FRPO/qwen3-1.7b-a20_seqmean_center_16k-k1-cGroupBoth-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-a20_seqmean_center_16k-k1-cGroupBoth-clip0.2-mb4-eta100-bs256x5-n2 is a 1.7 billion parameter language model based on the Qwen3 architecture, specifically fine-tuned using Reinforcement Learning (RL) through the KL-in-LLM-RL / FRPO experimental framework. Developed by FRPO, this model leverages the verl training library and is designed for applications benefiting from RL-based optimization. It features a 32K context length and is provided in fp32 safetensors format, suitable for direct integration into RL-tuned workflows.

Loading preview...

Model Overview

This model, qwen3-1.7b-a20_seqmean_center_16k-k1-cGroupBoth-clip0.2-mb4-eta100-bs256x5-n2, is a 1.7 billion parameter language model derived from the Qwen3-1.7B base architecture. It has undergone Reinforcement Learning (RL) fine-tuning as part of the KL-in-LLM-RL / FRPO experimental series, utilizing the verl training framework.

Key Characteristics

  • Base Model: Built upon Qwen/Qwen3-1.7B.
  • Fine-tuning Method: RL fine-tuned using the KL-in-LLM-RL / FRPO experimental setup.
  • Checkpoint Availability: Includes global_step_201 at the repository root and global_step_200 in a subfolder.
  • Weights Format: Provided in fp32 safetensors, directly as saved by the trainer without additional post-processing.
  • Configuration: The specific run configuration is encoded within the repository name itself.

Intended Use Cases

This model is particularly suited for research and development in:

  • Reinforcement Learning Applications: Ideal for exploring the effects of KL-in-LLM-RL / FRPO fine-tuning on language models.
  • Experimental Setups: Useful for replicating or extending experiments conducted with the verl training library.
  • Specific Task Optimization: Potentially beneficial for tasks where RL-based fine-tuning can yield performance improvements over standard supervised fine-tuning.

Finetunes
1 models