graf/Llama-3.1-GSM8K-8B-RM

reward

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on the gsm8k_llama3.1-8B_128_1ep dataset. It achieves the following results on the evaluation set:

• Loss: 0.2467
• val Accuracy: 0.8810

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: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 16
• total_train_batch_size: 128
• total_eval_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	val Accuracy
0.609	0.0856	5	0.4890	0.8135
0.3044	0.1711	10	0.2622	0.9204
0.3091	0.2567	15	0.1574	0.9060
0.2377	0.3422	20	0.2161	0.9090
0.2227	0.4278	25	0.2810	0.8696
0.3034	0.5134	30	0.2796	0.8832
0.2101	0.5989	35	0.2074	0.9022
0.2027	0.6845	40	0.1866	0.9075
0.2683	0.7701	45	0.2167	0.8976
0.1873	0.8556	50	0.2340	0.8878
0.2984	0.9412	55	0.2451	0.8825

Framework versions

• Transformers 4.46.1
• Pytorch 2.5.1
• Datasets 3.1.0
• Tokenizers 0.20.1