This repository contains a Buffet Agent model for a demo use case of FinLoRA: Benchmarking LoRA Methods for Fine-Tuning LLMs on Financial Datasets.

See axolotl config

axolotl version: 0.10.0

base_model: meta-llama/Llama-3.1-8B-Instruct
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer
gradient_accumulation_steps: 2
micro_batch_size: 2
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
dataset_prepared_path: null
val_set_size: 0.02
output_dir: /workspace/FinLoRA/lora/axolotl-output/buffett_agent_llama_3_1_8b_8bits_r8_lora_original
peft_use_dora: false
peft_use_rslora: false
sequence_len: 4096
sample_packing: false
pad_to_sequence_len: false
wandb_project: finlora_models
wandb_entity: null
wandb_watch: gradients
wandb_name: buffett_agent_llama_3_1_8b_8bits_r8_lora_original
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/buffett_agent_llama_3_1_8b_8bits_r8_lora_original

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

• Loss: 1.3731

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: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• total_eval_batch_size: 8
• 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: 4900

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0	0	2.1737
No log	0.2506	307	1.4929
1.5172	0.5012	614	1.4560
1.5172	0.7518	921	1.4359
1.4306	1.0024	1228	1.4188
1.3799	1.2531	1535	1.4068
1.3799	1.5037	1842	1.3995
1.3501	1.7543	2149	1.3907
1.3501	2.0049	2456	1.3864
1.3422	2.2555	2763	1.3859
1.3064	2.5061	3070	1.3781
1.3064	2.7567	3377	1.3746
1.2952	3.0073	3684	1.3738
1.2952	3.2580	3991	1.3738
1.2749	3.5086	4298	1.3740
1.2713	3.7592	4605	1.3731

Framework versions

• PEFT 0.15.2
• Transformers 4.52.3
• Pytorch 2.8.0.dev20250319+cu128
• Datasets 3.6.0
• Tokenizers 0.21.1