smollm2-17b-dpo-cai-v1

This model is a fine-tuned version of moodlep/smollm2-1.7b-instr-sft-cai on the HuggingFaceH4/ultrafeedback_binarized and the HuggingFaceH4/cai-conversation-harmless datasets. It achieves the following results on the evaluation set:

• Loss: 0.6931
• Rewards/chosen: 0.0000
• Rewards/rejected: -0.0001
• Rewards/accuracies: 0.4339
• Rewards/margins: 0.0002
• Logps/rejected: -228.5849
• Logps/chosen: -245.0489
• Logits/rejected: -1.4322
• Logits/chosen: -1.3905

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-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• 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	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6933	0.3883	100	0.6930	-0.0013	-0.0015	0.4358	0.0002	-228.7255	-245.1841	-1.6940	-1.6433
0.6931	0.7767	200	0.6929	-0.0007	-0.0011	0.4487	0.0004	-228.6854	-245.1230	-1.2768	-1.2404

Framework versions

• PEFT 0.14.0
• Transformers 4.45.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.20.3