Improve model card for openbmb/MiniCPM-o-2_6-gguf: Add metadata, links, and usage examples

This PR significantly enhances the model card for `openbmb/MiniCPM-o-2_6-gguf` by:

- Adding `license: apache-2.0` and `library_name: transformers` to the metadata. The `library_name` is justified by the presence of `transformers`-based code examples in the official GitHub repository, enabling automated "How to use" snippets on the Hub.
- Updating the model description with key highlights and a clear introduction of MiniCPM-o 2.6, linking it to its foundational paper [MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe](https://huggingface.co/papers/2509.18154).
- Including direct links to the main GitHub repository (`https://github.com/OpenBMB/MiniCPM-V`) and the project homepage (`https://minicpm-omni-webdemo-us.modelbest.cn/`).
- Adding a "Quickstart" section with a Python code snippet demonstrating how to use the *base* MiniCPM-o 2.6 model with the Hugging Face `transformers` library, as found in the original project's GitHub README.
- Retaining and clearly separating the existing, valuable instructions for converting to and using the GGUF format with `llama.cpp`.
- Adding a citation section for proper attribution.

These changes aim to make the model card more informative, discoverable, and user-friendly for a wider audience, catering to both `transformers` users and those utilizing the GGUF format.

README.md

@@ -1,16 +1,100 @@
 ---
-tags:
-- minicpm-o
-pipeline_tag: any-to-any
 base_model:
 - openbmb/MiniCPM-o-2_6
+pipeline_tag: any-to-any
+tags:
+- minicpm-o
+license: apache-2.0
+library_name: transformers
 ---
 
-## MiniCPM-o 2.6
+<div align="center">
+<img src="https://github.com/OpenBMB/MiniCPM-V/raw/main/assets/minicpm_v_and_minicpm_o_title.png" width="500em" ></img>
+
+**A GPT-4o Level MLLM for Single Image, Multi Image and High-FPS Video Understanding on Your Phone**
+</div>
+
+This repository provides the GGUF model weights for **MiniCPM-o 2.6**, an 8B parameter end-to-end multimodal large language model (MLLM). MiniCPM-o 2.6 is part of the MiniCPM-V family, with its foundational advancements described in the paper [MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe](https://huggingface.co/papers/2509.18154).
+
+MiniCPM-o 2.6 achieves comparable performance to GPT-4o-202405 in vision, speech, and multimodal live streaming, making it one of the most versatile and performant models in the open-source community. It supports various modalities including images, videos, text, and audio inputs, producing high-quality text and speech outputs.
+
+-   **Project Homepage:** [https://minicpm-omni-webdemo-us.modelbest.cn/](https://minicpm-omni-webdemo-us.modelbest.cn/)
+-   **Main GitHub Repository:** [https://github.com/OpenBMB/MiniCPM-V](https://github.com/OpenBMB/MiniCPM-V)
+
+<br>
+
+### ✨ Key Highlights
+-   **Leading Visual Capability:** Achieves state-of-the-art performance on various image, multi-image, and video understanding benchmarks.
+-   **State-of-the-art Speech Capability:** Supports bilingual real-time speech conversation with configurable voices, outperforming GPT-4o-realtime on audio understanding tasks.
+-   **Strong Multimodal Live Streaming Capability:** Can accept continuous video and audio streams independent of user queries, supporting real-time speech interaction.
+-   **Superior Efficiency:** Features state-of-the-art token density, improving inference speed, first-token latency, memory usage, and power consumption.
+
+<br>
+
+## 🚀 Quickstart: Using the Base Model with Hugging Face Transformers
+
+The underlying PyTorch model, `openbmb/MiniCPM-o-2_6`, is compatible with the Hugging Face `transformers` library using `trust_remote_code=True`. This allows for flexible multi-turn conversations with images, videos, and audio.
 
-This repository contains the [MiniCPM-o 2.6](https://huggingface.co/openbmb/MiniCPM-o-2_6) model weights in GGUF format, used for llama.cpp.
+```python
+import torch
+from PIL import Image
+from transformers import AutoModel, AutoTokenizer
 
-Currently, this readme only supports minicpm-omni's vision capabilities, and we will update the full-mode support as soon as possible.
+torch.manual_seed(100)
+
+# Load the base MiniCPM-o 2.6 model (PyTorch version)
+# Ensure you have 'openbmb/MiniCPM-o-2_6' in your Hugging Face cache or local path
+model_name = 'openbmb/MiniCPM-o-2_6'
+model = AutoModel.from_pretrained(
+    model_name,
+    trust_remote_code=True,
+    attn_implementation='sdpa', # or 'flash_attention_2', no 'eager'
+    torch_dtype=torch.bfloat16
+)
+model = model.eval().cuda()
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+
+# Example: Multi-turn chat with an image
+# For local execution, you might need to provide a local image path
+# For example, you can use an image from the model's assets on GitHub:
+# image = Image.open('https://github.com/OpenBMB/MiniCPM-V/raw/main/assets/minicpmo2_6/show_demo.jpg').convert('RGB')
+# Placeholder for an actual image if you run it locally:
+image = Image.new('RGB', (224, 224), color = 'red') # Replace with your actual image loading logic
+
+enable_thinking = False # If `enable_thinking=True`, the long-thinking mode is enabled.
+
+# First round chat
+question_round1 = "What is the landform in the picture?"
+msgs_round1 = [{'role': 'user', 'content': [image, question_round1]}]
+
+answer_round1 = model.chat(
+    msgs=msgs_round1,
+    tokenizer=tokenizer,
+    enable_thinking=enable_thinking
+)
+print(f"User: {question_round1}
+Assistant: {answer_round1}
+")
+
+# Second round chat, pass history context of multi-turn conversation
+msgs_round2 = msgs_round1 + [{"role": "assistant", "content": [answer_round1]}]
+question_round2 = "What should I pay attention to when traveling here?"
+msgs_round2.append({"role": "user", "content": [question_round2]})
+
+answer_round2 = model.chat(
+    msgs=msgs_round2,
+    tokenizer=tokenizer
+)
+print(f"User: {question_round2}
+Assistant: {answer_round2}")
+
+# For more detailed usage, including multi-image, video, and audio conversations,
+# please refer to the main GitHub repository: https://github.com/OpenBMB/MiniCPM-V
+```
+
+## MiniCPM-o 2.6 GGUF Version for llama.cpp
+
+This repository specifically contains the `MiniCPM-o 2.6` model weights in GGUF format, used for `llama.cpp`. This allows for efficient CPU inference on local devices.
 
 ### Prepare models and code
 
@@ -23,7 +107,7 @@ cd llama.cpp
 git checkout minicpm-omni
 ```
 
-### Usage of MiniCPM-o 2.6
+### Usage of MiniCPM-o 2.6 (GGUF)
 
 Convert PyTorch model to gguf files (You can also download the converted [gguf](https://huggingface.co/openbmb/MiniCPM-o-2_6-gguf) by us)
 
@@ -36,8 +120,8 @@ python ./convert_hf_to_gguf.py ../MiniCPM-o-2_6/model
 ./llama-quantize ../MiniCPM-o-2_6/model/ggml-model-f16.gguf ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf Q4_K_M
 ```
 
-Build llama.cpp using `CMake`:
-https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md
+Build `llama.cpp` using `CMake`:
+[https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md](https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md)
 
 ```bash
 cmake -B build
@@ -54,4 +138,17 @@ Inference on Linux or Mac
 
 # or run in interactive mode
 ./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -i
+```
+
+## 📝 Citation
+
+If you find our model/code/paper helpful, please consider citing our paper:
+
+```bib
+@article{yao2024minicpm,
+  title={MiniCPM-V: A GPT-4V Level MLLM on Your Phone},
+  author={Yao, Yuan and Yu, Tianyu and Zhang, Ao and Wang, Chongyi and Cui, Junbo and Zhu, Hongji and Cai, Tianchi and Li, Haoyu and Zhao, Weilin and He, Zhihui and others},
+  journal={arXiv preprint arXiv:2408.01800},
+  year={2024}
+}
 ```