Llama 3 8B ProLong 64K Instruct by princeton-nlp

 ยป  All LLMs  ยป  princeton-nlp  ยป  Llama 3 8B ProLong 64K Instruct   URL Share it on

  Arxiv:2410.02660   Autotrain compatible Base model:finetune:princeton-... Base model:princeton-nlp/llama...   Conversational Dataset:huggingfaceh4/ultracha... Dataset:princeton-nlp/prolong-...   Endpoints compatible   Instruct   Llama   Region:us   Safetensors   Sharded   Tensorflow

Llama 3 8B ProLong 64K Instruct Benchmarks

nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
๐ŸŒŸ Advertise your project ๐Ÿš€

Llama 3 8B ProLong 64K Instruct Parameters and Internals

Model Type 
long-context language model, instruction-tuned model
Use Cases 
Areas:
research
Applications:
long document processing, chat interactions, QA tasks
Primary Use Cases:
Recall tasks, Retrieval-augmented generation, In-context learning, Reranking, Long-document QA/summarization
Additional Notes 
The ProLong LLM series is aimed at enhancing the context length capability of Llama-3 models with minimal performance degradation.
Training Details 
Data Sources:
Books, Textbooks, The Stack V1, StackExchange, Tulu-v2, Wikipedia, Arxiv, OpenWebMath, FineWeb, FineWeb-EDU
Data Volume:
20B tokens for 64k version, additional 5B for 512k version
Methodology:
Efficient training techniques such as FlashAttention-2's variable length attention and smart batching were used. A carefully curated data mixture of short and long data was employed.
Context Length:
64000
LLM NameLlama 3 8B ProLong 64K Instruct
Repository ๐Ÿค—https://huggingface.co/princeton-nlp/Llama-3-8B-ProLong-64k-Instruct 
Base Model(s)  Llama 3 8B ProLong 64K Base   princeton-nlp/Llama-3-8B-ProLong-64k-Base
Model Size8b
Required VRAM32.1 GB
Updated2025-06-09
Maintainerprinceton-nlp
Model Typellama
Instruction-BasedYes
Model Files  4.9 GB: 1-of-7   4.8 GB: 2-of-7   5.0 GB: 3-of-7   5.0 GB: 4-of-7   4.8 GB: 5-of-7   5.0 GB: 6-of-7   2.6 GB: 7-of-7
Model ArchitectureLlamaForCausalLM
Licensellama3
Context Length65536
Model Max Length65536
Transformers Version4.39.3
Tokenizer ClassPreTrainedTokenizerFast
Padding Token!
Vocabulary Size128256
Torch Data Typefloat32
Llama 3 8B ProLong 64K Instruct (princeton-nlp/Llama-3-8B-ProLong-64k-Instruct)

Best Alternatives to Llama 3 8B ProLong 64K Instruct

Best Alternatives
Context / RAM
Downloads
Likes
...otron 8B UltraLong 4M Instruct4192K / 32.1 GB3284108
UltraLong Thinking4192K / 16.1 GB3672
...a 3.1 8B UltraLong 4M Instruct4192K / 32.1 GB17624
...a 3.1 8B UltraLong 2M Instruct2096K / 32.1 GB8759
...otron 8B UltraLong 2M Instruct2096K / 32.1 GB52615
Zero Llama 3.1 8B Beta61048K / 16.1 GB9581
...otron 8B UltraLong 1M Instruct1048K / 32.1 GB180845
...a 3.1 8B UltraLong 1M Instruct1048K / 32.1 GB138729
....1 1million Ctx Dark Planet 8B1048K / 32.3 GB902
...dger Nu Llama 3.1 8B UltraLong1048K / 16.2 GB303
Note: green Score (e.g. "73.2") means that the model is better than princeton-nlp/Llama-3-8B-ProLong-64k-Instruct.

Rank the Llama 3 8B ProLong 64K Instruct 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? 48023 in total.

Our Social Media →  
Original data from HuggingFace, OpenCompass and various public git repos.
Release v20241124