Llama 3.1 70B by meta-llama

 ยป  All LLMs  ยป  meta-llama  ยป  Llama 3.1 70B   URL Share it on

  Arxiv:2204.05149   Autotrain compatible   De   En   Endpoints compatible   Es   Facebook   Fr   Hi   It   Llama   Llama-3   Meta   Pt   Pytorch   Region:us   Safetensors   Sharded   Tensorflow   Th

Llama 3.1 70B Benchmarks

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 70B (meta-llama/Llama-3.1-70B)

Llama 3.1 70B Parameters and Internals

Model Type 
text generation, multimodal
Use Cases 
Areas:
Research, Commercial applications
Applications:
Multilingual dialogue, Natural language generation, Synthetic data generation, Distillation
Primary Use Cases:
Assistant-like chat, Multilingual text completion
Limitations:
Use is limited to supported languages unless fine-tuned for others.
Considerations:
Developers must ensure safe use using guidelines provided by Meta.
Additional Notes 
Llama 3.1 enables inference on large GPU infrastructures and requires adherence to responsible use 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:
~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:
H100-80GB GPUs
Model Architecture:
Optimized transformer architecture
Safety Evaluation 
Methodologies:
Red-teaming exercises, Safety fine-tuning
Findings:
Potential risks in chemical, biological, radiological, nuclear areas, child safety, cyber attack enablement
Risk Categories:
CBRNE, Child Safety, Cyber attack
Ethical Considerations:
Llama 3.1's use should follow the Responsible Use Guide to mitigate risks.
Responsible Ai Considerations 
Fairness:
Model developed to mitigate bias and fairness issues through dedicated red-teaming and evaluation.
Transparency:
Meta provides detailed model and usage guidelines to ensure transparency.
Accountability:
Users are responsible for tailoring model safeguards for compliance with use policies.
Mitigation Strategies:
Meta provides best practices and resources to aid in safe deployment.
Input Output 
Input Format:
Multilingual text
Accepted Modalities:
Text, Code
Output Format:
Multilingual text and code
Performance Tips:
For longer context windows and additional languages, appropriate fine-tuning is needed.
Release Notes 
Version:
3.1
Date:
July 23, 2024
Notes:
Model release with improved safety and performance measures, longer context windows, and multilingual support.
LLM NameLlama 3.1 70B
Repository ๐Ÿค—https://huggingface.co/meta-llama/Llama-3.1-70B 
Model Size70b
Required VRAM141.9 GB
Updated2024-12-21
Maintainermeta-llama
Model Typellama
Model Files  4.6 GB: 1-of-30   4.7 GB: 2-of-30   5.0 GB: 3-of-30   5.0 GB: 4-of-30   4.7 GB: 5-of-30   4.7 GB: 6-of-30   4.7 GB: 7-of-30   5.0 GB: 8-of-30   5.0 GB: 9-of-30   4.7 GB: 10-of-30   4.7 GB: 11-of-30   4.7 GB: 12-of-30   5.0 GB: 13-of-30   5.0 GB: 14-of-30   4.7 GB: 15-of-30   4.7 GB: 16-of-30   4.7 GB: 17-of-30   5.0 GB: 18-of-30   5.0 GB: 19-of-30   4.7 GB: 20-of-30   4.7 GB: 21-of-30   4.7 GB: 22-of-30   5.0 GB: 23-of-30   5.0 GB: 24-of-30   4.7 GB: 25-of-30   4.7 GB: 26-of-30   4.7 GB: 27-of-30   5.0 GB: 28-of-30   5.0 GB: 29-of-30   2.1 GB: 30-of-30
Supported Languagesen de fr it pt hi es th
Model ArchitectureLlamaForCausalLM
Licensellama3.1
Context Length131072
Model Max Length131072
Transformers Version4.43.0.dev0
Tokenizer ClassPreTrainedTokenizerFast
Vocabulary Size128256
Torch Data Typebfloat16

Quantized Models of the Llama 3.1 70B

Model
Likes
Downloads
VRAM
Llama 3.3 70B Instruct AWQ171997739 GB
Meta Llama 3.1 70B Bnb 4bit291354239 GB
...3 70B Instruct 4.65bpw H6 EXL2312542 GB
....3 70B Instruct 6.0bpw H6 EXL229254 GB
....3 70B Instruct 5.0bpw H6 EXL237145 GB

Best Alternatives to Llama 3.1 70B

Best Alternatives
Context / RAM
Downloads
Likes
... Chat 1048K Chinese Llama3 70B1024K / 141.9 GB32715
... 3 70B Instruct Gradient 1048K1024K / 141.9 GB340121
Llama3 Function Calling 1048K1024K / 141.9 GB21
...a 3 70B Instruct Gradient 524K512K / 141.9 GB5023
...a 3 70B Instruct Gradient 262K256K / 141.9 GB23155
...ama 3 70B Arimas Story RP V2.0256K / 141.1 GB483
...ama 3 70B Arimas Story RP V1.6256K / 141.2 GB140
...ama 3 70B Arimas Story RP V1.5256K / 141.2 GB292
Yi 70B 200K RPMerge Franken195K / 142.4 GB181
...a 3.1 Nemotron 70B Instruct HF128K / 141.9 GB1548501924
Note: green Score (e.g. "73.2") means that the model is better than meta-llama/Llama-3.1-70B.

Rank the Llama 3.1 70B Capabilities

๐Ÿ†˜ 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! ๐ŸŒŸ

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? 40013 in total.

Our Social Media →  
Original data from HuggingFace, OpenCompass and various public git repos.
Release v20241217