModel: 8 double + 48 single stream blocks - Uses AutoProcessor (PixtralProcessor) instead of regular tokenizer Changes: - Add flux2 arch_config with full gen_components - Add Flux2 to MODEL_REGISTRY (black-forest-labs/FLUX.2-dev) - Add BaseModelFamily::Flux2 on Rust side (presets, generate defaults) - Add flux2-dev to gated models list (HuggingFace auth required) - Bundle configs: flux2-dev-transformer, flux2-vae, flux2-dev-scheduler, flux2-text-encoder (Mistral3), flux2-processor (PixtralProcessor) - Support HF directory loading path in assemble_pipeline (for quantized text encoders like NF4 Mistral3) - Default params: 28 steps, guidance 4.0 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…der) - Add flux2_klein architecture with Flux2KleinPipeline (not Flux2Pipeline) - ComfyUI fp8 dequantization: apply weight_scale before stripping scale tensors - Skip enable_layerwise_casting for dequantized weights (already bf16) - Add pipeline_kwargs pass-through for is_distilled=True - Add Klein transformer + scheduler config files - Add Klein detection in CLI default_steps (4) and default_guidance (1.0) - Fix flux2 text encoder model_id to flux2-mistral-text-encoder Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
build.rsfor frontend build integrationDraft — UI is work in progress.
🤖 Generated with Claude Code