v1_5_mistral_lora32

This model is a fine-tuned version of peiyi9979/math-shepherd-mistral-7b-prm on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3153
• Accuracy: 0.8614
• Precision: 0.7273
• Recall: 0.7547
• F1: 0.7407

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-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 765837
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• 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	Accuracy	Precision	Recall	F1
No log	0	0	0.5958	0.7376	0.5	0.0660	0.1167
0.6819	0.0575	20	0.5694	0.7450	0.6154	0.0755	0.1345
0.4789	0.1149	40	0.4776	0.7550	0.5263	0.6604	0.5858
0.4108	0.1724	60	0.4043	0.8045	0.6286	0.6226	0.6256
0.3486	0.2299	80	0.3752	0.8366	0.6961	0.6698	0.6827
0.2936	0.2874	100	0.3650	0.8465	0.7115	0.6981	0.7048
0.3788	0.3448	120	0.3579	0.8465	0.6864	0.7642	0.7232
0.3918	0.4023	140	0.3483	0.8564	0.7449	0.6887	0.7157
0.3386	0.4598	160	0.3409	0.8614	0.7193	0.7736	0.7455
0.2864	0.5172	180	0.3257	0.8663	0.75	0.7358	0.7429
0.2581	0.5747	200	0.3223	0.8663	0.7407	0.7547	0.7477
0.3373	0.6322	220	0.3174	0.8663	0.75	0.7358	0.7429
0.3006	0.6897	240	0.3172	0.8564	0.7222	0.7358	0.7290
0.3157	0.7471	260	0.3143	0.8639	0.7339	0.7547	0.7442
0.291	0.8046	280	0.3137	0.8639	0.7339	0.7547	0.7442
0.2578	0.8621	300	0.3168	0.8639	0.7339	0.7547	0.7442
0.3223	0.9195	320	0.3157	0.8614	0.7273	0.7547	0.7407
0.3448	0.9770	340	0.3153	0.8614	0.7273	0.7547	0.7407

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3