Commit Graph

29 Commits

Author SHA1 Message Date
comfyanonymous f3a36e7484
Temporarily disable auto enabling triton by default on AMD. (#14878)
I get freezing issues on my test machine.
2026-07-10 18:37:59 -07:00
liminfei-amd 1377a2f729
Only auto-enable the ROCm comfy-kitchen Triton backend on matrix-core GPUs (#14869)
#14862 auto-enables the comfy-kitchen Triton backend whenever torch.version.hip
is set and Triton >= 3.7. The INT8 matmul kernels compile tl.dot to matrix-core
instructions (WMMA on RDNA3+/gfx11xx-gfx12xx, MFMA on CDNA/gfx9xx); RDNA1/RDNA2
(gfx10xx) have neither, so the auto-enabled INT8 path hangs the GPU there
(reported on RDNA2 + triton-windows 3.7.1: native and custom-node INT8 freeze
until reset).

Gate the automatic ROCm default on GPU architecture as well as Triton version so
RDNA1/RDNA2 stay on the working eager fallback. Add --disable-triton-backend as
an explicit override; --enable-triton-backend still force-enables on any arch.
2026-07-10 03:31:20 -07:00
liminfei-amd 099522f85b
Enable comfy-kitchen Triton backend by default on ROCm/AMD (#14862)
On AMD/ROCm the CUDA backend is unavailable, so Triton is the only accelerated
comfy-kitchen backend. It was disabled by default (opt-in --enable-triton-backend),
leaving AMD on the slow eager path. Enable it by default when torch.version.hip is
set AND Triton is >= 3.7 -- older Triton lacks libdevice.rint on the HIP backend and
hard-crashes the INT8 path, so on Triton < 3.7 it stays disabled with a log line.
NVIDIA behavior is unchanged; the explicit --enable-triton-backend flag still works
as an override.

Fixes #14861
2026-07-09 23:11:52 -04:00
comfyanonymous 73e84d5ec8
Support convrot int4 models. (#14859)
linear_dtype in comfy_quant metadata can be used to set if the int4 op does
the matrix multiplication in int8 or int4, the default is int4 on GPUs that
support it with fallback to int8 for GPUs that don't.
2026-07-09 18:57:09 -04:00
comfyanonymous 1a510f0423
Support int8 models. (#14636) 2026-06-25 11:23:58 -07:00
Silver b138133ffa
Enable triton comfy kitchen via cli-arg (#12730) 2026-05-03 14:07:21 -04:00
Jukka Seppänen 1c5db7397d
feat: Support mxfp8 (#12907) 2026-03-14 18:36:29 -04:00
comfyanonymous 6165c38cb5
Optimize nvfp4 lora applying. (#11866)
This changes results a bit but it also speeds up things a lot.
2026-01-14 00:49:38 -05:00
comfyanonymous b3c0e4de57
Make loras work on nvfp4 models. (#11837)
The initial applying is a bit slow but will probably be sped up in the
future.
2026-01-12 22:33:54 -05:00
comfyanonymous 21e8425087
Add warning for old pytorch. (#11718) 2026-01-07 21:07:26 -05:00
comfyanonymous edee33f55e
Disable comfy kitchen cuda if pytorch cuda less than 13 (#11681) 2026-01-06 22:13:43 -05:00
comfyanonymous 6da00dd899
Initial ops changes to use comfy_kitchen: Initial nvfp4 checkpoint support. (#11635)
---------

Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>
2026-01-05 21:48:58 -05:00
comfyanonymous 791e30ff50
Fix nan issue when quantizing fp16 tensor. (#11213) 2025-12-09 17:03:21 -05:00
comfyanonymous 43071e3de3
Make old scaled fp8 format use the new mixed quant ops system. (#11000) 2025-12-05 14:35:42 -05:00
Urle Sistiana 6484ac89dc
fix QuantizedTensor.is_contiguous (#10956) (#10959) 2025-11-28 16:33:07 -05:00
rattus 3f382a4f98
quant ops: Dequantize weight in-place (#10935)
In flux2 these weights are huge (200MB). As plain_tensor is a throw-away
deep copy, do this multiplication in-place to save VRAM.
2025-11-27 08:06:30 -08:00
comfyanonymous bdb10a583f
Fix loras not working on mixed fp8. (#10899) 2025-11-26 00:07:58 -05:00
comfyanonymous 015a0599d0
I found a case where this is needed (#10875) 2025-11-25 03:23:19 -05:00
comfyanonymous b6805429b9
Allow pinning quantized tensors. (#10873) 2025-11-25 02:48:20 -05:00
comfyanonymous 25022e0b09
Cleanup and fix issues with text encoder quants. (#10872) 2025-11-25 01:48:53 -05:00
contentis 3b3ef9a77a
Quantized Ops fixes (#10715)
* offload support, bug fixes, remove mixins

* add readme
2025-11-12 18:26:52 -05:00
comfyanonymous af4b7b5edb
More fp8 torch.compile regressions fixed. (#10625) 2025-11-03 22:14:20 -05:00
comfyanonymous 6b88478f9f
Bring back fp8 torch compile performance to what it should be. (#10622) 2025-11-03 19:22:10 -05:00
comfyanonymous e199c8cc67
Fixes (#10621) 2025-11-03 17:58:24 -05:00
comfyanonymous 958a17199a
People should update their pytorch versions. (#10618) 2025-11-03 17:08:30 -05:00
comfyanonymous c58c13b2ba
Fix torch compile regression on fp8 ops. (#10580) 2025-11-01 00:25:17 -04:00
comfyanonymous 906c089957
Fix small performance regression with fp8 fast and scaled fp8. (#10537) 2025-10-29 19:29:01 -04:00
comfyanonymous 1a58087ac2
Reduce memory usage for fp8 scaled op. (#10531) 2025-10-29 15:43:51 -04:00
contentis 8817f8fc14
Mixed Precision Quantization System (#10498)
* Implement mixed precision operations with a registry design and metadate for quant spec in checkpoint.

* Updated design using Tensor Subclasses

* Fix FP8 MM

* An actually functional POC

* Remove CK reference and ensure correct compute dtype

* Update unit tests

* ruff lint

* Implement mixed precision operations with a registry design and metadate for quant spec in checkpoint.

* Updated design using Tensor Subclasses

* Fix FP8 MM

* An actually functional POC

* Remove CK reference and ensure correct compute dtype

* Update unit tests

* ruff lint

* Fix missing keys

* Rename quant dtype parameter

* Rename quant dtype parameter

* Fix unittests for CPU build
2025-10-28 16:20:53 -04:00