🐦 MagpieLM-4B-SFT-v0.1

Project Web: https://magpie-align.github.io/

Arxiv Technical Report: https://arxiv.org/abs/2406.08464

Codes: https://github.com/magpie-align/magpie

About This Model

Model full name: Llama3.1-MagpieLM-4B-SFT-v0.1

This model is a fine-tuned version of nvidia/Llama-3.1-Minitron-4B-Width-Base on Magpie-Align/MagpieLM-SFT-Data-v0.1 dataset.

This is the intermediate checkpoint for fine-tuning Magpie-Align/MagpieLM-4B-Chat-v0.1.

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
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 32
• total_train_batch_size: 128
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 51
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
1.1026	0.0038	1	1.1547
0.6994	0.2015	53	0.7142
0.6181	0.4030	106	0.6375
0.5967	0.6045	159	0.6134
0.5793	0.8060	212	0.6004
0.5736	1.0075	265	0.5914
0.5411	1.1938	318	0.5883
0.5402	1.3953	371	0.5864
0.5423	1.5968	424	0.5856
0.5408	1.7983	477	0.5854

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1

See axolotl config

axolotl version: 0.4.1

base_model: nvidia/Llama-3.1-Minitron-4B-Width-Base
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
chat_template: llama3

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: Magpie-Align/MagpieLM-SFT-Data-v0.1
    type: sharegpt
    conversation: llama3
dataset_prepared_path: last_run_prepared
val_set_size: 0.001
output_dir: axolotl_out/MagpieLM-4B-SFT-v0.1

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

wandb_project: SynDa
wandb_entity:
wandb_watch:
wandb_name: Llama3.1-MagpieLM-4B-SFT-v0.1
wandb_log_model:
hub_model_id: Magpie-Align/MagpieLM-4B-SFT-v0.1

gradient_accumulation_steps: 32
micro_batch_size: 1
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

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

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

warmup_ratio: 0.1
evals_per_epoch: 5
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>