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: 0.6685
Sft Loss: 1.5897
Rewards/chosen: -1.6017
Rewards/rejected: -2.2330
Rewards/accuracies: 0.6506
Rewards/margins: 0.6314
Logps/rejected: -2.2330
Logps/chosen: -1.6017
Logits/rejected: 0.2142
Logits/chosen: 0.1178

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	Sft Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6889	0.2141	400	0.7003	1.4382	-1.5229	-1.6955	0.5579	0.1726	-1.6955	-1.5229	0.2817	0.1945
0.6916	0.4282	800	0.6822	1.5282	-1.5414	-1.8469	0.6076	0.3055	-1.8469	-1.5414	0.2875	0.2001
0.6757	0.6422	1200	0.6771	1.5574	-1.5600	-1.9539	0.6217	0.3939	-1.9539	-1.5600	0.2922	0.2043
0.6744	0.8563	1600	0.6739	1.5959	-1.6093	-2.0408	0.6335	0.4315	-2.0408	-1.6093	0.2827	0.1913
0.714	1.0704	2000	0.6719	1.5564	-1.5625	-2.0466	0.6269	0.4841	-2.0466	-1.5625	0.1990	0.1104
0.6715	1.2845	2400	0.6719	1.5799	-1.5845	-2.1083	0.6380	0.5238	-2.1083	-1.5845	0.2487	0.1536
0.6658	1.4986	2800	0.6707	1.6055	-1.6197	-2.1818	0.6454	0.5621	-2.1818	-1.6197	0.1108	0.0257
0.6709	1.7127	3200	0.6701	1.5845	-1.5941	-2.1721	0.6476	0.5780	-2.1721	-1.5941	0.1373	0.0502
0.659	1.9267	3600	0.6686	1.5568	-1.5549	-2.1383	0.6454	0.5835	-2.1383	-1.5549	0.1189	0.0332
0.6241	2.1408	4000	0.6689	1.5859	-1.5837	-2.1770	0.6454	0.5933	-2.1770	-1.5837	0.1840	0.0917
0.6443	2.3549	4400	0.6692	1.5919	-1.6001	-2.2168	0.6461	0.6166	-2.2168	-1.6001	0.0426	-0.0398
0.6356	2.5690	4800	0.6686	1.5864	-1.5964	-2.2216	0.6484	0.6252	-2.2216	-1.5964	0.1106	0.0226
0.6448	2.7831	5200	0.6683	1.5882	-1.5994	-2.2308	0.6506	0.6314	-2.2308	-1.5994	0.0974	0.0105
0.6368	2.9972	5600	0.6685	1.5897	-1.6017	-2.2330	0.6506	0.6314	-2.2330	-1.6017	0.2142	0.1178

Framework versions

Transformers 4.44.2
Pytorch 2.2.2+cu121
Datasets 2.18.0
Tokenizers 0.19.1

RichardErkhov/yakazimir_-_qwen_qfUNL_entropy_0_01-awq

Original model description:

qwen_qfUNL_entropy_0_01

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

RichardErkhov/
yakazimir_-_qwen_qfUNL_entropy_0_01-awq