yakazimir/simpo-exps_qwen05b

trl-lib/qwen1.5-0.5b-sft

This model is a fine-tuned version of trl-lib/qwen1.5-0.5b-sft on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7797
• Rewards/chosen: -1.7866
• Rewards/rejected: -1.8726
• Rewards/accuracies: 0.5230
• Rewards/margins: 0.0860
• Logps/rejected: -1.8726
• Logps/chosen: -1.7866
• Logits/rejected: -0.3259
• Logits/chosen: -0.3565
• Semantic Entropy: 0.6461

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-08
• train_batch_size: 1
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 16
• 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
• training_steps: 60

Training results

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1