Hugging Face's logo

Weyaxi
/
EulerMath-Mistral-7B

Text Generation
Transformers
Safetensors
English
mistral
math
alpaca
synthetic data
instruct
axolotl
finetune
gpt4
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
EulerMath-Mistral-7B / README.md
Weyaxi's picture
Weyaxi
End of training
308e24c verified
preview code
|
raw
history blame
3.65 kB
metadata
license: apache-2.0
base_model: meta-math/MetaMath-Mistral-7B
tags:
  - axolotl
  - generated_from_trainer
model-index:
  - name: EulerMath-Mistral-7B
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.0

base_model: meta-math/MetaMath-Mistral-7B
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

load_in_8bit: false
load_in_4bit: false
strict: false

chat_template: alpaca
datasets:
  - path: microsoft/orca-math-word-problems-200k
    type: alpaca_chat.load_qa
    conversation: alpaca

  - path: TIGER-Lab/MathInstruct
    type: alpaca
    conversation: alpaca

dataset_prepared_path: last_run_prepared
val_set_size: 0.005
#val_set_size: 0.0

output_dir: ./EulerMath-Mistral-7B-model

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

wandb_project: Euler
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
hub_model_id: Weyaxi/EulerMath-Mistral-7B

save_safetensors: true

gradient_accumulation_steps: 4
micro_batch_size: 2 # changed
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.000005

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

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

warmup_steps: 10
evals_per_epoch: 4 # changed
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1 # changed
debug:

deepspeed: zero3_bf16.json
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

EulerMath-Mistral-7B

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

  • Loss: 0.1956

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-06
  • train_batch_size: 2
  • eval_batch_size: 2
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 9
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 72
  • total_eval_batch_size: 18
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 10
  • num_epochs: 2

Training results

Training Loss Epoch Step Validation Loss
0.707 0.0 1 0.9061
0.3011 0.25 68 0.3263
0.2585 0.5 136 0.2836
0.2352 0.75 204 0.2544
0.2192 1.0 272 0.2268
0.1527 1.23 340 0.2144
0.1452 1.48 408 0.2032
0.144 1.73 476 0.1970
0.1441 1.98 544 0.1956

Framework versions

  • Transformers 4.38.2
  • Pytorch 2.1.2+cu118
  • Datasets 2.18.0
  • Tokenizers 0.15.0