qwen2.5-0.5b-expo-DPO-25-2

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-25-1 on the hZzy/train_pairwise_all_new dataset. It achieves the following results on the evaluation set:

• Loss: 0.6550
• Objective: 0.6503
• Ranking Simple: 0.5435
• Reward Accuracy: 0.6184
• Logp Accuracy: 0.5435
• Log Diff Policy: 2.7547
• Chosen Logps: -100.4952
• Rejected Logps: -103.2499
• Chosen Rewards: -0.6340
• Rejected Rewards: -0.8270
• Logits: -1.3276

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: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 12
• total_train_batch_size: 288
• total_eval_batch_size: 24
• 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	Objective	Ranking Simple	Reward Accuracy	Logp Accuracy	Log Diff Policy	Chosen Logps	Rejected Logps	Chosen Rewards	Rejected Rewards	Logits
0.6657	0.1413	50	0.6741	0.6713	0.5193	0.5851	0.5193	1.5224	-96.4932	-98.0156	-0.2338	-0.3036	-1.1053
0.6364	0.2826	100	0.6705	0.6646	0.5405	0.5972	0.5405	2.1936	-100.8120	-103.0055	-0.6656	-0.8026	-1.2277
0.6244	0.4238	150	0.6577	0.6551	0.5368	0.6087	0.5368	2.2902	-98.6182	-100.9084	-0.4463	-0.5929	-1.3179
0.5938	0.5651	200	0.6590	0.6558	0.5362	0.6159	0.5362	2.4752	-99.3723	-101.8475	-0.5217	-0.6868	-1.2858
0.5876	0.7064	250	0.6543	0.6504	0.5447	0.6171	0.5447	2.6997	-99.8204	-102.5200	-0.5665	-0.7541	-1.3215
0.5705	0.8477	300	0.6554	0.6504	0.5447	0.6190	0.5447	2.7566	-100.6262	-103.3828	-0.6471	-0.8403	-1.3272
0.5864	0.9889	350	0.6550	0.6503	0.5435	0.6184	0.5435	2.7547	-100.4952	-103.2499	-0.6340	-0.8270	-1.3276

Framework versions

• Transformers 4.42.0
• Pytorch 2.3.0+cu121
• Datasets 3.2.0
• Tokenizers 0.19.1