INSAIT-Institute/GRPO-Think-14B-8k
INSAIT-Institute/GRPO-Think-14B-8k is a 14.8 billion parameter GRPO-Think code verifier developed by INSAIT-Institute. This model is fine-tuned from DeepSeek-R1-Distill-Qwen-14B and trained with Reinforcement Learning with Verifiable Rewards (RLVR), including negative samples and intermediate thinking traces, using an 8k reasoning-token budget. It specializes in judging and ranking candidate solutions for competitive programming problems, serving as a plug-and-play reward function for code generation policy optimization and automated evaluation.
Loading preview...
Aletheia: GRPO-Think-14B-8k Code Verifier
GRPO-Think-14B-8k is a 14.8 billion parameter code verifier developed by INSAIT-Institute, fine-tuned from deepseek-ai/DeepSeek-R1-Distill-Qwen-14B. It is trained using the GRPO (Generative Reinforcement Preference Optimization) method, which incorporates on-policy learning, negative samples, and intermediate thinking traces, with an 8k reasoning-token budget during training. This model is part of the Aletheia project, which investigates the optimal training recipes for code verifiers across different scales and application scenarios.
Key Capabilities
- Code Verification: Judges and ranks candidate solutions for competitive programming problems, given a problem statement and multiple solutions.
- RLVR Training: Utilizes Reinforcement Learning with Verifiable Rewards (RLVR) to learn robust verification capabilities.
- Thinking Traces: Incorporates intermediate thinking traces during training, allowing for more nuanced reasoning.
- Negative Samples: Learns from negative samples to improve its ability to distinguish correct from incorrect solutions.
Good For
- RLHF / RLAIF: Serving as a plug-and-play reward function for optimizing code generation policies.
- Automated Evaluation: Acting as an LLM-as-a-judge for various code-related tasks.
- Research: Studying the impact of thinking traces, on-policy learning, and negative samples on the performance of code verifiers.