hp_ablations_grid_mistral_bsz4096_lr2e-6_scheduler-cosine-warmup0.05-minlr5e-7

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the mlfoundations-dev/oh-dcft-v3-llama3.1-nemotron-70b_shareGPT_format dataset. It achieves the following results on the evaluation set:

• Loss: 0.0622

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: 2e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 64
• gradient_accumulation_steps: 8
• total_train_batch_size: 4096
• total_eval_batch_size: 512
• 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_with_min_lr
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.5488	1.0	84	0.0679
0.4983	2.0	168	0.0636
0.4707	3.0	252	0.0622

Framework versions

• Transformers 4.46.1
• Pytorch 2.3.0
• Datasets 3.1.0
• Tokenizers 0.20.3