metadata

library_name: transformers
datasets:
  - Joctor/cn_bokete_oogiri_caption
base_model:
  - Qwen/Qwen2-VL-7B-Instruct
pipeline_tag: image-to-text

Model Card for Model ID

AI大喜利，简介 https://www.gcores.com/articles/188405

How to Get Started with the Model

Use the code below to get started with the model.

from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info

model_id = "Joctor/qwen2-vl-7b-instruct-ogiri"

# default: Load the model on the available device(s)
model = Qwen2VLForConditionalGeneration.from_pretrained(
    model_id, torch_dtype="auto", device_map="auto"
)

# default processer
processor = AutoProcessor.from_pretrained(model_id)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "path/to/image",
            },
            {"type": "text", "text": "根据图片给出有趣巧妙的回答"},
        ],
    }
]

# Preparation for inference
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)
inputs = inputs.to("cuda")

# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)

Training Details

Training Data

https://huggingface.co/datasets/Joctor/cn_bokete_oogiri_caption

Training Procedure

基础模型：qwen2vl
微调方式：数据量充足，采用SFT微调
微调参数：max_length=1024(短就是好!), num_train_epochs=1, per_device_train_batch_size=1, gradient_accumulation_steps=1
训练设备：10 * 4090D
训练时长：22小时