metadata

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

tvl-mini

Description

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

tvl was trained in bf16

Data

Train dataset contains:

GrandMaster-PRO-MAX dataset (60k samples)
Translated, humanized and merged by image subset of GQA (TODO)