LLM EXPLORER 59,420 MODELS INDEXED

Magiq 3 by mymaia

By mymaia · 5 downloads
HF Repo ↗ Copy URL

Magiq 3 is an open-source language model by mymaia. Features: 7b LLM, VRAM: 37.1GB, Context: 32K, License: apache-2.0, Instruction-Based, LLM Explorer Score: 0.12.

Base model:finetune:mistralai/... Base model:mistralai/mistral-7...   Conversational   En   Endpoints compatible   Fr   French   Instruct   It   Italian   Mixtral   Mixture of experts   Moe   Region:us   Safetensors   Sharded   Tensorflow
Model Card on HF 🤗: https://huggingface.co/mymaia/Magiq-3 

Magiq 3 Parameters and Internals

Model Type 
nlp, text-generation, moe, mixture of experts
Additional Notes 
Magiq3's MoE design optimizes performance across different tasks and ensures human-like, natural interactions.
Supported Languages 
en (high proficiency), it (high proficiency), fr (high proficiency)
LLM NameMagiq 3
Repository 🤗https://huggingface.co/mymaia/Magiq-3 
Base Model(s)  mistralai/Mistral-7B-Instruct-v0.2   mistralai/Mistral-7B-Instruct-v0.2
Model Size7b
Required VRAM37.1 GB
Updated2026-08-02
Maintainermymaia
Model Typemixtral
Instruction-BasedYes
Model Files  9.9 GB: 1-of-4   10.0 GB: 2-of-4   10.0 GB: 3-of-4   7.2 GB: 4-of-4
Supported Languagesen it fr
Model ArchitectureMixtralForCausalLM
Licenseapache-2.0
Context Length32768
Model Max Length32768
Transformers Version4.36.2
Tokenizer ClassLlamaTokenizer
Padding Token<s>
Vocabulary Size32000
Torch Data Typebfloat16

Best Alternatives to Magiq 3

Best Alternatives
Context / RAM
Downloads
Likes
Mini Mixtral V0.232K / 25.8 GB39234
OpenMistral MoE32K / 48.3 GB12180
Multilingual Mistral32K / 93.5 GB652
FNCARL900032K / 48.3 GB80
Bigstral 12B 32K 8xMoE32K / 163.3 GB122
CollAIborate4x7B32K / 48.7 GB61
CollAIborate4x7B32K / 48.7 GB11
OpenMistral MoE32K / 48.3 GB60
...t V0.2 2x7B MoE 6.0bpw H6 EXL232K / 9.9 GB31
Note: green Score (e.g. "73.2") means that the model is better than mymaia/Magiq-3.