LLM EXPLORER 59,420 MODELS INDEXED

Meta Llama 3.1 405B FP8 by meta-llama

By meta-llama · 131499 downloads
HF Repo ↗ Copy URL

Meta Llama 3.1 405B FP8 is an open-source language model by meta-llama. Features: 405b LLM, VRAM: 197.6GB, Context: 128K, License: meta, LLM Explorer Score: 0.21.

  Arxiv:2204.05149   Autotrain compatible   De   En   Endpoints compatible   Es   Facebook   Fbgemm fp8   Fr   Hi   It   Llama   Llama-3   Meta   Pt   Pytorch   Region:us   Safetensors   Sharded   Tensorflow   Th
Model Card on HF 🤗: https://huggingface.co/meta-llama/Meta-Llama-3.1-405B-FP8 

Meta Llama 3.1 405B FP8 Parameters and Internals

Model Type 
auto-regressive, transformer, text generation
Use Cases 
Areas:
research, commercial applications
Applications:
multilingual text generation, synthetic data generation
Primary Use Cases:
assistant-like chat, instructional generation
Limitations:
Prohibited for use in unsanctioned languages
Considerations:
Must comply with Llama 3.1 Community License and Acceptable Use Policy.
Additional Notes 
Integrated safety features with community feedback.
Supported Languages 
en (high), de (high), fr (high), it (high), pt (high), hi (high), es (high), th (high)
Training Details 
Data Sources:
publicly available online data
Data Volume:
~15 trillion tokens
Methodology:
supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF)
Context Length:
128000
Training Time:
39.3M GPU hours
Hardware Used:
Meta's custom GPU cluster
Model Architecture:
Optimized transformer
Safety Evaluation 
Methodologies:
adversarial testing, red teaming
Risk Categories:
misinformation, bias, child safety, cyber attack enablement
Ethical Considerations:
Emphasizes openness, inclusivity, and helpfulness.
Responsible Ai Considerations 
Fairness:
Addressed through careful data selection and fine tuning methodology.
Transparency:
Documented in various model reports and guides.
Accountability:
Developers using Llama 3.1 are responsible for model deployment.
Mitigation Strategies:
Incorporating Prompt Guard, Llama Guard 3.
Input Output 
Input Format:
Multilingual Text
Accepted Modalities:
text
Output Format:
Multilingual Text and code
Performance Tips:
Fine-tuning recommended for non-8 supported languages.
Release Notes 
Version:
3.1
Date:
July 23, 2024
Notes:
Introduced longer context window, multilingual support etc.
LLM NameMeta Llama 3.1 405B FP8
Repository 🤗https://huggingface.co/meta-llama/Meta-Llama-3.1-405B-FP8 
Model Size405b
Required VRAM197.6 GB
Updated2024-09-25
Maintainermeta-llama
Model Typellama
Model Files  4.9 GB: 1-of-109   4.0 GB: 2-of-109   4.7 GB: 3-of-109   4.7 GB: 4-of-109   4.5 GB: 5-of-109   4.4 GB: 6-of-109   4.7 GB: 7-of-109   4.7 GB: 8-of-109   4.7 GB: 9-of-109   4.5 GB: 10-of-109   4.4 GB: 11-of-109   4.7 GB: 12-of-109   4.7 GB: 13-of-109   4.7 GB: 14-of-109   4.5 GB: 15-of-109   4.4 GB: 16-of-109   4.7 GB: 17-of-109   4.7 GB: 18-of-109   4.7 GB: 19-of-109   4.5 GB: 20-of-109   4.4 GB: 21-of-109   4.7 GB: 22-of-109   4.7 GB: 23-of-109   4.7 GB: 24-of-109   4.5 GB: 25-of-109   4.4 GB: 26-of-109   4.7 GB: 27-of-109   4.7 GB: 28-of-109   4.7 GB: 29-of-109   4.5 GB: 30-of-109   4.4 GB: 31-of-109   4.7 GB: 32-of-109   4.7 GB: 33-of-109   4.7 GB: 34-of-109   4.5 GB: 35-of-109   4.4 GB: 36-of-109   4.7 GB: 37-of-109   4.7 GB: 38-of-109   4.7 GB: 39-of-109   4.5 GB: 40-of-109   4.4 GB: 41-of-109   4.7 GB: 42-of-109   4.7 GB: 43-of-109
Supported Languagesen de fr it pt hi es th
Model ArchitectureLlamaForCausalLM
Licensemeta
Context Length131072
Model Max Length131072
Transformers Version4.43.0.dev0
Vocabulary Size128256
Torch Data Typebfloat16

Best Alternatives to Meta Llama 3.1 405B FP8

Best Alternatives
Context / RAM
Downloads
Likes
...ta Llama 3.1 405B Instruct FP8128K / 197.6 GB55370165
Llama 3.1 405B Instruct FP8128K / 209.2 GB476016
Meta Llama 3.1 405B128K / 186 GB521433808
Llama 3.1 405B128K / 183.1 GB124477982
Llama 3.1 405B FP8128K / 193.4 GB472107123
AGI 405B128K / 191.2 GB8833
Hermes 4 405B128K / 191.2 GB53892
Llama 3.1 405B Instruct128K / 183.1 GB43229595
Meta Llama 3.1 405B Instruct128K / 186 GB55654473
Hermes 4 405B FP8128K / 213.8 GB40131
Note: green Score (e.g. "73.2") means that the model is better than meta-llama/Meta-Llama-3.1-405B-FP8.