Add model card and metadata for MemoBrain-14B

README.md

@@ -0,0 +1,77 @@
+---
+license: mit
+library_name: transformers
+pipeline_tag: text-generation
+---
+
+# MemoBrain-14B: Executive Memory as an Agentic Brain for Reasoning
+
+MemoBrain-14B is an executive memory model designed for tool-augmented agents. It constructs a dependency-aware memory over reasoning steps, capturing salient intermediate states and their logical relations to sustain coherent, goal-directed reasoning over long horizons.
+
+- **Paper:** [MemoBrain: Executive Memory as an Agentic Brain for Reasoning](https://huggingface.co/papers/2601.08079)
+- **Repository:** [https://github.com/qhjqhj00/MemoBrain](https://github.com/qhjqhj00/MemoBrain)
+
+## Overview
+Complex reasoning in tool-augmented agent frameworks is often long-horizon, causing reasoning traces to strain the working context of LLMs. MemoBrain operates as a co-pilot alongside the reasoning agent, managing the working context by:
+- **Pruning** invalid steps.
+- **Folding** completed sub-trajectories into compact summaries.
+- **Preserving** a high-salience reasoning backbone under a fixed context budget.
+
+MemoBrain-14B is based on the Qwen3 architecture and has been specifically fine-tuned for these memory operations.
+
+## Quick Start
+
+### Deployment with vLLM
+We recommend deploying the model using [vLLM](https://github.com/vllm-project/vllm) for high-performance inference:
+
+```bash
+pip install vllm
+vllm serve TommyChien/MemoBrain-14B --port 8002
+```
+
+### Python Usage
+Once the model is served, you can interact with it using the `memobrain` package from the [official repository](https://github.com/qhjqhj00/MemoBrain).
+
+```python
+import asyncio
+from memobrain import MemoBrain
+
+async def main():
+    # Step 1: Initialize MemoBrain
+    memory = MemoBrain(
+        api_key="EMPTY",  # vLLM doesn't require API key
+        base_url="http://localhost:8002/v1",
+        model_name="TommyChien/MemoBrain-14B"
+    )
+    
+    # Step 2: Initialize memory with your task
+    memory.init_memory("Solve a complex research problem")
+    
+    # Step 3: Memorize conversation interactions
+    # The recommended unit is an episode: thinking → tool call → tool response
+    await memory.memorize([
+        {"role": "assistant", "content": "I need to search for information about Paris..."},
+        {"role": "user", "content": "Search results: Paris is the capital of France..."}
+    ])
+    
+    # Step 4: Optimize memory (flush invalid steps & fold completed sub-trajectories)
+    optimized_messages = await memory.recall()
+    print(f"Memory optimized: {len(optimized_messages)} messages")
+
+asyncio.run(main())
+```
+
+## Experimental Results
+MemoBrain-8B and 14B have demonstrated state-of-the-art performance on long-horizon reasoning benchmarks such as GAIA and WebWalker, showing significant improvements particularly on complex, multi-step tasks.
+
+## Citation
+If you find MemoBrain useful for your research, please cite:
+
+```bibtex
+@article{memobrain2026,
+  title={MemoBrain: Executive Memory as an Agentic Brain for Reasoning},
+  author={Hongjin Qian, Zhao Cao, Zheng Liu},
+  journal={arXiv preprint arXiv:2601.08079},
+  year={2026}
+}
+```