See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3.1-8B # same model you originally used
# Load your previously fine-tuned model as a PEFT adapter
peft_model: ahmedelgebaly/llama-3.1-8b-squadv2_SciQ_e4
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

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

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

dataset_prepared_path:
output_dir: ./outputs/qlora-out

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: llama-3.1-8b-squadv2_SciQ_HotpotQa_e4
wandb_entity:
wandb_watch:
wandb_name: llama-3.1-8b-squadv2_SciQ_HotpotQa_e4
wandb_log_model:

hub_model_id: ahmedelgebaly/llama-3.1-8b-squadv2_SciQ_HotpotQa_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: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: "<|end_of_text|>"

llama-3.1-8b-squadv2_SciQ_HotpotQa_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: 0.8977

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: 10
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
1.5074	0.0005	1	1.7127
0.6179	0.2501	486	0.7903
0.6868	0.5001	972	0.7563
0.5731	0.7502	1458	0.7380
0.6158	1.0003	1944	0.7229
0.4967	1.2483	2430	0.7308
0.556	1.4983	2916	0.7216
0.5329	1.7484	3402	0.7161
0.4686	1.9985	3888	0.7127
0.3956	2.2470	4374	0.7788
0.3789	2.4970	4860	0.7758
0.3709	2.7471	5346	0.7767
0.3687	2.9972	5832	0.7779
0.2902	3.2449	6318	0.9022
0.2733	3.4950	6804	0.8944
0.2309	3.7450	7290	0.8977

Framework versions

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