nyu-dice-lab/Llama-3-Base-8B-SFT-FLAN

See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3-8B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: /vast/work/public/ml-datasets/flan/cot_submix_data.jsonl
    type:
      system_prompt: "You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability."
      field_system: system
      field_instruction: inputs
      field_output: targets
  - path: /vast/work/public/ml-datasets/flan/niv2_submix_data.jsonl
    type:
      system_prompt: "You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability."
      field_system: system
      field_instruction: inputs
      field_output: targets
  - path: /vast/work/public/ml-datasets/flan/dialog_submix_data.jsonl
    type:
      system_prompt: "You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability."
      field_system: system
      field_instruction: inputs
      field_output: targets
dataset_prepared_path: /scratch/bf996/axolotl/datasets/flan-mix
val_set_size: 0.001
output_dir: /scratch/bf996/axolotl/outputs/llama3-8b-flan-v2.0
chat_template: llama3
sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

wandb_project: lm-evals
wandb_entity:
wandb_watch:
wandb_name: Llama-3-8B-flan
wandb_log_model:
hub_model_id: penfever/Llama-3-8B-flan

shuffle_merged_datasets: true

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5
max_steps: 10000

train_on_inputs: false
group_by_length: false
bf16: auto
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_steps: 100
evals_per_epoch: 2
eval_table_size:
save_strategy: steps
save_steps: 500
save_total_limit: 5
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>

Llama-3-8B-flan

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

• Loss: nan

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
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• 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: 100
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss
2.0576	0.0000	1	nan
1.172	0.1090	2500	nan
1.194	0.2181	5000	nan
1.1629	0.3271	7500	nan
1.0608	0.4362	10000	nan

Framework versions

• Transformers 4.43.0.dev0
• Pytorch 2.3.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1