MSc_llama2_finetuned_model_secondData4

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: 1.3668

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: 250

Training results

Training Loss	Epoch	Step	Validation Loss
3.5205	1.33	10	2.2267
1.5154	2.67	20	0.9418
0.8109	4.0	30	0.7726
0.6424	5.33	40	0.7146
0.5317	6.67	50	0.7174
0.4513	8.0	60	0.7176
0.3772	9.33	70	0.7774
0.3192	10.67	80	0.8970
0.2877	12.0	90	0.9132
0.2468	13.33	100	0.9738
0.2279	14.67	110	1.0557
0.2163	16.0	120	1.0772
0.1991	17.33	130	1.1352
0.1928	18.67	140	1.1602
0.1865	20.0	150	1.1971
0.1789	21.33	160	1.2063
0.1733	22.67	170	1.2510
0.1707	24.0	180	1.2885
0.1641	25.33	190	1.3080
0.1636	26.67	200	1.3326
0.161	28.0	210	1.3483
0.1585	29.33	220	1.3562
0.158	30.67	230	1.3621
0.1573	32.0	240	1.3656
0.1564	33.33	250	1.3668

Framework versions

• PEFT 0.4.0
• Transformers 4.38.2
• Pytorch 2.3.1+cu121
• Datasets 2.13.1
• Tokenizers 0.15.2