423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-02 03:32:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

[scripts] specify model class for qwen_omni merge (#8227)

Browse Source
This commit is contained in:
Kingsley 2025-05-30 14:20:12 +08:00 committed by GitHub
parent be02003d16
commit e31afdfd4f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 16 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
scripts/qwen_omni_merge.py
View File

@ -17,7 +17,11 @@ import shutil
import fire import fire
from peft import PeftModel from peft import PeftModel
from transformers import AutoModel, AutoProcessor, Qwen2_5OmniThinkerForConditionalGeneration # type: ignore from transformers import (
AutoProcessor,
Qwen2_5OmniForConditionalGeneration, # type: ignore
Qwen2_5OmniThinkerForConditionalGeneration,
)
def merge_lora( def merge_lora(
@ -27,7 +31,7 @@ def merge_lora(
submodule_name: str = "thinker", submodule_name: str = "thinker",
save_path: str = "./merged_model_checkpoint", save_path: str = "./merged_model_checkpoint",
): ):
"""Load the original model, tokenizer, and processor configuration, merge the LoRA weights. """Load the original model, merge the LoRA weights.
For a specified submodule, and save the final merged model along with its configurations. For a specified submodule, and save the final merged model along with its configurations.
@ -38,10 +42,9 @@ def merge_lora(
submodule_name (str): Name of the submodule to merge (default: "thinker"). submodule_name (str): Name of the submodule to merge (default: "thinker").
save_path (str): Directory where the merged model and configurations will be saved. save_path (str): Directory where the merged model and configurations will be saved.
""" """
# 1. Load the original model, tokenizer, and processor # 1. Load the original model
model = AutoModel.from_pretrained(base_model_path, torch_dtype="auto", device_map="cpu") model = Qwen2_5OmniForConditionalGeneration.from_pretrained(base_model_path, torch_dtype="auto", device_map="cpu")
processor = AutoProcessor.from_pretrained(base_model_path) print("Successfully loaded the original model.")
print("Successfully loaded the original model and tokenizer.")
# 2. Extract the submodule to be merged (e.g., model.thinker) # 2. Extract the submodule to be merged (e.g., model.thinker)
if not hasattr(model, submodule_name): if not hasattr(model, submodule_name):
@ -52,7 +55,8 @@ def merge_lora(
# 3. Load the LoRA weights onto the extracted submodule # 3. Load the LoRA weights onto the extracted submodule
lora_model = PeftModel.from_pretrained(base_submodule, lora_checkpoint_path) lora_model = PeftModel.from_pretrained(base_submodule, lora_checkpoint_path)
print("LoRA weights loaded successfully.") processor = AutoProcessor.from_pretrained(lora_checkpoint_path)
print("LoRA weights and processor loaded successfully.")
# 4. Merge the LoRA weights into the submodule and unload the LoRA modules # 4. Merge the LoRA weights into the submodule and unload the LoRA modules
merged_submodule = lora_model.merge_and_unload() merged_submodule = lora_model.merge_and_unload()
@ -95,14 +99,16 @@ def save_full_model(
thinker = Qwen2_5OmniThinkerForConditionalGeneration.from_pretrained( thinker = Qwen2_5OmniThinkerForConditionalGeneration.from_pretrained(
saved_thinker_path, torch_dtype="auto", device_map="cpu" saved_thinker_path, torch_dtype="auto", device_map="cpu"
) )
base_model = AutoModel.from_pretrained(base_model_path, torch_dtype="auto", device_map="cpu") base_model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
base_model_path, torch_dtype="auto", device_map="cpu"
)
base_model.thinker = thinker base_model.thinker = thinker
# 2. Save the complete model along with its tokenizer and processor configuration # 2. Save the complete model along with its tokenizer and processor configuration
processor = AutoProcessor.from_pretrained(base_model_path) processor = AutoProcessor.from_pretrained(saved_thinker_path)
base_model.save_pretrained(save_path) base_model.save_pretrained(save_path)
processor.save_pretrained(save_path) processor.save_pretrained(save_path)
print(f"Merged model and tokenizer saved to {save_path}.") print(f"Merged model and processor saved to {save_path}.")
# 3. Copy the extra file from the base model directory to the save_path # 3. Copy the extra file from the base model directory to the save_path
source_file = os.path.join(base_model_path, extra_file) source_file = os.path.join(base_model_path, extra_file)