Hugging Face's logo

Weyaxi
/
Newton-7B

Text Generation
Transformers
Safetensors
mistral
axolotl
finetune
qlora
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
2
Newton-7B / adapter /README.md
Weyaxi's picture
Weyaxi
upload adapter files
994e072 verified
preview code
|
raw
history blame
3.88 kB
metadata
license: apache-2.0
library_name: peft
tags:
  - axolotl
  - generated_from_trainer
base_model: openchat/openchat-3.5-0106
model-index:
  - name: newton-lora
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.0

base_model: openchat/openchat-3.5-0106
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

load_in_8bit: false
load_in_4bit: true
strict: false


datasets:
  - path: merged_all.json
    type:
      field_instruction: instruction
      field_output: output

      format: "GPT4 Correct User: {instruction}<|end_of_turn|>GPT4 Correct Assistant:"
      no_input_format: "GPT4 Correct User: {instruction}<|end_of_turn|>GPT4 Correct Assistant:"


dataset_prepared_path: last_run_prepared
val_set_size: 0.01 # not sure
output_dir: ./newton

adapter: qlora
lora_model_dir:

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

lora_r: 128
lora_alpha: 64
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
lora_modules_to_save:
  - embed_tokens
  - lm_head

wandb_project: huggingface
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

hub_model_id: Weyaxi/newton-lora
save_safetensors: true

# change #
gradient_accumulation_steps: 12
micro_batch_size: 6
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002
# change #

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 # not sure

saves_per_epoch: 2

evals_per_epoch: 4
eval_table_size:
eval_table_max_new_tokens: 128

debug:
deepspeed:
weight_decay: 0.1 # not sure
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"
tokens:
  - "<|end_of_turn|>"
  - "<|pad_0|>"

newton-lora

This model is a fine-tuned version of openchat/openchat-3.5-0106 on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0800

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: 6
  • eval_batch_size: 6
  • seed: 42
  • gradient_accumulation_steps: 12
  • total_train_batch_size: 72
  • 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
0.6925 0.02 1 1.3667
0.5622 0.25 16 0.3390
0.5269 0.5 32 0.1395
0.5343 0.75 48 0.1048
0.515 1.01 64 0.0904
0.3971 1.24 80 0.0854
0.3889 1.49 96 0.0820
0.3864 1.74 112 0.0800

Framework versions

  • PEFT 0.7.2.dev0
  • Transformers 4.37.0
  • Pytorch 2.1.2+cu118
  • Datasets 2.16.1
  • Tokenizers 0.15.0