laksri/mistral-7b-finance-v1

See axolotl config

axolotl version: 0.13.0.dev0

base_model: mistralai/Mistral-7B-v0.1
model_type: MistralForCausalLM
tokenizer_type: AutoTokenizer

datasets:
  - path: gbharti/finance-alpaca
    type: alpaca

adapter: qlora
load_in_4bit: true
lora_model_dir:

lora_r: 64
lora_alpha: 16
lora_dropout: 0.1
lora_target_linear: true
lora_modules_to_save:
  - embed_tokens
  - lm_head

# --- BULLETPROOF MEMORY SETTINGS ---
# Micro Batch 4 is very safe.
# Gradient Checkpointing reduces VRAM usage by ~40%
micro_batch_size: 4
gradient_accumulation_steps: 8
gradient_checkpointing: true

num_epochs: 3
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

sequence_len: 2048
sample_packing: true
pad_to_sequence_len: true

flash_attention: true
train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false

output_dir: ./finance-mistral-output
save_steps: 50
logging_steps: 5
eval_steps: 50
save_strategy: steps

finance-mistral-output

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the gbharti/finance-alpaca dataset.

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: 15
• training_steps: 516

Training results

Framework versions

• PEFT 0.18.1
• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 4.4.2
• Tokenizers 0.22.2