Skip to content

[misc] fix model merger #479

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
hiyouga merged 1 commit into from
Aug 23, 2025
+8 −8
Merged

[misc] fix model merger #479

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Binary file modified assets/wechat.jpg
View file
Open in desktop
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
6 changes: 3 additions & 3 deletions scripts/model_merger.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,7 @@
AutoConfig,
AutoModelForCausalLM,
AutoModelForTokenClassification,
AutoModelForVision2Seq,
AutoModelForImageTextToText,
PretrainedConfig,
PreTrainedModel,
)
Expand Down Expand Up @@ -165,10 +165,10 @@ def process_one_shard(rank, model_state_dict_lst):

if "ForTokenClassification" in architectures[0]:
AutoClass = AutoModelForTokenClassification
elif "ForConditionalGeneration" in architectures[0]:
AutoClass = AutoModelForImageTextToText
elif "ForCausalLM" in architectures[0]:
AutoClass = AutoModelForCausalLM
elif "ForConditionalGeneration" in architectures[0]:
AutoClass = AutoModelForVision2Seq
else:
raise NotImplementedError(f"Unknown architecture {architectures}.")

Expand Down
10 changes: 5 additions & 5 deletions verl/workers/fsdp_workers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,14 +191,14 @@ def _build_model_optimizer(
torch_dtype = PrecisionType.to_dtype(fsdp_config.torch_dtype)

if role == "critic":
auto_class = AutoModelForTokenClassification
AutoClass = AutoModelForTokenClassification
elif type(self.model_config) in AutoModelForImageTextToText._model_mapping.keys():
auto_class = AutoModelForImageTextToText
AutoClass = AutoModelForImageTextToText
else:
auto_class = AutoModelForCausalLM
AutoClass = AutoModelForCausalLM

if (not fsdp_config.enable_rank0_init) or self.device_mesh.get_local_rank("fsdp") == 0:
model = auto_class.from_pretrained(
model = AutoClass.from_pretrained(
model_config.model_path,
config=self.model_config,
torch_dtype=torch_dtype,
Expand All @@ -209,7 +209,7 @@ def _build_model_optimizer(
)
else:
with no_init_weights(), init_empty_weights():
model = auto_class.from_config(
model = AutoClass.from_config(
self.model_config,
torch_dtype=torch_dtype,
attn_implementation="flash_attention_2",
Expand Down