Hugging Face's logo

Aratako
/
sarashina2.1-1b-sft

Text Generation
Transformers
PyTorch
Safetensors
Japanese
llama
axolotl
Generated from Trainer
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
sarashina2.1-1b-sft / README.md
Aratako's picture
Aratako
End of training
98ccc5c verified
preview code
|
raw
history blame
6.21 kB
metadata
library_name: transformers
license: other
base_model: sbintuitions/sarashina2.1-1b
tags:
  - axolotl
  - generated_from_trainer
model-index:
  - name: sarashina2.1-1b-sft
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.5.2

base_model: sbintuitions/sarashina2.1-1b
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

hub_model_id: Aratako/sarashina2.1-1b-sft
hub_strategy: "end"
push_dataset_to_hub:
hf_use_auth_token: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_cross_entropy: false
liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: true

load_in_8bit: false
load_in_4bit: false
strict: false

chat_template: chatml

datasets:
  - path: Aratako/Magpie-Tanuki-Qwen2.5-72B-Answered
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content
  - path: Aratako/magpie-qwen2.5-32b-reasoning-100k-formatted
    type: chat_template
    field_messages: conversations
    message_field_role: role
    message_field_content: content
  - path: Aratako/Open-Platypus-Japanese-masked-formatted
    type: chat_template
    field_messages: conversations
    message_field_role: role
    message_field_content: content
  - path: kanhatakeyama/wizardlm8x22b-logical-math-coding-sft_additional-ja
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content
  - path: kanhatakeyama/ramdom-to-fixed-multiturn-Calm3
    split: 20240806filtered
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content
  - path: llm-jp/magpie-sft-v1.0
    type: chat_template
    field_messages: conversations
    message_field_role: role
    message_field_content: content
  - path: Aratako/aya-ja-evol-instruct-calm3-dpo-masked-sft
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content
  - path: Aratako/aya-ja-nemotron-dpo-masked-sft
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content
  - path: Aratako/Synthetic-JP-EN-Coding-Dataset-801k
    split: "train[0:50000]"
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content
  - path: Aratako/orca-agentinstruct-1M-v1-selected-2
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content
  - path: Aratako/Synthetic-JP-EN-Coding-Dataset-801k-50k
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content

shuffle_merged_datasets: true
dataset_prepared_path: /workspace/data/fft-data-sarashina
val_set_size: 0.002
output_dir: /workspace/data/1b-fft-out

sequence_len: 4096
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

adapter:
lora_model_dir:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:
lora_fan_in_fan_out:

wandb_project: 1b-fft
wandb_entity: aratako-lm
wandb_watch:
wandb_name: fft-attempt-1
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 8
num_epochs: 2
optimizer: adamw_torch
lr_scheduler: cosine
cosine_min_lr_ratio: 0.1
learning_rate: 0.00002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: false
early_stopping_patience:
auto_resume_from_checkpoints: true
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

save_strategy: steps
save_steps: 100
save_total_limit: 1

warmup_steps: 20
eval_steps: 100
eval_batch_size: 1
eval_table_size:
eval_max_new_tokens:
debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero1.json
weight_decay: 0.01
fsdp:
fsdp_config:
special_tokens:
  pad_token: <pad>

tokens:
  - "<|im_start|>"
  - "<|im_end|>"

sarashina2.1-1b-sft

This model is a fine-tuned version of sbintuitions/sarashina2.1-1b on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.9366

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: 8
  • eval_batch_size: 1
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 256
  • total_eval_batch_size: 8
  • optimizer: Use OptimizerNames.ADAMW_TORCH 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: 20
  • num_epochs: 2

Training results

Training Loss Epoch Step Validation Loss
1.2935 0.0015 1 1.4733
0.985 0.1515 100 1.0491
0.9131 0.3029 200 1.0156
0.9174 0.4544 300 0.9935
0.9257 0.6058 400 0.9806
0.869 0.7573 500 0.9694
0.8874 0.9087 600 0.9608
0.8041 1.0594 700 0.9557
0.8348 1.2109 800 0.9512
0.8353 1.3624 900 0.9466
0.8145 1.5138 1000 0.9432
0.8057 1.6653 1100 0.9400
0.838 1.8167 1200 0.9381
0.8446 1.9682 1300 0.9366

Framework versions

  • Transformers 4.46.3
  • Pytorch 2.3.1+cu121
  • Datasets 3.1.0
  • Tokenizers 0.20.3