INSAIT-Institute/GRPO-Think-14B-16k

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 12, 2026License:cc-by-nc-sa-4.0Architecture:Transformer Open Weights Featherless Exclusive Cold

INSAIT-Institute's GRPO-Think-14B-16k is a 14.8 billion parameter thinking code verifier, fine-tuned from DeepSeek-R1-Distill-Qwen-14B. It is trained using GRPO (on-policy, with negative samples and intermediate thinking traces) and a 16k reasoning-token budget. This model excels at judging and ranking candidate solutions for competitive programming problems, serving as a crucial component for post-training pipelines in large code generation models.

Loading preview...

Aletheia: GRPO-Think-14B-16k Code Verifier

This model, GRPO-Think-14B-16k, is a 14.8 billion parameter thinking code verifier developed by INSAIT-Institute. It is fine-tuned from deepseek-ai/DeepSeek-R1-Distill-Qwen-14B and trained using the GRPO (Reinforcement Learning with Verifiable Rewards) methodology. This includes on-policy learning, leveraging negative samples, and utilizing intermediate thinking traces, with a 16k reasoning-token budget during training.

Key Capabilities

  • Code Verification: Judges and ranks candidate solutions for competitive programming problems.
  • Thinking Traces: Incorporates intermediate thinking traces during its verification process, allowing for more robust analysis.
  • Scalable Training: Part of the Aletheia project, which explores optimal RLVR training recipes across different model scales.

Intended Uses

  • RLHF / RLAIF: Functions as a plug-and-play reward function for optimizing code generation policies.
  • Automated Evaluation: Can be used as an LLM-as-a-judge for various code-related tasks.
  • Research: Provides a testbed for studying the effects of different training components (thinking traces, on-policy learning, negative samples) in code verifier development.

Finetunes
1 models