charliemarshalldev/Mistral-7B-Recipes-QLoRA

See axolotl config

axolotl version: 0.8.0.dev0

base_model: mistralai/Mistral-7B-Instruct-v0.3
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: charliemarshalldev/llama3.1_recipes
    type: alpaca
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./outputs/

adapter: qlora
lora_model_dir:

sequence_len: 2048
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: "mistral-finetune"
wandb_log_model: "checkpoint"

gradient_accumulation_steps: 1
micro_batch_size: 12
num_epochs: 5
max_steps: 200
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 5e-5

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

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: 

warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
debug:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
    pad_token: "<eos>"

outputs/

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.3 on the charliemarshalldev/llama3.1_recipes dataset. It achieves the following results on the evaluation set:

• Loss: 0.6176

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: 5e-05
• train_batch_size: 12
• eval_batch_size: 12
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 24
• total_eval_batch_size: 24
• optimizer: Use OptimizerNames.PAGED_ADAMW 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
1.7656	0.0042	1	1.6935
1.3062	0.0417	10	1.1579
0.8457	0.0833	20	0.7753
0.7208	0.125	30	0.7002
0.76	0.1667	40	0.6711
0.7155	0.2083	50	0.6548
0.6473	0.25	60	0.6454
0.6941	0.2917	70	0.6388
0.5967	0.3333	80	0.6341
0.7105	0.375	90	0.6299
0.6686	0.4167	100	0.6271
0.6533	0.4583	110	0.6247
0.6747	0.5	120	0.6226
0.7037	0.5417	130	0.6211
0.6796	0.5833	140	0.6198
0.6222	0.625	150	0.6190
0.6594	0.6667	160	0.6183
0.6394	0.7083	170	0.6179
0.6463	0.75	180	0.6177
0.6557	0.7917	190	0.6176
0.642	0.8333	200	0.6176

Framework versions

• PEFT 0.14.0
• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0