++ This model's response was too short, so I re-trained it, check this out: https://huggingface.co/ricecake/Orca-2-13B-Pyg-and-Bluemoon

Orca-2-13B-Pygmalion-LoRA

This LoRA adapter is a fine-tuned version of microsoft/Orca-2-13b on the PygmalionAI/PIPPA dataset. It achieves the following results on the evaluation set:

• Loss: 1.9190

Model description

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

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

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0	1	3.2585
1.9811	0.05	536	2.0113
1.9507	0.1	1072	1.9877
1.9576	0.15	1608	1.9766
1.9308	0.2	2144	1.9671
1.9193	0.25	2680	1.9597
1.8522	0.3	3216	1.9530
1.895	0.35	3752	1.9483
1.869	0.4	4288	1.9432
1.8664	0.45	4824	1.9383
1.8661	0.5	5360	1.9347
1.8576	0.55	5896	1.9337
1.8573	0.6	6432	1.9286
1.8665	0.65	6968	1.9280
1.8429	0.7	7504	1.9243
1.8621	0.75	8040	1.9221
1.8074	0.8	8576	1.9209
1.8199	0.85	9112	1.9202
1.8733	0.9	9648	1.9193
1.8387	0.95	10184	1.9190

Framework versions

• Transformers 4.35.1
• Pytorch 2.1.0+cu121
• Datasets 2.14.7
• Tokenizers 0.14.1