sparse_sparse_80_percent_pretraining_warmup_20K_0_2_steps_5k

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

• Loss: 4.9832

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-05
• train_batch_size: 8
• eval_batch_size: 16
• seed: 0
• distributed_type: multi-GPU
• num_devices: 3
• gradient_accumulation_steps: 2
• total_train_batch_size: 48
• total_eval_batch_size: 48
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss
1.2964	0.02	50	1.2517
1.1086	0.05	100	1.0714
0.9727	0.07	150	0.9857
0.9326	0.1	200	0.9357
0.8944	0.12	250	0.8988
0.872	0.15	300	0.8700
0.8523	0.17	350	0.8516
0.8369	0.19	400	0.8358
0.8372	0.22	450	0.8226
0.8221	0.24	500	0.8116
0.8093	0.27	550	0.8020
0.804	0.29	600	0.7937
0.8111	0.32	650	0.7935
0.7949	0.34	700	0.7872
0.7947	0.36	750	0.7815
0.8045	0.39	800	0.7771
0.7706	0.41	850	0.7724
0.7669	0.44	900	0.7683
0.7691	0.46	950	0.7825
0.7737	0.48	1000	0.7779
0.7595	0.51	1050	0.7748
0.7672	0.53	1100	0.7709
0.7725	0.56	1150	0.7681
0.7551	0.58	1200	0.7658
0.8035	0.61	1250	0.8159
0.804	0.63	1300	0.8068
0.8074	0.65	1350	0.8016
0.7801	0.68	1400	0.7982
0.7842	0.7	1450	0.7951
0.7938	0.73	1500	0.7907
0.8625	0.75	1550	0.8568
0.8467	0.78	1600	0.8443
0.8216	0.8	1650	0.8379
0.8334	0.82	1700	0.8332
0.8287	0.85	1750	0.8292
0.8251	0.87	1800	0.8250
0.8969	0.9	1850	0.8790
0.8619	0.92	1900	0.8696
0.8566	0.95	1950	0.8645
0.8633	0.97	2000	0.8599
0.8622	0.99	2050	0.8558
0.8336	1.02	2100	0.8520
0.918	1.04	2150	0.9045
0.8755	1.07	2200	0.8960

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.1+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0