Update README.md

README.md

@@ -1,25 +1,63 @@
----
-title: Synapse-Base Inference API
-emoji: 🧠
-colorFrom: blue
-colorTo: purple
-sdk: docker
-pinned: true
-license: cc-by-nc-4.0
----
+---
+title: Synapse-Base Inference API
+emoji: 🧠
+colorFrom: blue
+colorTo: purple
+sdk: docker
+pinned: false
+license: cc-by-nc-4.0
+---
 
 # 🧠 Synapse-Base Inference API
 
-High-performance chess move prediction API powered by Synapse-Base v3.0.
+State-of-the-art chess move prediction powered by Synapse-Base v3.0.
+
+[![Model](https://img.shields.io/badge/Model-Synapse--Base-blue)](https://huggingface.co/GambitFlow/Synapse-Base)
+[![Parameters](https://img.shields.io/badge/Params-38.1M-orange)](https://huggingface.co/GambitFlow/Synapse-Base)
+[![Architecture](https://img.shields.io/badge/Arch-CNN--Transformer-red)](https://huggingface.co/GambitFlow/Synapse-Base)
+
+## 🎯 Model Details
+
+**Synapse-Base** is a hybrid CNN-Transformer chess engine implementing research-grade search techniques:
+
+- **Model:** [GambitFlow/Synapse-Base](https://huggingface.co/GambitFlow/Synapse-Base)
+- **Parameters:** 38.1 Million
+- **Architecture:** Hybrid CNN (ResNet) + Transformer (4 layers)
+- **Input:** 119-channel board representation
+- **Training Data:** [GambitFlow/Elite-Data](https://huggingface.co/datasets/GambitFlow/Elite-Data) (5M+ positions, 2000+ ELO)
+- **Strength:** 2400-2600 ELO estimated
+
+## 🔬 Search Algorithm
+
+Implements state-of-the-art techniques from chess engine research:
 
-## 🎯 Features
+### Core Algorithms
+- **Principal Variation Search (PVS)** [^1] - Enhanced alpha-beta with null-window searches
+- **Null Move Pruning** [^2] - Skip opponent moves for faster deep search
+- **Late Move Reductions (LMR)** [^3] - Reduce depth for unlikely moves
+- **Aspiration Windows** [^4] - Focused search around expected score
 
-- **Deep Search Algorithm**: Advanced alpha-beta pruning with move ordering
-- **CPU Optimized**: Runs efficiently on 2 vCPU + 16GB RAM
-- **REST API**: Simple POST endpoint for move generation
-- **Model**: 38.1M parameter hybrid CNN-Transformer
+### Optimization Techniques
+- **Transposition Table with Zobrist Hashing** [^5] - 256MB cache with 64-bit keys
+- **Advanced Move Ordering** [^6]:
+  - TT Move (from previous searches)
+  - MVV-LVA (Most Valuable Victim - Least Valuable Attacker)
+  - Killer Moves (quiet moves causing cutoffs)
+  - History Heuristic (move success tracking)
+  - Counter Moves (refutation table)
+- **Quiescence Search** [^7] - Resolve tactical sequences
+- **Iterative Deepening** [^8] - Progressive depth increase
 
-## 📡 API Endpoint
+## 📊 Performance
+
+| Metric | Value | Test Set |
+|--------|-------|----------|
+| **Depth 5 Search** | ~2 seconds | Standard positions |
+| **NPS** | 50K-100K nodes/sec | T4 GPU equivalent |
+| **Memory Usage** | ~5GB RAM | Peak inference |
+| **TT Hit Rate** | 60-80% | Typical positions |
+
+## 📡 API Endpoints
 
 ### `POST /get-move`
 
@@ -27,7 +65,7 @@ High-performance chess move prediction API powered by Synapse-Base v3.0.
 ```json
 {
   "fen": "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1",
-  "depth": 3,
+  "depth": 5,
   "time_limit": 5000
 }
 ```
@@ -37,35 +75,97 @@ High-performance chess move prediction API powered by Synapse-Base v3.0.
 {
   "best_move": "e2e4",
   "evaluation": 0.25,
-  "depth_searched": 3,
-  "nodes_evaluated": 15234,
-  "time_taken": 1247
+  "depth_searched": 5,
+  "seldepth": 12,
+  "nodes_evaluated": 87543,
+  "time_taken": 1847,
+  "nps": 47392,
+  "pv": ["e2e4", "e7e5", "g1f3"],
+  "tt_hit_rate": 68.4
 }
 ```
 
+### `GET /health`
+
+Health check endpoint.
+
 ## 🔧 Parameters
 
-- **fen** (required): Current board position in FEN notation
-- **depth** (optional): Search depth (1-5, default: 3)
-- **time_limit** (optional): Max time in milliseconds (default: 5000)
+- **fen** (required): Board position in FEN notation
+- **depth** (optional): Search depth (1-10, default: 5)
+- **time_limit** (optional): Max time in milliseconds (1000-30000, default: 5000)
+
+## 🚀 Quick Start
 
-## 💻 Local Testing
+```python
+import requests
 
-```bash
-docker build -t synapse-inference .
-docker run -p 7860:7860 synapse-inference
+response = requests.post(
+    "https://YOUR-SPACE.hf.space/get-move",
+    json={
+        "fen": "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1",
+        "depth": 5
+    }
+)
+
+data = response.json()
+print(f"Best move: {data['best_move']}")
+print(f"Evaluation: {data['evaluation']}")
 ```
 
-## 📊 Performance
+## 📚 Research References
 
-- **Average Response Time**: 1-3 seconds per move
-- **Memory Usage**: ~4GB RAM
-- **Concurrent Requests**: Up to 4 simultaneous
+[^1]: **Principal Variation Search**: Marsland, T. A. (1986). "A Review of Game-Tree Pruning". *ICCA Journal*, 9(1), 3-19.
 
-## ⚠️ Rate Limits
+[^2]: **Null Move Pruning**: Goetsch, G., & Campbell, M. S. (1990). "Experiments with the null-move heuristic". *Computers, Chess, and Cognition*, 159-168.
 
-Free tier: 100 requests/hour per IP
+[^3]: **Late Move Reductions**: Heinz, E. A. (2000). "Adaptive null-move pruning". *ICGA Journal*, 23(3), 123-134.
+
+[^4]: **Aspiration Windows**: Campbell, M., Hoane Jr, A. J., & Hsu, F. H. (2002). "Deep Blue". *Artificial Intelligence*, 134(1-2), 57-83.
+
+[^5]: **Zobrist Hashing**: Zobrist, A. L. (1970). "A new hashing method with application for game playing". *ICCA Journal*, 13(2), 69-73.
+
+[^6]: **Move Ordering**: Stockfish Team (2023). "Stockfish Chess Engine". https://github.com/official-stockfish/Stockfish
+
+[^7]: **Quiescence Search**: Shannon, C. E. (1950). "Programming a computer for playing chess". *Philosophical Magazine*, 41(314), 256-275.
+
+[^8]: **Iterative Deepening**: Korf, R. E. (1985). "Depth-first iterative-deepening: An optimal admissible tree search". *Artificial Intelligence*, 27(1), 97-109.
+
+## 📖 Additional Resources
+
+- **AlphaZero**: Silver, D., et al. (2018). "A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play". *Science*, 362(6419), 1140-1144.
+- **Leela Chess Zero**: https://lczero.org/
+- **Stockfish NNUE**: Nasu, Y. (2020). "Efficiently Updatable Neural Network for Computer Shogi". *The 28th Game Programming Workshop*.
+
+## ⚠️ Limitations
+
+- Not optimized for endgames with 7+ pieces (no tablebase integration yet)
+- Time management assumes stable network latency
+- Commercial use restricted under CC BY-NC 4.0
+
+## 🏆 Model Lineage
+
+**GambitFlow AI Engine Series:**
+1. Nexus-Nano (2.8M) - Ultra-fast baseline
+2. Nexus-Core (13M) - Balanced performance
+3. **Synapse-Base (38.1M)** - State-of-the-art ✨
+
+---
+
+**Developed by:** [GambitFlow](https://huggingface.co/GambitFlow) / Rafsan1711  
+**License:** CC BY-NC 4.0  
+**Citation:** If you use this model in research, please cite the GambitFlow repository.
+
+```bibtex
+@software{gambitflow_synapse_base_2025,
+  author = {Rafsan1711},
+  title = {Synapse-Base: Hybrid CNN-Transformer Chess Engine},
+  year = {2025},
+  publisher = {Hugging Face},
+  url = {https://huggingface.co/GambitFlow/Synapse-Base}
+}
+```
 
 ---
 
-Built with ❤️ by GambitFlow
+Part of the **GambitFlow Project** 🚀♟️