INSAIT-Institute/RAFT-1.5B

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/RAFT-1.5B is a 1.5 billion parameter RAFT (Rejection Sampling Fine-Tuning) code verifier model developed by INSAIT-Institute. Fine-tuned from DeepSeek-R1-Distill-Qwen-1.5B, it specializes in judging and ranking candidate solutions for competitive programming problems, learning exclusively from positive reasoning samples. This model is designed to efficiently evaluate and select correct code snippets, offering a compute-efficient strategy for code generation post-training pipelines.

Loading preview...

Model Overview

INSAIT-Institute/RAFT-1.5B is a 1.5 billion parameter code verifier model, part of the Aletheia project by INSAIT-Institute. It is specifically trained using RAFT (Rejection Sampling Fine-Tuning), which means it learns exclusively from positive reasoning samples, distinguishing it from verifiers that also incorporate negative samples or different training methodologies. The model is fine-tuned from deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B and utilizes the INSAIT-Institute/Aletheia-Train dataset.

Key Capabilities

  • Code Verification: Judges and ranks candidate solutions for competitive programming problems.
  • Positive-Only Learning: Trained solely on correct reasoning samples, simplifying the training process.
  • Execution-Grounded: Designed for scenarios where code execution feedback is available.

Intended Uses

  • RLHF / RLAIF: Can serve as a plug-and-play reward function for optimizing code generation policies.
  • Automated Evaluation: Functions as an LLM-as-a-judge for various code-related tasks.
  • Research: Useful for studying the impact of different training components (like thinking traces and negative samples) on code verifier performance.

Finetunes
1 models