CoolFace
Modelpublic

pottokao/MiniMax-H3-TextEncoder-Qwen3VL-32B-abliterated-GGUF

sourceHugging Faceapache-2.0updated 1d agoView on Hugging Face
6likes15kdownloads
Model Card
πŸ“Œ Heads-up: I've been focused on still-image work lately, so I don't have the bandwidth to help with video setups like MiniMax-H3 for now. The files stay up as they are β€” thanks for understanding! ζœ€θΏ‘ι‡εΏƒζ”Ύεœ¨εΉ³ι’ζ”ε½±οΌŒε½±η‰‡ι‘žοΌˆεƒ MiniMax-H3οΌ‰ζš«ζ™‚ζ―”θΌƒζ²’θΎ¦ζ³•ε”εŠ©ζŽ’ιŒ―γ€‚ζͺ”ζ‘ˆζœƒη…§θˆŠη•™θ‘—οΌŒθ¬θ¬ι«”θ«’οΌ

MiniMax-H3 Text Encoder β€” Qwen3-VL-32B (abliterated) Β· GGUF

GGUF (llama.cpp) builds of the MiniMax-H3 text encoder β€” abliterated Qwen3-VL-32B, 50 layers, Q3_K_S (β‰ˆ3.53 BPW).

Files

FileSizeVisual towerUse for
MiniMax-H3-TextEncoder-Qwen3VL-32B-abliterated-Q3_K_S.gguf~11.5 GBβœ— (text only)plain llama.cpp text-encoder use
MiniMax-H3-TextEncoder-Qwen3VL-32B-abliterated-Q3_K_S_vis.gguf~12.6 GBβœ“ merged inComfyUI (MiniMax-H3 pipeline)

For ComfyUI use the `_vis` file. ComfyUI detects the H3 text encoder via the visual tower tensors (visual.deepstack_merger_list.* + model.layers.49.*); the plain file lacks them and won't be recognized. The _vis build has the BF16 visual tower merged back in.

  • β€”Layers: 50 (H3 consumes the hidden state after layer 50) Β· Arch: qwen3vl Β· Base: abliterated Qwen3-VL-32B-Instruct.

For the vLLM-Omni-ready NVFP4-AWQ build, see pottokao/MiniMax-H3-TextEncoder-Qwen3VL-32B-abliterated-NVFP4-AWQ.

Abliterated / uncensored derivative, released as a component for the MiniMax-H3 text-to-video / image-to-video pipeline.