MYTH-Lab/GoT-R1-8B
MYTH-Lab/GoT-R1-8B is an 8 billion parameter causal language model developed by Wuhan University and Shanghai Jiao Tong University, featuring a 32768 token context length. It utilizes a novel Graph-of-Thought (GoT) reasoning framework with structural reinforcement, moving beyond linear Chain-of-Thought to internalize complex logic. This model excels at high-density reasoning with minimal verbosity, achieving superior accuracy and token efficiency on logical tasks compared to traditional methods. It is optimized for complex problem-solving and logical consistency, making it suitable for applications requiring precise and efficient reasoning.
Loading preview...
GoT-R1-8B: High-Density Reasoning with Graph-of-Thought
GoT-R1-8B is an 8 billion parameter causal language model developed by MYTH-Lab (Wuhan University & Shanghai Jiao Tong University) that redefines complex problem-solving. Unlike traditional Chain-of-Thought (CoT) models, GoT-R1 internalizes a Graph-of-Thought (GoT), constructing a high-density structured reasoning graph internally. This approach ensures that each reasoning step is an atomic logical primitive, leading to enhanced accuracy and significantly reduced token usage.
Key Capabilities & Differentiators
- High-Density Reasoning: Decouples pure logic from conversational filler, outputting graph topological steps instead of verbose paragraphs.
- Elimination of Redundant Narration: Avoids common "overthinking" loops by strictly penalizing verbosity during RL training.
- Automated Structural Synthesis: Trained on high-fidelity logical skeletons without requiring expensive manual graph labeling.
- Extreme Token Efficiency: Achieves state-of-the-art accuracy using only 1.8% of the token budget required by external search methods like Tree-of-Thought (ToT).
- Superior Logical Consistency: Demonstrates an 18% improvement on TruthfulQA at the 8B scale, drastically reducing logical inconsistencies and hallucinations.
Performance Highlights
GoT-R1-8B consistently outperforms Qwen3-8B across various benchmarks:
- GSM8K (ACC): 96.74% (vs. 94.62% for Qwen3-8B)
- IFEval (I-Strict): 92.31% (vs. 90.46% for Qwen3-8B)
- TruthfulQA: 84.82% (vs. 74.42% for Qwen3-8B)
- Winogrande: 84.77% (vs. 80.58% for Qwen3-8B)
Ideal Use Cases
This model is particularly well-suited for applications requiring:
- Complex Logical Problem-Solving: Tasks demanding precise, structured reasoning.
- Efficient AI Agents: Scenarios where minimal token usage and high accuracy are critical.
- Reducing LLM Verbosity: Applications where concise, fact-based outputs are preferred over lengthy explanations.