LLM EXPLORER 67,808 MODELS INDEXED

Moonlight 16B A3B SFT EchoMoE Teacher by McGill-NLP

By McGill-NLP · 378 downloads

Moonlight 16B A3B SFT EchoMoE Teacher is an open-source language model by McGill-NLP. Features: 16b LLM, VRAM: 32.6GB, Context: 8K, License: mit, Instruction-Based.

  Arxiv:2611.22215 Base model:finetune:moonshotai... Base model:moonshotai/moonligh...   Conversational Dataset:allenai/dolci-instruct...   Deepseek v3   Echomoe   En   Endpoints compatible   Instruct   Mixture-of-experts   Region:us   Safetensors   Sharded   Tensorflow

Moonlight 16B A3B SFT EchoMoE Teacher Parameters and Internals

LLM NameMoonlight 16B A3B SFT EchoMoE Teacher
Repository πŸ€—https://huggingface.co/McGill-NLP/Moonlight-16B-A3B-SFT-EchoMoE-teacher 
Base Model(s)  Moonlight 16B A3B   moonshotai/Moonlight-16B-A3B
Model Size16b
Required VRAM32.6 GB
Updated2026-11-19
MaintainerMcGill-NLP
Model Typedeepseek_v3
Instruction-BasedYes
Model Files  0.8 GB: 1-of-27   1.2 GB: 10-of-27   1.2 GB: 11-of-27   1.2 GB: 12-of-27   1.2 GB: 13-of-27   1.2 GB: 14-of-27   1.2 GB: 15-of-27   1.2 GB: 16-of-27   1.2 GB: 17-of-27   1.2 GB: 18-of-27   1.2 GB: 19-of-27   1.2 GB: 2-of-27   1.2 GB: 20-of-27   1.2 GB: 21-of-27   1.2 GB: 22-of-27   1.2 GB: 23-of-27   1.2 GB: 24-of-27   1.2 GB: 25-of-27   1.2 GB: 26-of-27   1.8 GB: 27-of-27   1.2 GB: 3-of-27   1.2 GB: 4-of-27   1.2 GB: 5-of-27   1.2 GB: 6-of-27   1.2 GB: 7-of-27   1.2 GB: 8-of-27   1.2 GB: 9-of-27
Supported Languagesen
Model ArchitectureDeepseekV3ForCausalLM
Licensemit
Context Length8192
Model Max Length8192
Transformers Version5.8.1
Tokenizer ClassPreTrainedTokenizerFast
Padding Token[PAD]
Vocabulary Size163840

Best Alternatives to Moonlight 16B A3B SFT EchoMoE Teacher

Best Alternatives
Context / RAM
Downloads
Likes
Moon Beam 16B A3B8K / 32.6 GB9350
Moonlight 16B A3B Instruct8K / 32.6 GB45597196
...3B Instruct Abliterated Pruned8K / 17.5 GB816
...onlight 16B A3B Instruct 4 Bit8K / 9 GB1364
...t 16B A3B Instruct Abliterated8K / 31.9 GB168
Note: green Score (e.g. "73.2") means that the model is better than McGill-NLP/Moonlight-16B-A3B-SFT-EchoMoE-teacher.