简介 | Overview

这是 CyberSpark2-3b-cabin-lora-v2 的可用 LoRA 适配器仓库，包含可直接挂载到 Qwen2.5-VL-3B-Instruct 的 adapter 权重。

This is the usable LoRA adapter repository for CyberSpark2-3b-cabin-lora-v2, containing adapter weights that can be loaded directly on top of Qwen2.5-VL-3B-Instruct.

包含文件 | Included Files

• adapter_model.safetensors：LoRA 权重。

• adapter_config.json：LoRA 配置。

• README.md：模型卡和使用说明。

• adapter_model.safetensors: LoRA adapter weights.

• adapter_config.json: LoRA adapter config.

• README.md: model card and usage instructions.

Python 挂载示例 | Python Loading Example

python
Copy code
from PIL import Image
from peft import PeftModel
from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration

base_model_id = "Qwen/Qwen2.5-VL-3B-Instruct"
adapter_repo_or_path = "qualcomm-ai-hub-community/CyberSpark2-3b-cabin-lora-v2"

processor = AutoProcessor.from_pretrained(base_model_id, trust_remote_code=True)
base_model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    base_model_id,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)
model = PeftModel.from_pretrained(base_model, adapter_repo_or_path)

image = Image.open("/path/to/your/image.jpg").convert("RGB")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image"},
            {"type": "text", "text": "Describe the driver state and suggest actions."},
        ],
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[text], images=[image], return_tensors="pt")
inputs = inputs.to(base_model.device)

generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids = [
    output_ids[len(input_ids):]
    for input_ids, output_ids in zip(inputs.input_ids, generated_ids)
]
print(processor.batch_decode(generated_ids, skip_special_tokens=False)[0])