README.md

---
license: apache-2.0
datasets:
- Vikhrmodels/GrandMaster-PRO-MAX
language:
- ru
- en
base_model:
- Qwen/Qwen2-VL-2B-Instruct
pipeline_tag: text2text-generation
tags:
- multimodal
library_name: transformers
---

# tvl-mini


## Description

This is LORA finetune of Qwen2-VL-2B on russian language.


## Data

Dataset contains:
 - GrandMaster-PRO-MAX dataset (68k samples)
 - Visual Reasoning (36k samples) #Training in progress
 - Captioning (34k samples) #Training in progress
 - Knowledgeable VQA (35k samples) #Training in progress
 - VQA (80k samples) #Training in progress
 - Classification (21k samples) #Training in progress
 - Conversations (11k samples) #Training in progress

## Bechmarks

### TODO

## Quickstart

Your can simply run [this notebook](https://www.kaggle.com/code/artemdzhalilov/tvl-hand-test) or run code below.

First install qwen-vl-utils and dev version of transformers:

```bash
pip install qwen-vl-utils
pip install --no-cache-dir git+https://github.com/huggingface/transformers@19e6e80e10118f855137b90740936c0b11ac397f
```

And then run:

```bash
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch

model = Qwen2VLForConditionalGeneration.from_pretrained(
    "2Vasabi/tvl-mini-0.1", torch_dtype=torch.bfloat16, device_map="auto"
)


processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct")
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://i.ibb.co/d0QL8s6/images.jpg",
            },
            {"type": "text", "text": "Кратко опиши что ты видишь на изображении"},
        ],
    }
]

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")

generated_ids = model.generate(**inputs, max_new_tokens=1000)
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)
```