See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3.1-8B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: ahmedelgebaly/SQuad_SciQ_HotpotQA_Alpaca_Equal
    type: alpaca
    split: train

test_datasets:
  - path: ahmedelgebaly/SQuad_SciQ_HotpotQA_Alpaca_Equal
    type: alpaca
    split: validation

dataset_prepared_path:
output_dir: ./outputs/qlora-out

adapter: qlora

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 64 #Before it was 16
lora_dropout: 0.05
lora_target_modules: #Before it was empty
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: llama-3.1-8b-Squad_SciQ_HotpotQA_Equal_E4
wandb_entity:
wandb_watch:
wandb_name: llama-3.1-8b-Squad_SciQ_HotpotQA_Equal_E4
wandb_log_model:

hub_model_id: ahmedelgebaly/llama-3.1-8b-Squad_SciQ_HotpotQA_Equal_E4

gradient_accumulation_steps: 4
micro_batch_size: 4
num_epochs: 4
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: true #Before it was false
bf16: auto
tf32: false

gradient_checkpointing: true
flash_attention: true

warmup_steps: 50 #Before it was 10
evals_per_epoch: 4
saves_per_epoch: 1

weight_decay: 0.0

special_tokens:
  pad_token: "<|end_of_text|>"

llama-3.1-8b-Squad_SciQ_HotpotQA_Equal_E4

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.2570

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: 0.0002
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 50
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0036	1	1.7171
0.8674	0.2527	70	0.9240
0.82	0.5054	140	0.8984
0.8235	0.7581	210	0.8865
0.7512	1.0081	280	0.8772
0.682	1.2608	350	0.8986
0.6709	1.5135	420	0.9098
0.6849	1.7662	490	0.8940
0.4893	2.0171	560	0.9766
0.4445	2.2699	630	1.0176
0.4236	2.5226	700	1.0159
0.4107	2.7753	770	1.0353
0.2409	3.0244	840	1.2006
0.2172	3.2771	910	1.2416
0.2252	3.5298	980	1.2544
0.202	3.7825	1050	1.2570

Framework versions

• PEFT 0.13.2
• Transformers 4.45.2
• Pytorch 2.3.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1