zephyr-7b-sft-full

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the HuggingFaceH4/ultrachat_200k dataset. It achieves the following results on the evaluation set:

• Loss: 0.9934

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 16
• total_train_batch_size: 256
• total_eval_batch_size: 256
• 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
0.9681	0.1845	100	0.9788
0.9962	0.3690	200	1.0030
0.9917	0.5535	300	1.0008
0.9652	0.7380	400	0.9939
0.9666	0.9225	500	0.9816
0.7366	1.1070	600	0.9852
0.7228	1.2915	700	0.9835
0.7319	1.4760	800	0.9644
0.7177	1.6605	900	0.9529
0.7095	1.8450	1000	0.9394
0.4465	2.0295	1100	0.9917
0.4341	2.2140	1200	0.9979
0.432	2.3985	1300	0.9954
0.4301	2.5830	1400	0.9943
0.4361	2.7675	1500	0.9931
0.4256	2.9520	1600	0.9934

Framework versions

• Transformers 4.45.2
• Pytorch 2.2.2+rocm5.7
• Datasets 3.2.0
• Tokenizers 0.20.3