--- language: - en pipeline_tag: text-generation ---

# Bespoke-MiniChart-7B [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1FEmlwGgn9209iQO-rs2-9UHPLoytwZMH?usp=sharing) This is an open‑source chart‑understanding Vision‑Language Model (VLM) developed at [Bespoke Labs](https://www.bespokelabs.ai/) and maintained by [Liyan Tang](https://www.tangliyan.com/) and Bespoke Labs. It sets a new state‑of‑the‑art in chart question‑answering (Chart‑QA) for 7 billion‑parameter models, outperforming much larger closed models such as Gemini‑1.5‑Pro and Claude‑3.5 on seven public benchmarks. Please check our blog for more information about how we trained the model # Model Performance Our model achieves state-of-the-art performance on chart understanding among models with similar sizes. In addition to that, our models can even surpass closed-models such as Gemini-1.5-Pro and Claude-3.5. | Model / Category | ChartQAPro (1637) | ChartQA (2500) | EvoChart (1250) | CharXiv (4000) | ChartX (1152) | ChartBench (2100) | MMC (808) | Average | |----------------------------------------|------------------:|---------------:|----------------:|---------------:|--------------:|------------------:|----------:|--------:| | **Open-Models (11 B and less)** | | | | | | | | | | InternVL-2.5-8B | –| **78.2** | 53.0 | 55.7 | 49.5 | 44.7 | **85.5** | – | | Qwen2-VL-7B | –| **82.1** | 54.5 | 53.5 | 50.8 | 50.8 | **83.9** | – | | Qwen2.5-VL-7B | **53.5** | **86.0** | 67.9 | 60.9 | 67.0 | 61.4 | **86.0** | **69.0**| | **Ours** | | | | | | | | | | Bespoke-MiniChart-7B | **56.7** | **89.5** | **71.8** | **66.4** | **68.9** | **66.1** | **88.4** | **72.5**| | **Open-Models (32 B and more)** | | | | | | | | | | QVQ-72B-Preview | –| **84.2** | 65.0 | 59.0 | 60.9 | 53.8 | **83.4** | – | | Qwen2.5-VL-32B | **58.4** | **89.5** | 74.3 | 66.9 | 64.5 | 59.8 | **89.6** | **71.9**| | Qwen2.5-VL-72B | **59.0** | **90.0** | **76.8** | 67.1 | **67.2** | 61.5 | **91.2** | **73.3**| | **Closed-Models** | | | | | | | | | | GPT-4o | **53.6** | 85.7 | 71.7 | 67.8 | 54.3 | 46.1 | **89.1** | 66.9 | | Gemini-1.5-flash | **53.8** | 85.6 | 67.5 | 67.7 | 63.5 | 58.1 | 82.1 | 68.3 | | Gemini-1.5-pro | **59.2** | **89.0** | 72.0 | **69.9** | 65.4 | 62.4 | 87.9 | **72.3**| | Claude-3.5 | **56.6** | 85.7 | **78.1** | **69.7** | 64.7 | 60.9 | **89.9** | **72.2**| | Claude-3.7 | **63.0** | 86.1 | **80.1** | **69.7** | **69.2** | **65.0** | 88.4 | **74.5**| # Model Use: ```python import requests from PIL import Image from io import BytesIO import base64 import matplotlib.pyplot as plt from vllm import LLM, SamplingParams QA_PROMPT = """Please answer the question using the chart image. Question: [QUESTION] Please first generate your reasoning process and then provide the user with the answer. Use the following format: ... your thinking process here ... ... your final answer (entity(s) or number) ... """ def get_image_from_url(image_url): try: response = requests.get(image_url, stream=True) response.raise_for_status() return Image.open(BytesIO(response.content)) except Exception as e: print(f"Error with image: {e}") return None def get_answer(image_url, question, display=True): image = get_image_from_url(image_url) if display: plt.figure(figsize=(10, 8)) plt.imshow(image) plt.axis('off') plt.show() if not image: return "Error downloading image" buffered = BytesIO() image.save(buffered, format=image.format or 'JPEG') encoded_image = base64.b64encode(buffered.getvalue()).decode('utf-8') messages = [{ "role": "user", "content": [ {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{encoded_image}"}}, {"type": "text", "text": QA_PROMPT.replace("[QUESTION]", question)} ] }] response = llm.chat([messages], sampling_params=SamplingParams(temperature=0, max_tokens=500)) return response[0].outputs[0].text # Initialize the LLM llm = LLM( model="bespokelabs/Bespoke-MiniChart-7B", tokenizer_mode="auto", max_model_len=15000, tensor_parallel_size=1, gpu_memory_utilization=0.9, mm_processor_kwargs={"max_pixels": 1600*28*28}, seed=2025, trust_remote_code=True, ) # Running inference image_url = "https://github.com/bespokelabsai/minichart-playground-examples/blob/main/images/ilyc9wk4jf8b1.png?raw=true" question = "How many global regions maintained their startup funding losses below 30% in 2022?" print("\n\n=================Model Output:===============\n\n", get_answer(image_url, question)) ``` # Licence This work is licensed under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). For commercial licensing, please contact company@bespokelabs.ai. # Citation ``` @misc{bespoke_minichart_7b, title = {Bespoke-MiniChart-7B: pushing the frontiers of open VLMs for chart understanding}, author = {Liyan Tang and Shreyas Pimpalgaonkar and Kartik Sharma and Alexandros G. Dimakis and Mahesh Sathiamoorthy and Greg Durrett}, howpublished = {blog post}, year = {2025}, url={https://huggingface.co/bespokelabs/Bespoke-MiniChart-7B}, } ``` # Acknowledgements **Bespoke Labs** team: - Liyan Tang - Shreyas Pimpalgaonkar - Kartik Sharma - Alex Dimakis - Mahesh Sathiamoorthy - Greg Durrett *Model perfected at Bespoke Labs — where careful curation meets cutting‑edge modeling.*