Cosmos3 FP8 ModelOpt/Diffusers remapping (#48952)
Signed-off-by: Wojciech Kutak <wkutak@nvidia.com> Signed-off-by: wkutak <wkutak@nvidia.com> Signed-off-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
co-authored by
Isotr0py
Roger Wang
parent
f878367898
commit
9dd62d80ab
@@ -37,6 +37,11 @@ class Cosmos3ForConditionalGeneration(Qwen3VLForConditionalGeneration):
|
||||
".to_out.": ".o_proj.",
|
||||
".norm_q.": ".q_norm.",
|
||||
".norm_k.": ".k_norm.",
|
||||
# ModelOpt-native dialect (diffusers/transformers read these; vLLM reads
|
||||
# weight_scale/input_scale instead), drop so AutoWeightsLoader passes
|
||||
".input_quantizer.": None,
|
||||
".weight_quantizer.": None,
|
||||
".output_quantizer.": None,
|
||||
},
|
||||
orig_to_new_prefix={
|
||||
"proj_in.": None,
|
||||
|
||||
Reference in New Issue
Block a user