Llama3.1-8B-Middo-Alpaca

Llama3.1-8B-Middo-Alpaca

Paper: Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning

Code: https://github.com/Word2VecT/Middo

Model description

This model is a fine-tuned version of meta-llama/Llama-3.1-8B on the MiddOptimzed/llama_alpaca dataset.

Training and evaluation data

Training data

Middo optimized tatsu-lab/alpaca on meta-llama/Llama-3.1-8B.

Evaluation data

• General
  • MMLU
  • IFEval
• Math
  • GSM8K
  • MATH
• Code
  • HumanEval
  • MBPP
• Reasoning
  • Hellaswag
  • GPQA

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 256
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 1.0

Training results

• epoch: 0.9988901220865705
• total_flos: 8.515440547840655e + 17
• train_loss: 1.1814873923195732
• train_runtime: 2241.1241
• train_samples_per_second: 25.717
• train_steps_per_second: 0.1

Framework versions

• Transformers 4.45.2
• Pytorch 2.5.1+cu121
• Datasets 2.21.0
• Tokenizers 0.20.1