FRPO/qwen3-1.7b-a15_global_token_norm-k1-cNone-globalTokNorm-clip0.2-mb4-eta100-bs256x5-n2-seed1

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-a15_global_token_norm-k1-cNone-globalTokNorm-clip0.2-mb4-eta100-bs256x5-n2-seed1 is a 1.7 billion parameter language model based on the Qwen3 architecture, developed through FRPO's KL-in-LLM-RL experiments. This model is an RL fine-tuned checkpoint, specifically optimized using reinforcement learning techniques. It is designed for applications benefiting from models fine-tuned with specific reinforcement learning configurations, offering a 32768 token context length.

Loading preview...

Model Overview

This model, FRPO/qwen3-1.7b-a15_global_token_norm-k1-cNone-globalTokNorm-clip0.2-mb4-eta100-bs256x5-n2-seed1, is a 1.7 billion parameter language model derived from the Qwen3-1.7B base model. It represents an RL fine-tuned checkpoint from the KL-in-LLM-RL / FRPO experimental series, developed using the verl framework.

Key Characteristics

  • Base Architecture: Built upon the robust Qwen3-1.7B model.
  • Fine-tuning Method: Utilizes Reinforcement Learning (RL) as part of the KL-in-LLM-RL / FRPO experiments.
  • Checkpoint: The specific checkpoint provided is global_step_200.
  • Weights: Stored in fp32 safetensors format, directly as saved by the trainer without additional post-processing.
  • Configuration: The run configuration details are embedded within the model's repository name, indicating specific training parameters like global_token_norm, k1, clip0.2, mb4, eta100, bs256x5, n2, and seed1.

Potential Use Cases

This model is particularly suited for research and development in:

  • Exploring the impact of specific RL fine-tuning strategies on Qwen3-based models.
  • Applications requiring a model with a 32768 token context length that has undergone reinforcement learning optimization.
  • Evaluating the performance of models trained with the specified global_token_norm and other configuration parameters.

Finetunes
1 models