nick/comfyui-serverless
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix RTX 5090 crash: use sdpa attention instead of sageattn
Some checks failed
Build and Push Docker Image / build (push) Has been cancelled
Details

Browse Source 
SageAttention was only compiled for A100 (sm80) and H100 (sm90).
RTX 5090 (Blackwell sm120) has no compatible kernel, causing ComfyUI
to crash during generation with "Connection reset by peer".

Switch to PyTorch's native SDPA which works on all architectures.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Debian
2026-01-11 03:10:06 +00:00
parent 3c421cf7b8
commit 1e60401679
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
workflows/Wan22-I2V-Remix-API.json
View File

@@ -83,7 +83,7 @@
"base_precision": "fp16_fast", "base_precision": "fp16_fast",
"quantization": "disabled", "quantization": "disabled",
"load_device": "offload_device", "load_device": "offload_device",
"attention_mode": "sageattn", "attention_mode": "sdpa",
"rms_norm_function": "default", "rms_norm_function": "default",
"block_swap_args": [ "block_swap_args": [
"128", "128",
@@ -101,7 +101,7 @@
"base_precision": "fp16_fast", "base_precision": "fp16_fast",
"quantization": "disabled", "quantization": "disabled",
"load_device": "offload_device", "load_device": "offload_device",
"attention_mode": "sageattn", "attention_mode": "sdpa",
"rms_norm_function": "default", "rms_norm_function": "default",
"block_swap_args": [ "block_swap_args": [
"128", "128",