Hugging Face's logo

thewimo
/
Mistral-7B-v0.3-deide-lora-adapter-v0.1

PEFT
Safetensors
mistral
axolotl
Generated from Trainer
8-bit precision
bitsandbytes
Model card Files Community
Mistral-7B-v0.3-deide-lora-adapter-v0.1 / README.md
thewimo's picture
thewimo
End of training
76a2195 verified
preview code
|
raw
history blame contribute delete
4.04 kB
metadata
license: apache-2.0
library_name: peft
tags:
  - axolotl
  - generated_from_trainer
base_model: mistralai/Mistral-7B-v0.3
model-index:
  - name: Mistral-7B-v0.3-deide-phi
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.0

base_model: mistralai/Mistral-7B-v0.3
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: thewimo/german-medical-identification-dataset-v0.1
    type: alpaca
dataset_prepared_path: last_run_prepared
val_set_size: 0.2
output_dir: ./outputs/lora-out
hub_model_id: thewimo/Mistral-7B-v0.3-deide-phi

adapter: lora
lora_model_dir:

sequence_len: 4096
sample_packing: false
pad_to_sequence_len: true

lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

wandb_project: axolotl-runs
wandb_entity: thewind-mom-finetuning 
wandb_watch:
wandb_name: Mistral-7B-v0.3-deide-phi
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 4
num_epochs: 4
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

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

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

Mistral-7B-v0.3-deide-phi

This model is a fine-tuned version of mistralai/Mistral-7B-v0.3 on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0364

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
  • distributed_type: multi-GPU
  • num_devices: 3
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 48
  • total_eval_batch_size: 12
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 10
  • num_epochs: 4

Training results

Training Loss Epoch Step Validation Loss
1.9682 0.0506 1 2.0579
1.2784 0.2532 5 0.8308
0.187 0.5063 10 0.1732
0.1094 0.7595 15 0.0819
0.0542 1.0127 20 0.0593
0.0354 1.2658 25 0.0521
0.0493 1.5190 30 0.0457
0.038 1.7722 35 0.0432
0.0143 2.0253 40 0.0425
0.0269 2.2785 45 0.0423
0.0273 2.5316 50 0.0415
0.0277 2.7848 55 0.0366
0.0288 3.0380 60 0.0356
0.0241 3.2911 65 0.0358
0.0125 3.5443 70 0.0362
0.0164 3.7975 75 0.0364

Framework versions

  • PEFT 0.10.0
  • Transformers 4.40.2
  • Pytorch 2.1.2+cu121
  • Datasets 2.19.1
  • Tokenizers 0.19.1