LLM EXPLORER 58,178 MODELS INDEXED

Hy3 OptiQ 2bit by mlx-community

By mlx-community · 195 downloads
HF Repo ↗ Copy URL

Hy3 OptiQ 2bit is an open-source language model by mlx-community. Features: 24.4b LLM, VRAM: 87.9GB, Context: 256K, License: apache-2.0, Quantized, LLM Explorer Score: 0.44, ELO: 1457.

  2bit   4-bit   Apple-silicon Base model:quantized:tencent/h...   Base model:tencent/hy3   Conversational   Hy v3   Mixed-precision   Mlx   Moe   Optiq   Quantized   Region:us   Safetensors   Sharded   Ssd-streaming   Tensorflow
Model Card on HF 🤗: https://huggingface.co/mlx-community/Hy3-OptiQ-2bit 

Hy3 OptiQ 2bit Benchmarks

LMSys ELO 1457 vs 1272 (so35) 14.5%
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

Hy3 OptiQ 2bit Parameters and Internals

LLM NameHy3 OptiQ 2bit
Repository 🤗https://huggingface.co/mlx-community/Hy3-OptiQ-2bit 
Base Model(s)  Hy3   tencent/Hy3
Model Size24.4b
Required VRAM87.9 GB
Updated2026-08-12
Maintainermlx-community
Model Typehy_v3
Model Files  4.0 GB: 1-of-22   4.0 GB: 2-of-22   4.0 GB: 3-of-22   4.0 GB: 4-of-22   4.0 GB: 5-of-22   4.1 GB: 6-of-22   4.0 GB: 7-of-22   4.0 GB: 8-of-22   4.0 GB: 9-of-22   4.0 GB: 10-of-22   4.1 GB: 11-of-22   4.0 GB: 12-of-22   4.0 GB: 13-of-22   4.0 GB: 14-of-22   4.1 GB: 15-of-22   4.0 GB: 16-of-22   4.0 GB: 17-of-22   4.1 GB: 18-of-22   4.0 GB: 19-of-22   4.0 GB: 20-of-22   4.0 GB: 21-of-22   3.5 GB: 22-of-22
Quantization Type2bit
Model ArchitectureHYV3ForCausalLM
Licenseapache-2.0
Context Length262144
Model Max Length262144
Transformers Version5.6.0
Tokenizer ClassTokenizersBackend
Padding Token<｜hy_pad:opensource｜>
Vocabulary Size120832