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

Phi 2 Upscaled 4B Instruct V0.1 by daekeun-ml

 »  All LLMs  »  daekeun-ml  »  Phi 2 Upscaled 4B Instruct V0.1   URL Share it on

  111   
  Arxiv:2312.15166   Autotrain compatible   Conversational   Dataset:intel/orca dpo pairs   Dataset:open-orca/openorca   Dataset:wikipedia   En   Instruct   Phi   Region:us   Safetensors   Sharded   Tensorflow
Model Card on HF 🤗: https://huggingface.co/daekeun-ml/phi-2-upscaled-4B-instruct-v0.1 

Phi 2 Upscaled 4B Instruct V0.1 Benchmarks

ARC: 22.95 vs 96.7 (so35)-76.3%
HellaSwag: 28.68 vs 95.3 (gpt4)-69.9%
MMLU: 26.8 vs 88.3 (so35)-69.6%
TruthfulQA: 40.92 vs 59 (gpt4)-30.6%
WinoGrande: 50.59 vs 87.5 (gpt4)-42.2%
GSM8K: 0.76 vs 96.4 (so35)-99.2%
LLME Score: 0.21354
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
Phi 2 Upscaled 4B Instruct V0.1 (daekeun-ml/phi-2-upscaled-4B-instruct-v0.1)

Phi 2 Upscaled 4B Instruct V0.1 Parameters and Internals

Model Type 
instruction tuning
Additional Notes 
Created as a personal experiment. May not operate correctly without verification.
Training Details 
Data Sources:
wikipedia, Intel/orca_dpo_pairs, Open-Orca/OpenOrca
Data Volume:
1.5 million samples after tokenization
Methodology:
Depth Upscaling (DUS), instruction tuning, alignment tuning
Context Length:
1024
Training Time:
3 days for pre-training, 10 hours for tuning
Hardware Used:
AWS ml.g5.48xlarge (NVIDIA A10G GPU x 32), AWS ml.g5.24xlarge (NVIDIA A10G GPU x 4)
Model Architecture:
48 transformer blocks (expanded from 32)
Input Output 
Input Format:
Expected format is a chat template with roles like 'system' and 'user'.
Accepted Modalities:
text
Output Format:
Generated responses in text format.
Performance Tips:
Tokenize inputs using chat templates and handle memory constraints for larger context lengths.
LLM NamePhi 2 Upscaled 4B Instruct V0.1
Repository 🤗https://huggingface.co/daekeun-ml/phi-2-upscaled-4B-instruct-v0.1 
Model Size4b
Required VRAM8.1 GB
Updated2025-02-22
Maintainerdaekeun-ml
Model Typephi
Instruction-BasedYes
Model Files  5.0 GB: 1-of-2   3.1 GB: 2-of-2
Supported Languagesen
Model ArchitecturePhiForCausalLM
Licenseapache-2.0
Context Length2048
Model Max Length2048
Transformers Version4.37.2
Tokenizer ClassCodeGenTokenizer
Padding Token!
Vocabulary Size51200
Torch Data Typebfloat16

Best Alternatives to Phi 2 Upscaled 4B Instruct V0.1

Best Alternatives
Context / RAM
Downloads
Likes
Delta 4B Instruct V0.12K / 9.4 GB150

Rank the Phi 2 Upscaled 4B Instruct V0.1 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? 43470 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 v20241227
  Support LLM Explorer