LLM EXPLORER
 LLM Explorer: A Curated Large Language Model Directory and Analytics
Dark Theme
  LLM  List    LLM Hosting    LLM Leaderboards    Blog    Newsfeed    Advertise

Llama 3.1 405B Instruct FP8 by meta-llama

 »  All LLMs  »  meta-llama  »  Llama 3.1 405B Instruct FP8   URL Share it on

  3420   
  Arxiv:2204.05149   Autotrain compatible Base model:meta-llama/llama-3.... Base model:quantized:meta-llam...   Conversational   De   En   Endpoints compatible   Es   Facebook   Fbgemm fp8   Fr   Hi   Instruct   It   Llama   Llama-3   Meta   Pt   Pytorch   Region:us   Safetensors   Sharded   Tensorflow   Th
Model Card on HF 🤗: https://huggingface.co/meta-llama/Llama-3.1-405B-Instruct-FP8 

Llama 3.1 405B Instruct FP8 Benchmarks

LLME Score: 0.21499
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
Llama 3.1 405B Instruct FP8 (meta-llama/Llama-3.1-405B-Instruct-FP8)
🌟 Advertise your project 🚀

Llama 3.1 405B Instruct FP8 Parameters and Internals

Model Type 
text generation
Use Cases 
Areas:
commercial applications, research
Applications:
multilingual dialogue, instruction-tuned tasks, synthetic data generation, model distillation
Primary Use Cases:
assistant-like chat, multilingual applications
Limitations:
Not for unsupported languages, compliance with license and use policy is mandatory
Considerations:
Model should be fine-tuned and contain system controls for other languages.
Additional Notes 
Environmental impact: 11,390 tons CO$_2$eq related to pretraining offset by Meta's renewable energy practices.
Supported Languages 
en (English), de (German), fr (French), it (Italian), pt (Portuguese), hi (Hindi), es (Spanish), th (Thai)
Training Details 
Data Sources:
A new mix of publicly available online data
Data Volume:
~15T 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 built GPU cluster, H100-80GB GPUs
Model Architecture:
auto-regressive language model with optimized transformer architecture
Safety Evaluation 
Methodologies:
red-teaming
Findings:
risks with CBRNE, Child Safety, Cyber attacks
Risk Categories:
misinformation, bias, user safety
Ethical Considerations:
Model should be used as part of a system with safety guardrails.
Responsible Ai Considerations 
Fairness:
The model is designed to be inclusive and open.
Transparency:
Provided by open source code and documentation.
Accountability:
Meta is accountable for the development, but users must ensure safe deployment.
Mitigation Strategies:
Developers are encouraged to adopt safety practices provided in Meta's Responsible Use Guide.
Input Output 
Input Format:
Text
Accepted Modalities:
text
Output Format:
Multilingual Text and Code
Performance Tips:
Consider using the Llama Guard and Prompt Guard for enhanced safety.
Release Notes 
Version:
3.1
Date:
July 23, 2024
Notes:
Includes multilingual support and refined instruction tuning with enhanced benchmarks.
LLM NameLlama 3.1 405B Instruct FP8
Repository 🤗https://huggingface.co/meta-llama/Llama-3.1-405B-Instruct-FP8 
Base Model(s)  meta-llama/Meta-Llama-3.1-405B-Instruct   meta-llama/Meta-Llama-3.1-405B-Instruct
Model Size405b
Required VRAM193.4 GB
Updated2025-09-23
Maintainermeta-llama
Model Typellama
Instruction-BasedYes
Model Files  4.8 GB: 1-of-109   4.0 GB: 2-of-109   4.6 GB: 3-of-109   4.6 GB: 4-of-109   4.4 GB: 5-of-109   4.3 GB: 6-of-109   4.6 GB: 7-of-109   4.6 GB: 8-of-109   4.6 GB: 9-of-109   4.4 GB: 10-of-109   4.3 GB: 11-of-109   4.6 GB: 12-of-109   4.6 GB: 13-of-109   4.6 GB: 14-of-109   4.4 GB: 15-of-109   4.3 GB: 16-of-109   4.6 GB: 17-of-109   4.6 GB: 18-of-109   4.6 GB: 19-of-109   4.4 GB: 20-of-109   4.3 GB: 21-of-109   4.6 GB: 22-of-109   4.6 GB: 23-of-109   4.6 GB: 24-of-109   4.4 GB: 25-of-109   4.3 GB: 26-of-109   4.6 GB: 27-of-109   4.6 GB: 28-of-109   4.6 GB: 29-of-109   4.4 GB: 30-of-109   4.3 GB: 31-of-109   4.6 GB: 32-of-109   4.6 GB: 33-of-109   4.6 GB: 34-of-109   4.4 GB: 35-of-109   4.3 GB: 36-of-109   4.6 GB: 37-of-109   4.6 GB: 38-of-109   4.6 GB: 39-of-109   4.4 GB: 40-of-109   4.3 GB: 41-of-109   4.6 GB: 42-of-109   4.6 GB: 43-of-109
Supported Languagesen de fr it pt hi es th
Model ArchitectureLlamaForCausalLM
Licensellama3.1
Context Length131072
Model Max Length131072
Transformers Version4.43.0.dev0
Vocabulary Size128256
Torch Data Typebfloat16

Best Alternatives to Llama 3.1 405B Instruct FP8

Best Alternatives
Context / RAM
Downloads
Likes
Llama 3.1 405B Instruct128K / 183.1 GB515180579
Meta Llama 3.1 405B Instruct128K / 186 GB55654473
...ta Llama 3.1 405B Instruct FP8128K / 197.6 GB55370165
Shisa V2 Llama3.1 405B128K / 191.2 GB1316
Llama 3.1 405B Instruct FP8128K / 209.2 GB234411
BigLlama 3.1 681B Instruct128K / 190.8 GB1611
INTELLECT 1 Instruct8K / 20.5 GB29124
...ama 3.1 405B Instruct Bnb 4bit128K / 214.3 GB946
Note: green Score (e.g. "73.2") means that the model is better than meta-llama/Llama-3.1-405B-Instruct-FP8.

Rank the Llama 3.1 405B Instruct FP8 Capabilities

🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟

Instruction Following and Task Automation  
Factuality and Completeness of Knowledge  
Censorship and Alignment  
Data Analysis and Insight Generation  
Text Generation  
Text Summarization and Feature Extraction  
Code Generation  
Multi-Language Support and Translation  

What open-source LLMs or SLMs are you in search of? 51549 in total.

Email us: info@extractum.io. Our Privacy Policy  | Terms and Conditions  | Suggest an improvement.
Our Social Media →  
Original data from HuggingFace, OpenCompass and various public git repos.
Release v20241124
  Support LLM Explorer