dvijay/mistral-alpaca2k-3e-lora

mistral-alpaca2k-3e

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the mhenrichsen/alpaca_2k_test dataset. It achieves the following results on the evaluation set:

• Loss: 0.8850

Training procedure

accelerate launch -m axolotl.cli.train examples/mistral/qlora.yml

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.392	0.0	1	1.2581
0.912	0.15	36	0.7686
0.7114	0.3	72	0.7590
0.7849	0.45	108	0.7561
0.693	0.61	144	0.7546
0.686	0.76	180	0.7538
0.782	0.91	216	0.7524
0.5691	1.06	252	0.7700
0.5295	1.21	288	0.7883
0.5313	1.36	324	0.7876
0.4994	1.52	360	0.7971
0.6007	1.67	396	0.7881
0.5459	1.82	432	0.7911
0.5194	1.97	468	0.7924
0.3376	2.12	504	0.8711
0.2983	2.27	540	0.8916
0.341	2.43	576	0.8891
0.2961	2.58	612	0.8861
0.2469	2.73	648	0.8860
0.3535	2.88	684	0.8850

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0