INSAIT-Institute/GRPO-Think-1.5B-16k

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 12, 2026License:cc-by-nc-sa-4.0Architecture:Transformer Open Weights Featherless Exclusive Cold

INSAIT-Institute/GRPO-Think-1.5B-16k is a 1.5 billion parameter GRPO-Think code verifier developed by INSAIT-Institute. It is fine-tuned from DeepSeek-R1-Distill-Qwen-1.5B and trained with Reinforcement Learning with Verifiable Rewards (RLVR), incorporating on-policy learning, negative samples, and intermediate thinking traces. This model is designed to judge and rank candidate solutions for competitive programming problems, utilizing a 16k reasoning-token budget during training. Its primary strength lies in providing a plug-and-play reward function for code generation policy optimization and automated evaluation.

Loading preview...

Aletheia: GRPO-Think-1.5B-16k Code Verifier

This model, GRPO-Think-1.5B-16k, is a 1.5 billion parameter code verifier developed by INSAIT-Institute as part of the Aletheia project. It is fine-tuned from deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B and trained using a method called GRPO (Reinforcement Learning with Verifiable Rewards), which includes on-policy learning, negative samples, and intermediate thinking traces. A key feature is its 16k reasoning-token budget, defining the maximum completion length used during training.

Key Capabilities

  • Code Verification: Judges and ranks candidate solutions for competitive programming problems, identifying the correct solution among several options.
  • RLVR Training: Utilizes Reinforcement Learning with Verifiable Rewards (RLVR) with specific components like on-policy learning, negative samples, and thinking traces.
  • Scalable Analysis: Part of a research effort to understand the performance-cost trade-offs in RLVR for code verifiers across different model sizes.
  • Dataset Utilization: Trained on the INSAIT-Institute/Aletheia-Train dataset, which consists of 50,000 training instances pairing competitive programming problems with candidate solutions.

Good For

  • RLHF / RLAIF: Serving as a 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 in training effective code verifiers.

Finetunes
1 models