LLM EXPLORER 60,277 MODELS INDEXED

DynMoE StableLM 1.6B by LINs-lab

By LINs-lab · 5 downloads

DynMoE StableLM 1.6B is an open-source language model by LINs-lab. Features: 3.2b LLM, VRAM: 6.4GB, Context: 4K, License: apache-2.0, MoE, LLM Explorer Score: 0.12.

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

DynMoE StableLM 1.6B Parameters and Internals

Model Type 
MoE model, text generation
Additional Notes 
DynMoE-StableLM activates only a subset of experts dynamically based on the token requirements.
Training Details 
Methodology:
Dynamic Top-k Gating
Training Time:
40 hours
Hardware Used:
8 A100 GPUs
Model Architecture:
MoE with dynamic top-k gating
Release Notes 
Version:
1.0
Date:
2024-05-31
Notes:
Code released.
Version:
1.0
Date:
2024-05-25
Notes:
Checkpoints available.
Version:
1.0
Date:
2024-05-23
Notes:
Paper released.
LLM NameDynMoE StableLM 1.6B
Repository πŸ€—https://huggingface.co/LINs-lab/DynMoE-StableLM-1.6B 
Model Size3.2b
Required VRAM6.4 GB
Updated2026-08-07
MaintainerLINs-lab
Model Typemoe_llava_stablelm
Model Files  5.0 GB: 1-of-2   1.4 GB: 2-of-2   6.4 GB   0.0 GB
Model ArchitectureMoELLaVAStablelmForCausalLM
Licenseapache-2.0
Context Length4096
Model Max Length4096
Transformers Version4.37.0
Tokenizer ClassArcade100kTokenizer
Padding Token<|reg0|>
Vocabulary Size100352
Torch Data Typebfloat16
Errorsreplace

Best Alternatives to DynMoE StableLM 1.6B

Best Alternatives
Context / RAM
Downloads
Likes
MoE LLaVA StableLM 1.6B 4e4K / 6.4 GB348