outputs_test_new

This model is a fine-tuned version of unsloth/Llama-3.2-3B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8289
• Rewards/chosen: -0.0732
• Rewards/rejected: -0.1379
• Rewards/accuracies: 0.8333
• Rewards/margins: 0.0648
• Logps/rejected: -1.3794
• Logps/chosen: -0.7316
• Logits/rejected: -0.7685
• Logits/chosen: -0.1043
• Nll Loss: 0.7863
• Log Odds Ratio: -0.3910
• Log Odds Chosen: 0.9677

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: 8e-06
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Nll Loss	Log Odds Ratio	Log Odds Chosen
0.7899	0.5747	50	0.8337	-0.0734	-0.1339	0.8333	0.0605	-1.3386	-0.7339	-0.7673	-0.1094	0.7900	-0.4029	0.9116
0.4669	1.1494	100	0.8308	-0.0734	-0.1373	0.8333	0.0640	-1.3735	-0.7335	-0.7682	-0.1034	0.7880	-0.3928	0.9564
0.6859	1.7241	150	0.8289	-0.0732	-0.1379	0.8333	0.0648	-1.3794	-0.7316	-0.7685	-0.1043	0.7863	-0.3910	0.9677

Framework versions

• PEFT 0.13.1
• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1