Aletheia-Bench/GRPO-Think-14B-8k

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Dec 17, 2025License:cc-by-nc-sa-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Aletheia-Bench/GRPO-Think-14B-8k is a 14.8 billion parameter GRPO-Think code verifier model developed by Aletheia-Bench. Fine-tuned from DeepSeek-R1-Distill-Qwen-14B, it is trained with Reinforcement Learning with Verifiable Rewards (RLVR) using an 8k reasoning-token budget. This model excels at judging and ranking candidate solutions for competitive programming problems, making it suitable for automated code evaluation and as a reward function in RLHF/RLAIF for code generation.

Loading preview...

Aletheia-Bench/GRPO-Think-14B-8k: A Code Verifier Model

GRPO-Think-14B-8k is a 14.8 billion parameter code verifier developed by Aletheia-Bench, fine-tuned from deepseek-ai/DeepSeek-R1-Distill-Qwen-14B. It is part of the Aletheia project, which investigates the effectiveness of Reinforcement Learning with Verifiable Rewards (RLVR) for code verifiers. This specific model is a GRPO-Think verifier, meaning it was trained using GRPO (on-policy, with negative samples and intermediate thinking traces) and utilized an 8k reasoning-token budget during training.

Key Capabilities

  • Code Solution Ranking: Given a competitive programming problem and multiple candidate solutions, the model can judge and rank these candidates to identify the correct one.
  • Thinking Traces: Incorporates intermediate thinking traces during its verification process, allowing for more robust reasoning.
  • RLVR Training: Benefits from a comprehensive RLVR pipeline, including on-policy learning and leveraging negative samples, which are crucial for stabilizing training at larger scales.
  • Execution-Grounded: Trained on the Aletheia-Train dataset, which consists of 50,000 competitive programming instances with execution-verified solutions across Python, C++, and Java.

Good For

  • RLHF / RLAIF: Can serve as a plug-and-play reward function for optimizing code generation policies.
  • Automated Evaluation: Ideal for LLM-as-a-judge applications in various code-related tasks.
  • Research: Provides a robust platform for studying the impact of thinking traces, on-policy learning, and negative samples in the development of effective code verifiers.

Finetunes
1 models