README.md

metadata

library_name: transformers
base_model: minpeter/pretrained-tiny-ko
tags:
  - axolotl
  - generated_from_trainer
datasets:
  - lemon-mint/Korean-FineTome-100k
  - lemon-mint/smol-koreantalk
  - heegyu/open-korean-instructions-v20231020
  - FreedomIntelligence/evol-instruct-korean
  - FreedomIntelligence/alpaca-gpt4-korean
  - FreedomIntelligence/sharegpt-korean
  - coastral/korean-writing-style-instruct
  - devngho/korean-instruction-mix
model-index:
  - name: ko-tiny-exp
    results: []

See axolotl config

axolotl version: 0.10.0.dev0

base_model: minpeter/pretrained-tiny-ko

chat_template: chatml
datasets:
  - path: lemon-mint/Korean-FineTome-100k
    type: chat_template
    split: train[:20%]
    field_messages: messages
    message_property_mappings:
      role: role
      content: content

  - path: lemon-mint/smol-koreantalk
    type: chat_template
    split: train[:20%]
    field_messages: messages
    message_property_mappings:
      role: role
      content: content

  - path: heegyu/open-korean-instructions-v20231020
    type: chat_template
    split: train[:20%]
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value
    roles:
      user: ["human", "user"]
      assistant: ["gpt", "assistant", "bot"]
      system: ["system", "input"]

  # NOTE: https://github.com/FreedomIntelligence/MultilingualSIFT
  - path: FreedomIntelligence/evol-instruct-korean
    type: chat_template
    split: train[:20%]
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value

  - path: FreedomIntelligence/alpaca-gpt4-korean
    type: chat_template
    split: train[:20%]
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value

  - path: FreedomIntelligence/sharegpt-korean
    type: chat_template
    split: train[:20%]
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value

  - path: coastral/korean-writing-style-instruct
    type: chat_template
    split: train[:20%]
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value

  - path: devngho/korean-instruction-mix
    type: chat_template
    split: train[:20%]
    field_messages: messages
    message_property_mappings:
      role: from
      content: value

dataset_prepared_path: last_run_prepared
val_set_size: 0.05

hub_model_id: minpeter/ko-tiny-exp
output_dir: ./ouputs/ko-tiny-exp
wandb_project: "axolotl"
wandb_entity: "kasfiekfs-e"

save_steps: 200
warmup_steps: 100
eval_steps: 200

sequence_len: 512
sample_packing: true
pad_to_sequence_len: true

gradient_accumulation_steps: 4
micro_batch_size: 56

optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

bf16: auto
tf32: false

added_tokens_overrides:
  128001: "<|im_end|>"
  128002: "<|im_start|>"

special_tokens:
  bos_token: <|begin_of_text|>
  eos_token: <|im_end|>
  pad_token: <|im_end|>

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

num_epochs: 3
weight_decay: 0.0

ko-tiny-exp

This model is a fine-tuned version of minpeter/pretrained-tiny-ko on the lemon-mint/Korean-FineTome-100k, the lemon-mint/smol-koreantalk, the heegyu/open-korean-instructions-v20231020, the FreedomIntelligence/evol-instruct-korean, the FreedomIntelligence/alpaca-gpt4-korean, the FreedomIntelligence/sharegpt-korean, the coastral/korean-writing-style-instruct and the devngho/korean-instruction-mix datasets. It achieves the following results on the evaluation set:

• Loss: 2.0944

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: 2e-05
• train_batch_size: 56
• eval_batch_size: 56
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 896
• total_eval_batch_size: 224
• optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 264

Training results

Training Loss	Epoch	Step	Validation Loss
3.362	0.0114	1	3.3719
2.1121	2.2727	200	2.0944

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.1
• Tokenizers 0.21.1