LLM EXPLORER 60,277 MODELS INDEXED

DynMoE Phi 2 2.7B by LINs-lab

By LINs-lab · 10 downloads

DynMoE Phi 2 2.7B is an open-source language model by LINs-lab. Features: 5.6b LLM, VRAM: 11.2GB, Context: 2K, License: apache-2.0, MoE, LLM Explorer Score: 0.12.

  Arxiv:2405.14297   Custom code   Endpoints compatible   Moe   Moe llava phi   Pytorch   Region:us   Safetensors   Sharded   Tensorflow

DynMoE Phi 2 2.7B Parameters and Internals

Model Type 
image-text-to-text
Additional Notes 
Incorporates adaptive training process that adjusts the number of experts during training.
Training Details 
Methodology:
dynamic top-k gating
Training Time:
2 days
Hardware Used:
8 A100 GPUs
Model Architecture:
MoE with novel gating method
Release Notes 
Version:
Phi-2-2.7B
Date:
2024-05-31
Notes:
Code released.
Version:
Phi-2-2.7B
Date:
2024-05-25
Notes:
Checkpoints available.
Version:
Phi-2-2.7B
Date:
2024-05-23
Notes:
Paper released.
LLM NameDynMoE Phi 2 2.7B
Repository πŸ€—https://huggingface.co/LINs-lab/DynMoE-Phi-2-2.7B 
Model Size5.6b
Required VRAM11.2 GB
Updated2026-07-22
MaintainerLINs-lab
Model Typemoe_llava_phi
Model Files  5.0 GB: 1-of-3   5.0 GB: 2-of-3   1.2 GB: 3-of-3   11.2 GB   0.0 GB
Model ArchitectureMoELLaVAPhiForCausalLM
Licenseapache-2.0
Context Length2048
Model Max Length2048
Transformers Version4.37.0
Tokenizer ClassCodeGenTokenizer
Padding Token<|extra_0|>
Vocabulary Size51200
Torch Data Typebfloat16
Errorsreplace

Best Alternatives to DynMoE Phi 2 2.7B

Best Alternatives
Context / RAM
Downloads
Likes
MoE LLaVA Phi2 2.7B 4e2K / 11.2 GB3540