Skip to content

GitLab

Unverified Commit a258ad8b authored by Jinzhen Lin's avatar Jinzhen Lin Committed by GitHub
Browse files

[Bugfix] fix qwen3 moe fp8 accuracy issue (#23031) 


Signed-off-by: default avatarJinzhen Lin <jinzhen.ljz@antgroup.com>
parent bf7f470b
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 0 deletions
vllm/model_executor/layers/quantization/fp8.py
View file @ a258ad8b
...@@ -125,6 +125,10 @@ class Fp8Config(QuantizationConfig): ...@@ -125,6 +125,10 @@ class Fp8Config(QuantizationConfig):
ignored_layers = cls.get_from_keys_or(config, ["ignored_layers"], None) ignored_layers = cls.get_from_keys_or(config, ["ignored_layers"], None)
weight_block_size = cls.get_from_keys_or(config, ["weight_block_size"], weight_block_size = cls.get_from_keys_or(config, ["weight_block_size"],
None) None)
if not ignored_layers:
ignored_layers = cls.get_from_keys_or(config,
["modules_to_not_convert"],
None)
return cls(is_checkpoint_fp8_serialized=is_checkpoint_fp8_serialized, return cls(is_checkpoint_fp8_serialized=is_checkpoint_fp8_serialized,
activation_scheme=activation_scheme, activation_scheme=activation_scheme,
ignored_layers=ignored_layers, ignored_layers=ignored_layers,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment