Model Overview

This model, laion/r2egym-unified-1000__Qwen3-8B, is an 8 billion parameter language model built upon the robust Qwen/Qwen3-8B architecture. It has been fine-tuned using the /e/data1/datasets/playground/ot/hf_hub/datasets--laion--r2egym-unified-1000/snapshots/5f3bc7d941f44406d18e2d31cdb42df47890e5f5_thinking_preprocessed dataset.

Training Details

The fine-tuning process involved specific hyperparameters aimed at optimizing performance:

• Learning Rate: 4e-05
• Batch Sizes: A train_batch_size of 1 and eval_batch_size of 8, with a total_train_batch_size of 96 and total_eval_batch_size of 256, achieved through a gradient_accumulation_steps of 3 across 32 GPUs.
• Optimizer: Utilized ADAMW_TORCH_FUSED with betas=(0.9, 0.98) and epsilon=1e-08.
• Scheduler: A cosine learning rate scheduler with a warmup ratio of 0.1 was employed over 7 epochs.

Framework Versions

The model was trained using:

• Transformers 4.57.6
• Pytorch 2.9.1+cu130
• Datasets 4.7.0
• Tokenizers 0.22.2

Further information regarding the model's specific description, intended uses, limitations, and detailed training/evaluation data is currently pending.