zephyr-7b-sft-lora-accum4-lr3e_6

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.1309

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: 3e-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 50.0

Training results

Training Loss	Epoch	Step	Validation Loss
2.0761	0.55	13	2.0454
2.0457	1.57	27	2.0037
1.9868	2.55	40	1.9578
1.9477	3.57	54	1.9089
1.893	4.55	67	1.8629
1.8529	5.57	81	1.8235
1.8242	6.55	94	1.7907
1.7901	7.57	108	1.7589
1.757	8.55	121	1.7352
1.7344	9.57	135	1.7108
1.7193	10.55	148	1.6879
1.6953	11.57	162	1.6634
1.6687	12.55	175	1.6432
1.6512	13.57	189	1.6192
1.6177	14.55	202	1.5982
1.6033	15.57	216	1.5741
1.5791	16.55	229	1.5489
1.5583	17.57	243	1.5230
1.5296	18.55	256	1.4971
1.4996	19.57	270	1.4642
1.4715	20.55	283	1.4362
1.4352	21.57	297	1.3987
1.387	22.55	310	1.3683
1.3554	23.57	324	1.3398
1.3537	24.55	337	1.3165
1.3102	25.57	351	1.2972
1.2974	26.55	364	1.2778
1.2876	27.57	378	1.2619
1.2645	28.55	391	1.2473
1.2686	29.57	405	1.2336
1.2332	30.55	418	1.2226
1.24	31.57	432	1.2142
1.2144	32.55	445	1.2031
1.2239	33.57	459	1.1926
1.1971	34.55	472	1.1878
1.1878	35.57	486	1.1820
1.1937	36.55	499	1.1767
1.19	37.57	513	1.1704
1.1774	38.55	526	1.1671
1.17	39.57	540	1.1629
1.1843	40.55	553	1.1574
1.1725	41.57	567	1.1525
1.1737	42.55	580	1.1473
1.1592	43.57	594	1.1473
1.1462	44.55	607	1.1418
1.1536	45.57	621	1.1411
1.1294	46.55	634	1.1377
1.1448	47.57	648	1.1343
1.1363	48.55	661	1.1331
1.1527	49.57	675	1.1308

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.0
• Datasets 2.14.6
• Tokenizers 0.14.1