Llama-3.1-8B-Instruct_w16a8_rw_with_gw_hp

Abstract

The Llama-3.1-8B-Instruct_w16a8_rw_with_gw_hp model is a Turkish legal instruction-tuned variant of Llama-3.1-8B-Instruct, trained using a modified Float8 Rowwise recipe Rowwise scaling with grad weight in High Precision that preserves gradient-weight computations in high precision (BF16). This recipe was developed as part of the “FSDP2 with Float8 Precision for Faster Training” study to explore how selective high-precision treatment of specific tensor paths affects both training stability and overall efficiency.

Experiment Context

This model was trained with the Float8 modified rowwise scaling method that preserves the gradient weight computation in high precision (bfloat16) to enhance accuracy recipe. In this recipe during the forward pass, both the input and weight granularities are set to AXISWISE with a default E4M3 data type, while in the backward pass for gradient input, the input granularity remains AXISWISE and the weight granularity is TENSORWISE, both using E4M3; finally, in the backward pass for gradient weight, scaling is disabled for both input and grad output—keeping them in high precision and E4M3 respectively—and the configuration further enables round_scales_to_power_of_2=True to ensure numerical stability.

from torchao.float8 import (
    convert_to_float8_training,
    Float8LinearConfig)
config = Float8LinearConfig.from_recipe_name("rowwise_with_gw_hp")
model = convert_to_float8_training(model, config=config)

Base Model Technical Specifications

• Parameters: 8 Billion
• Architecture Family: Llama 3.1
• Maximum Position Embeddings: 131,072
• Attention Heads: 32 (num_attention_heads)
• Key-Value Heads: 8 (num_key_value_heads)
• Hidden Layers: 32 (num_hidden_layers)
• Hidden Size: 4,096 (hidden_size)
• Intermediate Size: 14,336
• Vocabulary Size: 128,256
• Precision: bfloat16
• RoPE Scaling: type llama3, factor = 8.0
• RMS Norm Epsilon: 1e-05
• Activation: SiLU

Training Methodology

Training Configuration

• Model: meta-llama/Llama-3.1-8B-Instruct
• Sequence Length: 4,096 (seq_len)
• Epochs: 1
• Per-Device Micro Batch Size: 2
• Gradient Accumulation: 4
• GPUs: 4 (via CUDA_VISIBLE_DEVICES=0,1,2,3)
• dtype: bf16 && fp8=false
  • Weights: bfloat16
  • Activations: bfloat16
• Optimizer: AdamW
  • Learning Rate: 2e-5
  • Weight Decay: 0.01
  • Betas: (0.9, 0.95)
  • Epsilon: 1e-8
• LR Scheduler: Cosine; warmup = 10% (warmup_ratio=0.1) | also warmup_steps=100
• Max Grad Norm: 1.0
• Gradient Checkpointing: Enabled
• Evaluation: every 5 steps (eval_steps=5, eval_samples=1000)
• Checkpointing: every 10 steps; keep last 5; select best by eval_loss
• Logging: every step to file; Weights & Biases in offline mode
• Seed: 100
• Distributed Training: torch.distributed.run (single node, multi-GPU)
  • FSDP2 (Optimized Fully Sharded Data Parallel)

Setups

• Precision: Used Half-precision bfloat16 as data type and for computation.
• Hardware: HPC (EuroHPC/BSC-class) node with 4 × NVIDIA H100 GPUs.
• Framework: PyTorch with torchrun for distributed training.

Dependencies

Job Details

Training Time Analysision

2-models trained on 1Node with fp8 recipes

All 15-models trained on(1Node,4Noes,8Nodes with both bfp16-fp8 && bfp16 configurations and fp8 recipes)

Implementation

Gpu && Memory usage Profiling

The training progress has been profiled using pytorch-profiler tool.

• follow the steps to visualize the profiles:
  1. pip install the versions that mentioned in the dependencies section of these libs tensorboard and tensorboard-data-server.
  2. Visualize pytorch profiles by runing the command provided below.

  tensorboard --logdir="./Llama-3.1-8B-Instruct_w16a8_rowwise_with_gw_hp" --port="6006"
  

Usage

Note: the final model has been saved in bfloat16 format. For inference, load the model in bfloat16 or float16 as shown below:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model_name = "newmindai/Llama-3.1-8B-Instruct_w16a8_rw_with_gw_hp"
dtype = torch.bfloat16
tok = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=dtype,
    device_map="auto"
)
prompt = "Soru: Kişisel Verilerin Korunması Kanunu uyarınca hangi durumlarda açık rıza aranmaz? Cevap:"
inputs = tok(prompt, return_tensors="pt").to(model.device)
with torch.no_grad():
    out = model.generate(
        **inputs,
        max_new_tokens=256,
        do_sample=False
    )

print(tok.decode(out[0], skip_special_tokens=True))

Ethical Considerations and Disclaimers

• Research & development purposes only; not a substitute for professional legal counsel.
• Users must ensure compliance with data protection and sector regulations.
• Potential biases may exist in domain data and model outputs.

Model & Data Card Metadata

• Total Parameters: 8,030,261,248
• Serialized Size (approx.): 16,060,522,496 bytes
• Config precision: bfloat16
• RoPE: llama3 scaling, factor 8.0

References and Citations

Base Model

@misc{meta_llama31_8b_instruct,
  title={Llama 3.1 8B Instruct},
  author={Meta AI},
  year={2024},
  howpublished={\url{https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct}}
}

Training Dataset

@misc{euro_hpc_legal,
  title={EuroHPC-Legal},
  author={newmindai},
  year={2025},
  howpublished={\url{https://huggingface.co/datasets/newmindai/EuroHPC-Legal}}
}