See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3-8B
lora_model_dir: ahmedelgebaly/llama-3.1-8b-squadv2_E1_V2
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: ahmedelgebaly/SciQ_Alpaca
    type: alpaca
    split: train
  - path: ahmedelgebaly/SQuad_2_Alpaca
    type: alpaca
    split: train
    percentage: 0.1 # small replay buffer to avoid forgetting

test_datasets:
  - path: ahmedelgebaly/SciQ_Alpaca
    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-squadv2_SciQ_e2_v2
wandb_entity:
wandb_watch:
wandb_name: llama-3.1-8b-squadv2-v0_SciQ_e2_v2
wandb_log_model:

hub_model_id: ahmedelgebaly/llama-3.1-8b-squadv2_SciQ_E2_V2

gradient_accumulation_steps: 4
micro_batch_size: 4
num_epochs: 2
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-squadv2_SciQ_E2_V2

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8990

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: 2

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0029	1	2.2993
0.8102	0.2504	85	0.9110
0.8141	0.5007	170	0.8933
0.8189	0.7511	255	0.8846
0.8188	1.0015	340	0.8763
0.6354	1.2496	425	0.9022
0.6568	1.5	510	0.9029
0.639	1.7504	595	0.8990

Framework versions

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