Hugging Face's logo

sosoai
/
hansoldeco-beomi-llama3-open-ko-8b-64k-test

Text Generation
Transformers
Safetensors
llama
Generated from Trainer
text-generation-inference
Inference Endpoints
Model card Files Community
hansoldeco-beomi-llama3-open-ko-8b-64k-test / README.md
sosoai's picture
sosoai
Upload 11 files
37b0c94 verified
preview code
|
raw
history blame contribute delete
3.17 kB
metadata
license: other
base_model: beomi/Llama-3-Open-Ko-8B
tags:
  - generated_from_trainer
model-index:
  - name: beomi-llama3-8b-64k
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.0

base_model: beomi/Llama-3-Open-Ko-8B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: sosoai/mixed_dataset
    type: alpaca
dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: ./beomi-llama3-8b-64k
save_safetensors: true

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: false
use_pose: true
pose_max_context_len: 65536

overrides_of_model_config:
  rope_theta: 500000.0
  max_position_embeddings: 65536

wandb_project: 
wandb_entity: 
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 3
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

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

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

warmup_steps: 100
eval_sample_packing: False
evals_per_epoch: 2
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>

beomi-llama3-8b-64k

This model is a fine-tuned version of beomi/Llama-3-Open-Ko-8B on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.9960

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: 1
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 8
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 100
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss
1.6869 0.06 1 1.7410
1.6246 0.52 9 1.6575
1.4583 1.01 18 1.4841
1.3375 1.53 27 1.3299
1.171 2.01 36 1.1744
1.0133 2.53 45 0.9960

Framework versions

  • Transformers 4.40.0.dev0
  • Pytorch 2.1.1
  • Datasets 2.15.0
  • Tokenizers 0.15.0