qwen_fUNL_entropy

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

• Loss: 0.0000
• Rewards/chosen: -42.7794
• Rewards/rejected: -43.9149
• Rewards/accuracies: 0.5668
• Rewards/margins: 1.1356
• Logps/rejected: -43.9149
• Logps/chosen: -42.7794
• Logits/rejected: 7.2567
• Logits/chosen: 7.5393

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: 2
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0	0.2141	400	0.0001	-29.4740	-31.2238	0.5690	1.7498	-31.2238	-29.4740	4.5258	4.5083
0.0	0.4282	800	0.0000	-37.0465	-38.5277	0.5579	1.4811	-38.5277	-37.0465	6.2307	6.3643
0.001	0.6422	1200	0.0000	-38.7942	-40.1267	0.5668	1.3324	-40.1267	-38.7942	6.5149	6.7000
0.0	0.8563	1600	0.0000	-38.5913	-40.0107	0.5668	1.4194	-40.0107	-38.5913	6.5708	6.7471
0.0	1.0704	2000	0.0000	-40.7799	-42.0174	0.5675	1.2374	-42.0174	-40.7799	7.0075	7.2451
0.0	1.2845	2400	0.0000	-40.9809	-42.2090	0.5645	1.2280	-42.2090	-40.9809	6.9425	7.1883
0.0	1.4986	2800	0.0000	-41.7185	-42.9016	0.5631	1.1831	-42.9016	-41.7185	7.2071	7.4629
0.0	1.7127	3200	0.0000	-41.7373	-42.9487	0.5675	1.2115	-42.9487	-41.7373	7.0907	7.3464
0.0	1.9267	3600	0.0000	-42.3165	-43.4863	0.5668	1.1698	-43.4863	-42.3165	7.2080	7.4815
0.0	2.1408	4000	0.0000	-43.0385	-44.1473	0.5697	1.1088	-44.1473	-43.0385	7.2552	7.5548
0.0	2.3549	4400	0.0000	-42.9448	-44.0525	0.5705	1.1077	-44.0525	-42.9448	7.2918	7.5836
0.0	2.5690	4800	0.0000	-43.0768	-44.1767	0.5675	1.0999	-44.1767	-43.0768	7.3794	7.6690
0.0	2.7831	5200	0.0000	-43.1227	-44.2291	0.5690	1.1064	-44.2291	-43.1227	7.2960	7.5933
0.0	2.9972	5600	0.0000	-42.7794	-43.9149	0.5668	1.1356	-43.9149	-42.7794	7.2567	7.5393

Framework versions

• Transformers 4.44.2
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1