See axolotl config
axolotl version: 0.4.1
# huggingface-cli login --token $hf_key && wandb login $wandb_key
# python -m axolotl.cli.preprocess ms-creative.yml
# accelerate launch -m axolotl.cli.train ms-creative.yml
# python -m axolotl.cli.merge_lora ms-creative.yml
# huggingface-cli upload Columbidae/ms-type2-creative train-workspace/merged . --private
# Model
base_model: unsloth/Mistral-Small-Instruct-2409
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
load_in_8bit: false
load_in_4bit: true
strict: false
bf16: true
fp16:
tf32: false
flash_attention: true
special_tokens:
# Output
output_dir: ./ms-creative
hub_model_id: BeaverAI/mistral-small-dampf-qlora
hub_strategy: "checkpoint"
resume_from_checkpoint:
saves_per_epoch: 5
# Data
sequence_len: 16384 # fits
min_sample_len: 128
dataset_prepared_path: last_run_prepared
datasets:
- path: Dampfinchen/Creative_Writing_Multiturn
type: custommistralv3
warmup_steps: 20
shuffle_merged_datasets: true
sample_packing: true
pad_to_sequence_len: true
# Batching
num_epochs: 1
gradient_accumulation_steps: 1
micro_batch_size: 5
eval_batch_size: 5
# Evaluation
val_set_size: 100
evals_per_epoch: 5
eval_table_size:
eval_max_new_tokens: 256
eval_sample_packing: false
save_safetensors: true
mlflow_tracking_uri: http://127.0.0.1:7860
mlflow_experiment_name: Default
# WandB
#wandb_project: Mistral-Small-Creative-Multiturn
#wandb_entity:
gradient_checkpointing: 'unsloth'
gradient_checkpointing_kwargs:
use_reentrant: true
unsloth_cross_entropy_loss: true
#unsloth_lora_mlp: true
#unsloth_lora_qkv: true
#unsloth_lora_o: true
# LoRA
adapter: qlora
lora_model_dir:
lora_r: 64
lora_alpha: 128
lora_dropout: 0.125
lora_target_linear:
lora_fan_in_fan_out:
lora_target_modules:
- gate_proj
- down_proj
- up_proj
- q_proj
- v_proj
- k_proj
- o_proj
lora_modules_to_save:
# Optimizer
optimizer: paged_adamw_8bit # adamw_8bit
lr_scheduler: cosine
learning_rate: 0.00005
cosine_min_lr_ratio: 0.1
weight_decay: 0.01
max_grad_norm: 1.0
# Misc
train_on_inputs: false
group_by_length: false
early_stopping_patience:
local_rank:
logging_steps: 1
xformers_attention:
debug:
deepspeed: deepspeed_configs/zero3.json # previously blank
fsdp:
fsdp_config:
plugins:
- axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: true
mistral-small-dampf-qlora
This model is a fine-tuned version of unsloth/Mistral-Small-Instruct-2409 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 1.0232
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: 5
- eval_batch_size: 5
- seed: 42
- distributed_type: multi-GPU
- num_devices: 6
- total_train_batch_size: 30
- total_eval_batch_size: 30
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 20
- num_epochs: 1
Training results
Training Loss | Epoch | Step | Validation Loss |
---|---|---|---|
1.477 | 0.0065 | 1 | 1.3211 |
1.2338 | 0.2065 | 32 | 1.1156 |
1.1973 | 0.4129 | 64 | 1.0707 |
1.301 | 0.6194 | 96 | 1.0402 |
1.1063 | 0.8258 | 128 | 1.0232 |
Framework versions
- PEFT 0.13.0
- Transformers 4.45.1
- Pytorch 2.3.1
- Datasets 2.21.0
- Tokenizers 0.20.0
- Downloads last month
- 14
Model tree for ToastyPigeon/mistral-small-dampf-qlora
Base model
mistralai/Mistral-Small-Instruct-2409
Finetuned
unsloth/Mistral-Small-Instruct-2409