Skip to content
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

[Misc] Optimize Qwen2-VL LoRA test #11663

Merged
merged 4 commits into from
Jan 1, 2025
Merged

[Misc] Optimize Qwen2-VL LoRA test #11663

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
ccb8ef8
Init
jeejeelee Dec 31, 2024
16e8e28
format
jeejeelee Dec 31, 2024
2ac660b
Done
jeejeelee Jan 1, 2025
46a082c
Fix typo
jeejeelee Jan 1, 2025
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
5 changes: 2 additions & 3 deletions tests/lora/test_qwen2vl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
from vllm.lora.request import LoRARequest
from vllm.platforms import current_platform

MODEL_PATH = "Qwen/Qwen2-VL-7B-Instruct"
MODEL_PATH = "Qwen/Qwen2-VL-2B-Instruct"

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I switched to the 2B model to reduce training time and decrease the computational pressure on CI testing pipeline.

PROMPT_TEMPLATE = (
"<|im_start|>system\nYou are a helpful assistant.<|im_end|>"
Expand Down Expand Up @@ -49,10 +49,9 @@ def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> List[str]:
# Print the outputs.
generated_texts: List[str] = []
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text.strip()
generated_texts.append(generated_text)
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
print(f"Generated text: {generated_text!r}")
return generated_texts


Expand Down
20 changes: 19 additions & 1 deletion vllm/model_executor/models/qwen2_vl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,7 @@
GPTQMarlinConfig)
from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.model_executor.models.module_mapping import MultiModelKeys
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.inputs import (ImageItem, ModalityData,
MultiModalFieldConfig, MultiModalKwargs,
Expand Down Expand Up @@ -925,15 +926,23 @@ class Qwen2VLForConditionalGeneration(nn.Module, SupportsMultiModal,
}

# LoRA specific attributes
# TODO Support LoRA for the visual encoder in the future.
supported_lora_modules = [
"qkv_proj",
"o_proj",
"gate_up_proj",
"down_proj",
# vison tower
DarkLight1337 marked this conversation as resolved.
Show resolved Hide resolved
"qkv",
"attn.proj", # Distinguish patch_embed.proj
"fc1",
"fc2",
# projector
"mlp.0",
"mlp.2"
]
embedding_modules = {}
embedding_padding_modules = []

# To ensure correct weight loading and mapping.
hf_to_vllm_mapper = WeightsMapper(orig_to_new_prefix={
"lm_head.": "language_model.lm_head.",
Expand Down Expand Up @@ -1230,3 +1239,12 @@ def load_weights(self, weights: Iterable[Tuple[str,

loader = AutoWeightsLoader(self)
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)

def get_mm_mapping(self) -> MultiModelKeys:
"""
Get the module prefix in multimodal models
"""
return MultiModelKeys.from_string_field(
language_model="language_model",
connector="visual.",
tower_model="visual.merger.")
Loading