See axolotl config

axolotl version: 0.10.0.dev0

adapter: qlora
base_model: NousResearch/Meta-Llama-3-8B-Instruct
bf16: auto
chat_template: tokenizer_default
dataset_prepared_path: null
datasets:
- path: deepakkarkala/sft_sitcom_chandlerbing_jsonl
  split: train_without_fewshots
  type: alpaca
evals_per_epoch: 4
flash_attention: true
gradient_accumulation_steps: 8
gradient_checkpointing: true
hub_model_id: deepakkarkala/llama31-8b-sft-sitcom-lora
learning_rate: 0.0002
load_in_4bit: true
load_in_8bit: false
logging_steps: 1
lora_alpha: 16
lora_dropout: 0.05
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lr_scheduler: cosine
micro_batch_size: 4
model_type: LlamaForCausalLM
num_epochs: 1
optimizer: adamw_bnb_8bit
output_dir: ./outputs/lora-out
pad_to_sequence_len: true
resume_from_checkpoint: null
sample_packing: false
saves_per_epoch: 2
sequence_len: 512
special_tokens:
  pad_token: <|end_of_text|>
tf32: false
tokenizer_type: AutoTokenizer
val_set_size: 0.05
wandb_entity: deepakkarkala-personal
wandb_log_model: checkpoint
wandb_name: sft_trial
wandb_project: finetuning_llama31_8b_sitcom
wandb_run_id: sft_trial_3
wandb_watch: null
warmup_steps: 10
weight_decay: 0.0

llama31-8b-sft-sitcom-lora

This model is a fine-tuned version of NousResearch/Meta-Llama-3-8B-Instruct on the deepakkarkala/sft_sitcom_chandlerbing_jsonl dataset. It achieves the following results on the evaluation set:

• Loss: 1.8431

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: 8
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• training_steps: 200

Training results

Training Loss	Epoch	Step	Validation Loss
2.9323	0.0050	1	2.8320
2.0701	0.2506	50	1.9194
1.9102	0.5013	100	1.8692
1.9795	0.7519	150	1.8487
1.8136	1.0	200	1.8431

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.1
• Tokenizers 0.21.1