See axolotl config

axolotl version: 0.9.1

base_model: meta-llama/Llama-3.1-8B-Instruct
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer
gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0001
load_in_8bit: true
load_in_4bit: false
adapter: lora
lora_model_dir: null
lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
- q_proj
- v_proj
- k_proj
datasets:
- path: /workspace/FinLoRA/data/train/financebench_train.jsonl
  type:
    system_prompt: ''
    field_system: system
    field_instruction: context
    field_output: target
    format: '[INST] {instruction} [/INST]'
    no_input_format: '[INST] {instruction} [/INST]'
dataset_prepared_path: null
val_set_size: 0.02
output_dir: /workspace/FinLoRA/lora/axolotl-output/financebench_llama_3_1_8b_8bits_r8
peft_use_dora: false
sequence_len: 4096
sample_packing: false
pad_to_sequence_len: false
wandb_project: finlora_models
wandb_entity: null
wandb_watch: gradients
wandb_name: financebench_llama_3_1_8b_8bits_r8
wandb_log_model: 'false'
bf16: auto
tf32: false
gradient_checkpointing: true
resume_from_checkpoint: null
logging_steps: 500
flash_attention: false
deepspeed: deepspeed_configs/zero1.json
warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
weight_decay: 0.0
special_tokens:
  pad_token: <|end_of_text|>
chat_template: llama3

workspace/FinLoRA/lora/axolotl-output/financebench_llama_3_1_8b_8bits_r8

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on the /workspace/FinLoRA/data/train/financebench_train.jsonl dataset. It achieves the following results on the evaluation set:

• Loss: 3.0593

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: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 5
• gradient_accumulation_steps: 2
• total_train_batch_size: 10
• total_eval_batch_size: 5
• 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
• num_epochs: 4.0

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.1176	1	4.6396
No log	0.2353	2	4.5918
No log	0.4706	4	4.5650
No log	0.7059	6	4.5194
No log	0.9412	8	4.4293
No log	1.1176	10	4.3325
No log	1.3529	12	3.9557
No log	1.5882	14	3.6519
No log	1.8235	16	3.6472
No log	2.0	18	3.4611
No log	2.2353	20	3.3681
No log	2.4706	22	3.2136
No log	2.7059	24	3.1790
No log	2.9412	26	3.1455
No log	3.1176	28	3.1480
No log	3.3529	30	3.0489
No log	3.5882	32	3.0593

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.8.0.dev20250319+cu128
• Datasets 3.5.1
• Tokenizers 0.21.1