FRPO/qwen3-1.7b-a13_entropy_adv-entAdv-alpha0.4-kappa2-clip0.2-mb4-eta100-bs256x5-n2

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 15, 2026Architecture:Transformer Featherless Exclusive Cold

FRPO/qwen3-1.7b-a13_entropy_adv-entAdv-alpha0.4-kappa2-clip0.2-mb4-eta100-bs256x5-n2 is a 1.7 billion parameter language model based on the Qwen3 architecture, fine-tuned using Reinforcement Learning (RL) with the FRPO method. Developed as part of the KL-in-LLM-RL experiments, this checkpoint is specifically designed for research and development in RL-based language model optimization. It offers a 32768 token context length and is suitable for evaluating the effects of entropy-regularized adversarial training in RL fine-tuning.

Loading preview...

Overview

This model, FRPO/qwen3-1.7b-a13_entropy_adv-entAdv-alpha0.4-kappa2-clip0.2-mb4-eta100-bs256x5-n2, is a 1.7 billion parameter language model derived from the Qwen/Qwen3-1.7B base model. It has undergone Reinforcement Learning (RL) fine-tuning using the FRPO (Fictitious Reinforcement Policy Optimization) method, specifically within the context of KL-in-LLM-RL experiments. The model's name encodes its run configuration, indicating parameters like entropy_adv, entAdv, alpha0.4, kappa2, clip0.2, mb4, eta100, bs256x5, and n2.

Key Characteristics

  • Base Model: Qwen3-1.7B architecture.
  • Parameter Count: 1.7 billion parameters.
  • Context Length: Supports a substantial context window of 32768 tokens.
  • Fine-tuning Method: Utilizes Reinforcement Learning (RL) with the FRPO algorithm, part of the KL-in-LLM-RL research.
  • Weights: Provided in fp32 safetensors format, directly as saved by the trainer without post-processing.
  • Experimental Focus: Designed for exploring advanced RL fine-tuning techniques, particularly those involving entropy-regularized adversarial training.

Good For

  • RL Research: Ideal for researchers and developers investigating Reinforcement Learning applications in large language models.
  • Experimental Evaluation: Suitable for evaluating the impact of specific RL parameters and configurations (e.g., entropy_adv, alpha, kappa, clip) on model performance and behavior.
  • Comparative Studies: Can be used as a checkpoint in studies comparing different RL fine-tuning strategies for Qwen3-1.7B.

Finetunes
1 models