From 735bb4bdb186bd4f39b9c924c24b8b39a7ef8b0d Mon Sep 17 00:00:00 2001 From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com> Date: Fri, 8 Aug 2025 01:21:00 -0700 Subject: [PATCH 1/7] Users report gfx1201 is buggy on flux with pytorch attention. (#9244) --- comfy/model_management.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/comfy/model_management.py b/comfy/model_management.py index 9e6149d60..dc5b4711d 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@ -321,9 +321,9 @@ try: if torch_version_numeric >= (2, 7): # works on 2.6 but doesn't actually seem to improve much if any((a in arch) for a in ["gfx90a", "gfx942", "gfx1100", "gfx1101", "gfx1151"]): # TODO: more arches, TODO: gfx950 ENABLE_PYTORCH_ATTENTION = True - if torch_version_numeric >= (2, 8): - if any((a in arch) for a in ["gfx1201"]): - ENABLE_PYTORCH_ATTENTION = True +# if torch_version_numeric >= (2, 8): +# if any((a in arch) for a in ["gfx1201"]): +# ENABLE_PYTORCH_ATTENTION = True if torch_version_numeric >= (2, 7) and rocm_version >= (6, 4): if any((a in arch) for a in ["gfx1201", "gfx942", "gfx950"]): # TODO: more arches SUPPORT_FP8_OPS = True From 5828607ccfef82a82931d8b66f3fd1176e04588f Mon Sep 17 00:00:00 2001 From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com> Date: Sat, 9 Aug 2025 09:49:25 -0700 Subject: [PATCH 2/7] Not sure if AMD actually support fp16 acc but it doesn't crash. (#9258) --- comfy/model_management.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/comfy/model_management.py b/comfy/model_management.py index dc5b4711d..c08f759e5 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@ -340,7 +340,7 @@ if ENABLE_PYTORCH_ATTENTION: PRIORITIZE_FP16 = False # TODO: remove and replace with something that shows exactly which dtype is faster than the other try: - if is_nvidia() and PerformanceFeature.Fp16Accumulation in args.fast: + if (is_nvidia() or is_amd()) and PerformanceFeature.Fp16Accumulation in args.fast: torch.backends.cuda.matmul.allow_fp16_accumulation = True PRIORITIZE_FP16 = True # TODO: limit to cards where it actually boosts performance logging.info("Enabled fp16 accumulation.") From 0552de7c7d6bcdd515da115d6756fd30494c7ff4 Mon Sep 17 00:00:00 2001 From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com> Date: Sun, 10 Aug 2025 02:03:47 -0700 Subject: [PATCH 3/7] Bump pytorch cuda and rocm versions in readme instructions. (#9273) --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 119098f5c..e4cff01a9 100644 --- a/README.md +++ b/README.md @@ -203,7 +203,7 @@ Put your VAE in: models/vae ### AMD GPUs (Linux only) AMD users can install rocm and pytorch with pip if you don't have it already installed, this is the command to install the stable version: -```pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.3``` +```pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.4``` This is the command to install the nightly with ROCm 6.4 which might have some performance improvements: @@ -237,7 +237,7 @@ Additional discussion and help can be found [here](https://github.com/comfyanony Nvidia users should install stable pytorch using this command: -```pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu128``` +```pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu129``` This is the command to install pytorch nightly instead which might have performance improvements. From 966f3a52061b5e300f36c6de0d07c47d6ad12f76 Mon Sep 17 00:00:00 2001 From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com> Date: Mon, 11 Aug 2025 02:53:01 -0700 Subject: [PATCH 4/7] Only show feature flags log when verbose. (#9281) --- server.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/server.py b/server.py index 0553a0dd7..8f9c88ebf 100644 --- a/server.py +++ b/server.py @@ -235,7 +235,7 @@ class PromptServer(): sid, ) - logging.info( + logging.debug( f"Feature flags negotiated for client {sid}: {client_flags}" ) first_message = False From fa340add552497a264071fd7f6c407ff4aa10449 Mon Sep 17 00:00:00 2001 From: Alexander Piskun <13381981+bigcat88@users.noreply.github.com> Date: Mon, 11 Aug 2025 23:48:17 +0300 Subject: [PATCH 5/7] remove creation of non-used asyncio_loop (#9284) --- execution.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/execution.py b/execution.py index 952f0cc5c..1dc35738b 100644 --- a/execution.py +++ b/execution.py @@ -646,8 +646,6 @@ class PromptExecutor: self.add_message("execution_error", mes, broadcast=False) def execute(self, prompt, prompt_id, extra_data={}, execute_outputs=[]): - asyncio_loop = asyncio.new_event_loop() - asyncio.set_event_loop(asyncio_loop) asyncio.run(self.execute_async(prompt, prompt_id, extra_data, execute_outputs)) async def execute_async(self, prompt, prompt_id, extra_data={}, execute_outputs=[]): From 629b17383718e1f46dbba101ea83ec897fbe3082 Mon Sep 17 00:00:00 2001 From: ComfyUI Wiki Date: Tue, 12 Aug 2025 04:52:12 +0800 Subject: [PATCH 6/7] Update template & embedded docs (#9283) * Update template & embedded docs * Update embedded docs to 0.2.6 --- requirements.txt | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/requirements.txt b/requirements.txt index 2f4692b03..2fb38ef27 100644 --- a/requirements.txt +++ b/requirements.txt @@ -1,6 +1,6 @@ comfyui-frontend-package==1.24.4 -comfyui-workflow-templates==0.1.52 -comfyui-embedded-docs==0.2.4 +comfyui-workflow-templates==0.1.53 +comfyui-embedded-docs==0.2.6 torch torchsde torchvision From 2208aa616d3ad193cd37ef57076d4f5243cecdd3 Mon Sep 17 00:00:00 2001 From: PsychoLogicAu Date: Tue, 12 Aug 2025 06:56:16 +1000 Subject: [PATCH 7/7] Support SimpleTuner lycoris lora for Qwen-Image (#9280) --- comfy/lora.py | 1 + 1 file changed, 1 insertion(+) diff --git a/comfy/lora.py b/comfy/lora.py index 6686b7229..00358884b 100644 --- a/comfy/lora.py +++ b/comfy/lora.py @@ -301,6 +301,7 @@ def model_lora_keys_unet(model, key_map={}): key_map["{}".format(key_lora)] = k # Support transformer prefix format key_map["transformer.{}".format(key_lora)] = k + key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = k #SimpleTuner lycoris format return key_map