fix: fp8 quantization failure of qwen 2.5 VL 7B model (#10112)

Signed-off-by: PanJason <pyyjason@gmail.com>
This commit is contained in:
Yueyang Pan
2025-09-27 05:05:23 +00:00
committed by GitHub
parent 37f3325b06
commit 8260574729
5 changed files with 81 additions and 14 deletions
+1 -1
View File
@@ -265,7 +265,7 @@ class Qwen2_5_VisionTransformer(nn.Module):
self.fullatt_block_indexes = vision_config.fullatt_block_indexes
self.window_size = vision_config.window_size
self.patch_size = vision_config.patch_size
mlp_hidden_size: int = vision_config.intermediate_size
mlp_hidden_size: int = ((vision_config.intermediate_size + 7) // 8) * 8
self.patch_embed = Qwen2_5_VisionPatchEmbed(
patch_size=patch_size,
temporal_patch_size=temporal_patch_size,