TiramisuQiao/LLama3-8B-Ins-SFT-WebRL-LoRA

This model is a fine-tuned version of Meta-Llama-3-8B-Instruct on the web_policy_sft dataset. It achieves the following results on the evaluation set:

• Loss: 0.4631

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
1.0437	0.1878	50	1.0606
0.5727	0.3756	100	0.5909
0.5279	0.5634	150	0.4970
0.4961	0.7512	200	0.4700
0.4797	0.9390	250	0.4632

Framework versions

• PEFT 0.15.2
• Transformers 4.53.1
• Pytorch 2.3.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.1