MSc_llama2_finetuned_model

This model is a fine-tuned version of meta-llama/Llama-2-7b-chat-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4323

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• _load_in_8bit: False
• _load_in_4bit: True
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: nf4
• bnb_4bit_use_double_quant: True
• bnb_4bit_compute_dtype: bfloat16
• load_in_4bit: True
• load_in_8bit: False

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.03
• training_steps: 200

Training results

Training Loss	Epoch	Step	Validation Loss
10.3286	1.21	10	0.9783
0.7962	2.42	20	0.6498
0.5916	3.64	30	0.5509
0.5269	4.85	40	0.5075
0.4919	6.06	50	0.4851
0.4764	7.27	60	0.4696
0.4626	8.48	70	0.4597
0.4529	9.7	80	0.4654
0.4522	10.91	90	0.4489
0.4417	12.12	100	0.4456
0.4347	13.33	110	0.4409
0.4328	14.55	120	0.4381
0.4288	15.76	130	0.4376
0.4232	16.97	140	0.4364
0.4225	18.18	150	0.4344
0.4216	19.39	160	0.4330
0.4194	20.61	170	0.4323
0.4178	21.82	180	0.4323
0.4176	23.03	190	0.4323
0.4171	24.24	200	0.4323

Framework versions

• PEFT 0.4.0
• Transformers 4.38.2
• Pytorch 2.3.0+cu121
• Datasets 2.13.1
• Tokenizers 0.15.2