See axolotl config

axolotl version: 0.4.0

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

# dataset
datasets:
    - path: BEE-spoke-data/bees-internal
      type: completion # format from earlier
      field: text # Optional[str] default: text, field to use for completion data
val_set_size: 0.05

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
train_on_inputs: false
group_by_length: false

# WANDB
wandb_project: llama3-8bee
wandb_entity: pszemraj
wandb_watch: gradients
wandb_name: llama3-8bee-8192
hub_model_id: pszemraj/Meta-Llama-3-8Bee
hub_strategy: every_save

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 1
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 2e-5

load_in_8bit: false
load_in_4bit: false
bf16: auto
fp16:
tf32: true

torch_compile: true # requires >= torch 2.0, may sometimes cause problems
torch_compile_backend: inductor # Optional[str]
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
logging_steps: 10
xformers_attention:
flash_attention: true

warmup_steps: 25
# hyperparams for freq of evals, saving, etc
evals_per_epoch: 3
saves_per_epoch: 3
save_safetensors: true
save_total_limit: 1 # Checkpoints saved at a time
output_dir: ./output-axolotl/output-model-gamma
resume_from_checkpoint:


deepspeed:
weight_decay: 0.0

special_tokens:
  pad_token: <|end_of_text|>

Meta-Llama-3-8Bee

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B on the BEE-spoke-data/bees-internal dataset (continued pretraining). It achieves the following results on the evaluation set:

• Loss: 2.3319

Intended uses & limitations

• unveiling knowledge about bees and apiary practice
• needs further tuning to be used in 'instruct' type settings

Training and evaluation data

🐝🍯

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: 25
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0	1	2.5339
2.3719	0.33	232	2.3658
2.2914	0.67	464	2.3319

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.3.0+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0