This commit introduces `fast_sampler.py`, a new module designed to enhance the performance of sampling and VAE decoding in ComfyUI. It replaces or augments functionality previously handled in `model_management.py`, providing better VRAM management, FP16 support, and tiled decoding for low-memory scenarios. Additionally, it improves the `PreviewImage` node in `nodes.py` for faster and more efficient preview generation. These changes improve efficiency, stability, and usability, particularly for GPU-based workflows.
**Key Changes:**
- Implemented `fast_ksampler` for optimized sampling with improved memory management, FP16 support via `torch.amp.autocast`, and `channels_last` memory format for better GPU performance.
- Added `fast_vae_decode` for efficient VAE decoding, incorporating FP16 support, `channels_last`, and selective VRAM clearing to prevent out-of-memory errors.
- Introduced `fast_vae_tiled_decode` for tiled VAE decoding, enabling processing of large latents on GPUs with limited VRAM by using configurable tile sizes and overlaps.
- Added profiling and debugging utilities (`profile_section`, `profile_cuda_sync`) to track execution times and VRAM usage when `--profile` or `--debug` flags are enabled.
- Improved VRAM management with `clear_vram`, ensuring sufficient free memory before loading models or VAE, with configurable thresholds and minimum free memory requirements.
- Implemented `is_fp16_safe` to check GPU compatibility for FP16 operations, disabling them on unsupported hardware (e.g., GTX 1660/Turing).
- Optimized tensor transfers with `optimized_transfer` and `optimized_conditioning` for synchronous device placement and dtype casting.
- Enhanced model preloading with `preload_model`, which unloads VAE before loading U-Net to conserve VRAM and checks for already-loaded VAE to avoid redundant transfers.
- Integrated `cudnn.benchmark` for for tests, disabled by default.
- VRAM should now be managed efficiently.
- Updated `PreviewImage` node in `nodes.py` to support adaptive resizing of preview images to a maximum dimension of ~512 pixels while preserving aspect ratio, using `Image.LANCZOS` for quality. Increased `compress_level` from 1 to 4 for faster PNG compression, optimizing preview generation.
**Impact:**
- Significantly reduces VRAM usage during sampling and VAE decoding, making ComfyUI more stable on GPUs with limited memory.
- Improves performance for large-scale image generation through tiled decoding and FP16 optimizations.
- Enhances debugging capabilities with detailed profiling and logging, aiding development and optimization.
**Dependencies:**
- Relies on `nodes.py` for integration with `KSampler`, `VAEDecode`, `VAEDecodeTiled`, and `PreviewImage` nodes.
- Assumes compatibility with existing `ModelPatcher` functionality for model patching (e.g., in `LoraLoader`).
**Notes:**
- Users should enable `--profile` or `--debug` flags to access detailed performance logs.
- FP16 support requires compatible GPU hardware (compute capability ≥ 8 or > 7).
- Tiled decoding parameters (`tile_size`, `overlap`, etc.) may need tuning for specific workflows.
- Preview images are now smaller and faster to generate, but users can adjust `max_size` in `PreviewImage` if higher resolution previews are needed.
This is a foundational change to improve ComfyUI's performance and scalability, particularly for resource-constrained environments.
Thanks to Grok @ xAI for help.
This commit fixes the temporal tile size calculation, and removes
a redundant tile at the end of the range when its elements are
completely covered by the previous tile.
Co-authored-by: Andrew Kvochko <a.kvochko@lightricks.com>
* fix attention OOM in xformers
* allow passing attention mask in flux attention
* allow an attn_mask in flux
* attn masks can be done using replace patches instead of a separate dict
* fix return types
* fix return order
* enumerate
* patch the right keys
* arg names
* fix a silly bug
* fix xformers masks
* replace match with if, elif, else
* mask with image_ref_size
* remove unused import
* remove unused import 2
* fix pytorch/xformers attention
This corrects a weird inconsistency with skip_reshape.
It also allows masks of various shapes to be passed, which will be
automtically expanded (in a memory-efficient way) to a size that is
compatible with xformers or pytorch sdpa respectively.
* fix mask shapes
To use:
"Load CLIP" node with t5xxl + type mochi
"Load Diffusion Model" node with the mochi dit file.
"Load VAE" with the mochi vae file.
EmptyMochiLatentVideo node for the latent.
euler + linear_quadratic in the KSampler node.
Previously when a list of 3 images [0, 1, 2] was used for a 6 frame video
they were concated like this:
[0, 1, 2, 0, 1, 2]
now they are concated like this:
[0, 0, 1, 1, 2, 2]
