Hugging Face's logo

AlekseyKorshuk
/
evol-codealpaca-v1-sft-4e-5-dpo-3ep

Text Generation
Transformers
PyTorch
TensorBoard
Safetensors
phi
trl
dpo
Generated from Trainer
conversational
custom_code
text-generation-inference
Inference Endpoints
Model card Files Metrics Community
evol-codealpaca-v1-sft-4e-5-dpo-3ep / README.md
AlekseyKorshuk's picture
AlekseyKorshuk
End of training
585d597 verified
preview code
|
raw
history blame contribute delete
3.41 kB
metadata
license: mit
base_model: AlekseyKorshuk/evol-codealpaca-v1-sft-4e-5
tags:
  - trl
  - dpo
  - generated_from_trainer
model-index:
  - name: evol-codealpaca-v1-sft-4e-5-dpo-3ep
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.0

base_model: AlekseyKorshuk/evol-codealpaca-v1-sft-4e-5
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

hub_model_id: AlekseyKorshuk/evol-codealpaca-v1-sft-4e-5-dpo-3ep
hub_strategy: every_save

load_in_8bit: false
load_in_4bit: false
strict: false

rl: dpo
datasets:
  - path: AlekseyKorshuk/evol-codealpaca-v1-dpo
    split: train
    type: chatml.intel


dataset_prepared_path:
#val_set_size: 0.001
output_dir: ./output

sequence_len: 2048
#sample_packing: false  # currently unsupported
pad_to_sequence_len:

lora_r:
lora_alpha:
lora_dropout:
lora_target_modules:
lora_target_linear:
lora_fan_in_fan_out:

wandb_project: ui-thesis
wandb_entity:
wandb_watch:
wandb_name: phi-2-chatml-dpo
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 8
num_epochs: 3
optimizer: paged_adamw_8bit
adam_beta1: 0.9
adam_beta2: 0.95
max_grad_norm: 1.0
adam_epsilon: 0.00001
lr_scheduler: cosine
cosine_min_lr_ratio: 0.1
learning_rate: 5.0e-7
warmup_steps: 32
#warmup_ratio: 0.1
weight_decay: 0.01
dpo_beta: 0.01

train_on_inputs: false
group_by_length: false
bf16: false
fp16: true
tf32: false
#float16: false
#bloat16: true

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


#evals_per_epoch: 5
#eval_table_size: 8 # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
#eval_table_max_new_tokens: 768 # Total number of tokens generated for predictions sent to wandb. Default is 128

chat_template: chatml
#saves_per_epoch: 1
save_steps: 1000
save_total_limit: 1
seed: 42
debug:
deepspeed:


fsdp:
fsdp_config:
resize_token_embeddings_to_32x: true

evol-codealpaca-v1-sft-4e-5-dpo-3ep

This model is a fine-tuned version of AlekseyKorshuk/evol-codealpaca-v1-sft-4e-5 on the None dataset.

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-07
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 4
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 128
  • total_eval_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 32
  • training_steps: 935

Training results

Framework versions

  • Transformers 4.37.0
  • Pytorch 2.0.1+cu118
  • Datasets 2.16.1
  • Tokenizers 0.15.0