laion/a3-rl-laion_nemotron-gym-identity-following-v2-65-8B

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 3, 2026Architecture:Transformer Featherless Exclusive Cold

The laion/a3-rl-laion_nemotron-gym-identity-following-v2-65-8B model is an 8 billion parameter language model fine-tuned using Reinforcement Learning (RL) with GRPO/rloo_n methods. Based on laion/GLM-4_7-swesmith-sandboxes-with_tests-oracle_verified_120s-maxeps-131k-fixthink, it was trained on the open-athena/nemotron-gym-identity-following-v2 dataset. This model is specifically optimized for identity-following tasks within a gym environment, demonstrating near-perfect reward achievement during its 80-step training run.

Loading preview...

Model Overview

laion/a3-rl-laion_nemotron-gym-identity-following-v2-65-8B is an 8 billion parameter language model developed by LAION, specifically fine-tuned using Reinforcement Learning (RL) techniques. It leverages the GRPO/rloo_n algorithms within the SkyRL framework on the Jupiter (a3 series #22) platform.

Key Characteristics

  • Base Model: Built upon laion/GLM-4_7-swesmith-sandboxes-with_tests-oracle_verified_120s-maxeps-131k-fixthink.
  • Training Data: Fine-tuned using the open-athena/nemotron-gym-identity-following-v2 dataset, indicating a focus on identity-following tasks.
  • RL Optimization: Utilizes advanced RL methods (GRPO/rloo_n) for performance enhancement.
  • Performance: The model achieved near-perfect reward during its 80-step training run, with the selected checkpoint (global_step 65) showing a high 5-period EMA of reward/avg_raw_reward (0.9996).

Training Details

The training process involved 80 steps, with checkpoint selection based on the highest 5-period Exponential Moving Average (EMA) of the average raw reward. Detailed training traces, including the last episode of each trial, are available as a companion dataset: open-athena/a3-rl-laion_nemotron-gym-identity-following-v2.

Intended Use

This model is particularly suited for research and development in RL environments, especially for tasks requiring precise identity-following or similar policy adherence, given its specialized training on the nemotron-gym-identity-following-v2 dataset.

Finetunes
1 models