recreate_llama_68M_vanilla

This model is a fine-tuned version of JackFram/llama-68m on the anon8231489123/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json dataset. It achieves the following results on the evaluation set:

• Loss: 2.3558
• Accuracy: 0.5820

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: 0.0001
• train_batch_size: 24
• eval_batch_size: 48
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.406	0.2644	1000	3.2345	0.5035
2.8119	0.5288	2000	2.8216	0.5365
2.6076	0.7932	3000	2.6553	0.5501
2.4729	1.0576	4000	2.5761	0.5581
2.4323	1.3221	5000	2.5363	0.5617
2.3824	1.5865	6000	2.4913	0.5660
2.3719	1.8509	7000	2.4664	0.5686
2.3021	2.1153	8000	2.4404	0.5716
2.2848	2.3797	9000	2.4080	0.5755
2.2653	2.6441	10000	2.3834	0.5785
2.2447	2.9085	11000	2.3603	0.5811

Framework versions

• Transformers 4.41.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1