nthakur/Meta-Llama-3-8B-Instruct-mirage-bench-sft-teacher-llama-3

Meta-Llama-3-8B-Instruct-mirage-meta-llama-3-sft-instruct

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the nthakur/mirage-meta-llama-3-sft-instruct dataset. It achieves the following results on the evaluation set:

• Loss: 0.2431

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: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
0.3403	0.0597	200	0.3074
0.3224	0.1195	400	0.2954
0.3055	0.1792	600	0.2886
0.2899	0.2389	800	0.2804
0.3116	0.2987	1000	0.2772
0.3101	0.3584	1200	0.2728
0.2913	0.4182	1400	0.2679
0.2765	0.4779	1600	0.2625
0.2697	0.5376	1800	0.2601
0.2759	0.5974	2000	0.2557
0.264	0.6571	2200	0.2524
0.2705	0.7168	2400	0.2490
0.2694	0.7766	2600	0.2466
0.2639	0.8363	2800	0.2450
0.2598	0.8961	3000	0.2435
0.2483	0.9558	3200	0.2432

Framework versions

• PEFT 0.10.0
• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1