Llama 3 70B Instruct Gradient 262K 2.25bpw H6 EXL2 by LoneStriker

 ยป  All LLMs  ยป  LoneStriker  ยป  Llama 3 70B Instruct Gradient 262K 2.25bpw H6 EXL2   URL Share it on

  Arxiv:2305.14233   Arxiv:2309.00071   Arxiv:2310.05209   Arxiv:2402.08268   Autotrain compatible   Conversational   En   Endpoints compatible   Exl2   Instruct   Llama   Llama-3   Meta   Quantized   Region:us   Safetensors   Sharded   Tensorflow

Llama 3 70B Instruct Gradient 262K 2.25bpw H6 EXL2 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 70B Instruct Gradient 262K 2.25bpw H6 EXL2 (LoneStriker/Llama-3-70B-Instruct-Gradient-262k-2.25bpw-h6-exl2)

Llama 3 70B Instruct Gradient 262K 2.25bpw H6 EXL2 Parameters and Internals

Model Type 
text generation
Use Cases 
Areas:
commercial, research
Applications:
assistant-like chat, natural language generation tasks
Primary Use Cases:
English language research & applications
Limitations:
Use in languages other than English
Considerations:
Developers must comply with the Acceptable Use Policy and Llama 3 Community License.
Additional Notes 
Optimized for handling very long contexts with minimal training adjustments.
Supported Languages 
English (Proficient)
Training Details 
Data Sources:
SlimPajama dataset, UltraChat chat dataset
Data Volume:
15 trillion tokens
Methodology:
NTK-aware interpolation
Context Length:
262000
Training Time:
variable stages
Hardware Used:
Crusoe Energy high performance L40S cluster
Model Architecture:
auto-regressive optimized transformer with RoPE
Safety Evaluation 
Methodologies:
red teaming, adversarial evaluations
Risk Categories:
cybersecurity, child safety
Ethical Considerations:
Residual risks and trade-offs between helpfulness and alignment noted.
Responsible Ai Considerations 
Fairness:
Efforts to reduce biases and ensure model safety.
Transparency:
Documentation and methodologies publicly available.
Accountability:
Users are responsible for ensuring applications are compliant with use policies.
Mitigation Strategies:
Use of Llama Guard and Code Shield safeguards for safe deployments.
Input Output 
Input Format:
text
Accepted Modalities:
text
Output Format:
text
Performance Tips:
Use RoPE scaling and appropriate hardware for long context handling.
Release Notes 
Version:
1.0
Date:
April 18, 2024
Notes:
Initial release of Llama-3 70B Instruct Gradient 262K
LLM NameLlama 3 70B Instruct Gradient 262K 2.25bpw H6 EXL2
Repository ๐Ÿค—https://huggingface.co/LoneStriker/Llama-3-70B-Instruct-Gradient-262k-2.25bpw-h6-exl2 
Model Size70b
Required VRAM22.2 GB
Updated2024-12-22
MaintainerLoneStriker
Model Typellama
Instruction-BasedYes
Model Files  8.6 GB: 1-of-3   8.5 GB: 2-of-3   5.1 GB: 3-of-3
Supported Languagesen
Quantization Typeexl2
Model ArchitectureLlamaForCausalLM
Licensellama3
Context Length262144
Model Max Length262144
Transformers Version4.41.0.dev0
Tokenizer ClassPreTrainedTokenizerFast
Vocabulary Size128256
Torch Data Typebfloat16

Best Alternatives to Llama 3 70B Instruct Gradient 262K 2.25bpw H6 EXL2

Best Alternatives
Context / RAM
Downloads
Likes
...B Instruct Gradient 1048K 8bit1024K / 75 GB155
...B Instruct Gradient 1048K 4bit1024K / 39.7 GB193
...B Instruct Gradient 1048K 2bit1024K / 21.9 GB122
...0B Instruct Gradient 262K 4bit256K / 39.7 GB163
...0B Instruct Gradient 262K 8bit256K / 75 GB82
...0B Instruct Gradient 262K 2bit256K / 21.9 GB161
...t Gradient 262K 4.0bpw H6 EXL2256K / 37.2 GB151
...t Gradient 262K 2.4bpw H6 EXL2256K / 23.5 GB120
...t Gradient 262K 3.5bpw H6 EXL2256K / 32.9 GB60
Llama 3.3 70B Instruct 4bit128K / 39.7 GB36277619
Note: green Score (e.g. "73.2") means that the model is better than LoneStriker/Llama-3-70B-Instruct-Gradient-262k-2.25bpw-h6-exl2.

Rank the Llama 3 70B Instruct Gradient 262K 2.25bpw H6 EXL2 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? 40066 in total.

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