reasonrag/Qwen2.5-7B-Instruct-ReasonRAG

dpo_v16

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the dpo_mcts_rag_v8 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8564
• Rewards/chosen: 1.0146
• Rewards/rejected: -0.5767
• Rewards/accuracies: 0.6204
• Rewards/margins: 1.5913
• Logps/chosen: -65.8051
• Logps/rejected: -74.8917
• Logits/chosen: -0.5108
• Logits/rejected: -0.5206

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: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 3
• gradient_accumulation_steps: 4
• total_train_batch_size: 12
• total_eval_batch_size: 3
• optimizer: Use 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: 0.2
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/chosen	Logps/rejected	Logits/chosen	Logits/rejected
0.9257	0.4352	500	0.9691	1.0721	-0.2211	0.6171	1.2933	-65.6133	-73.7064	-0.5575	-0.5659
0.7847	0.8703	1000	0.8689	0.7341	-0.9130	0.6223	1.6471	-66.7400	-76.0126	-0.5218	-0.5319

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0