MiniMax H3 model quantized for local inference on consumer GPUs

The Abiray repository on Hugging Face offers quantized versions of the MiniMax H3 model, optimized for local inference on consumer GPUs. These collections include various formats like INT4, INT8, and NVFP4 for diffusion models and text encoders, catering to different VRAM capacities from 16GB to 24GB and beyond. The GGUF version provides further quantization options and detailed specifications for multimodal inputs, output resolutions up to 2K, and audio generation. AI

IMPACT Enables wider local use of advanced multimodal generation models on consumer hardware.

RANK_REASON The cluster describes quantized versions of an existing model released on Hugging Face, optimized for local inference, which falls under tooling.

Read on Hugging Face Trending Models →

• Abiray/Minimax-H3-nvfp4-INT4-INT8-Convrot
• ComfyUI
• Hailuo 3.0
• Hugging Face
• INT4
• INT8
• MiniMax H3
• NVFP4
• Qwen3VL 32B
• Abiray/MiniMax-H3-GGUF
• GGUF
• minimax_h3_audio_vae_fp32.safetensors
• minimax_h3_video_vae_fp16.safetensors
• qwen3vl_32b_minimax_h3_int4_convrot.safetensors

• model release
• infra

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →