d1

This model is a fine-tuned version of deepseek-ai/deepseek-coder-1.3b-base on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 0.1395
• Rewards/chosen: -0.1261
• Rewards/rejected: -19.6585
• Rewards/accuracies: 0.9737
• Rewards/margins: 19.5324
• Logps/rejected: -369.5529
• Logps/chosen: -169.7162
• Logits/rejected: -9.2987
• Logits/chosen: -8.0855

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: 5e-05
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 64
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0841	1.7149	100	0.0542	0.8716	-12.2566	0.9737	13.1282	-295.5336	-159.7391	-16.8791	-17.5293
0.0013	3.4298	200	0.1395	-0.1261	-19.6585	0.9737	19.5324	-369.5529	-169.7162	-9.2987	-8.0855

Framework versions

• Transformers 4.43.0.dev0
• Pytorch 2.2.2+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1