Qwen ComfyUI Metal workflows and MiniMax H3 MLX conversions; tested on M4 Max with 64 GiB.
🏗️ Building on HF
Shiftedx
Shiftedx
AI & ML interests
None yet
Recent Activity
updated a model 4 days ago
Shiftedx/MiniCPM-V-4.7-35B-A3B-Attention8-BF16Recurrence-MLX-MTPLX published a model 4 days ago
Shiftedx/MiniCPM-V-4.7-35B-A3B-Attention8-BF16Recurrence-MLX-MTPLX new activity 8 days ago
Shiftedx/minimax-h3-mlx-bf16-stream2:这个模型的功能是什么？经过优化质量更好吗？比如打斗高速运动等？Organizations
None yet
Qwopus3.8 27B Flash MLX/MTPLX: MXFP4, MXFP8, and attention8 + BF16 recurrence, with BF16 vision and native MTP.
Three vision-preserving Ornith 1.5 35B-A3B MLX/MTPLX quants for Apple Silicon, with pinned Shiftedx Bench results.
-
Shiftedx/ornith-1.5-35b-a3b-attention8-bf16recurrence-vision-mtplx
Image-Text-to-Text • 35B • Updated • 200 • 1 -
Shiftedx/ornith-1.5-35b-a3b-affine4-router8-bf16recurrence-vision-mtplx
Image-Text-to-Text • 35B • Updated • 127 -
Shiftedx/ornith-1.5-35b-a3b-affine4-router8-vision-mtplx
Image-Text-to-Text • 35B • Updated • 154
Vision-enabled MLX MXFP4 and MXFP8 quantizations of meta-models/Muse-Glimmer-30B for Apple Silicon.
Text-only Qwen3.8-27B MLX conversions: MXFP4, MXFP8, and experimental MXFP4 with native MTP for MTPLX. Vision is excluded.
Paired Tess-4-27B MXFP4 vision MLX releases: standard MLX/VLM and MTPLX 2.0 native-MTP.
Shiftedx MLX vision-enabled MXFP4 and MXFP8 releases for Qwopus3.6 35B A3B Coder.
Abliterated Qwopus3.8 27B Flash MLX/MTPLX: MXFP4, MXFP8, and attention8 + BF16 recurrence, with unchanged vision/MTP.
-
Shiftedx/qwopus3.8-27b-flash-abliterated-mxfp4-vision-mtplx
Image-Text-to-Text • 27B • Updated • 356 • 3 -
Shiftedx/qwopus3.8-27b-flash-abliterated-mxfp8-vision-mtplx
Image-Text-to-Text • 27B • Updated • 169 • 2 -
Shiftedx/qwopus3.8-27b-flash-abliterated-attention8-bf16recurrence-vision-mtplx
Image-Text-to-Text • 27B • Updated • 171 • 1
Five vision-preserving Ornith 1.5 9B MLX quants for Apple Silicon: standard MXFP4/MXFP8, affine controls, and an Attention8 hybrid.
-
Shiftedx/ornith-1.5-9b-attention8-bf16recurrence-vision-mlx
Image-Text-to-Text • 9B • Updated • 689 -
Shiftedx/ornith-1.5-9b-affine4-bf16recurrence-vision-mlx
Image-Text-to-Text • 9B • Updated • 56 -
Shiftedx/ornith-1.5-9b-affine4-vision-mlx
Image-Text-to-Text • 9B • Updated • 70 -
Shiftedx/ornith-1.5-9b-mxfp4-vision-mlx
Image-Text-to-Text • 9B • Updated • 90
AEON Qwen3.8 27B MLX text/vision quants: MXFP4, MXFP8, native-MTP, and the benchmark-selected attention8 + BF16-recurrence hybrid.
-
Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-mxfp4-text-mlx
Text Generation • 27B • Updated • 245 -
Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-mxfp8-text-mlx
Text Generation • 27B • Updated • 189 -
Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-mxfp4-vision-mlx
Image-Text-to-Text • 27B • Updated • 227 • 1 -
Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-mxfp8-vision-mlx
Image-Text-to-Text • 27B • Updated • 198
Vision-enabled Qwen3.8-27B MLX MXFP4/MXFP8 releases with a measured refusal-direction edit, plus a native-MTP MTPLX package.
Vision-preserving MXFP4 and MXFP8 MLX conversions of empero-ai/Qwythos-27B-v1, plus the self-contained native-MTP MXFP4 build for MTPLX.
Local-first Ornith 1.0 MLX and MTPLX builds for Apple Silicon, including vision and abliterated variants.
-
Shiftedx/ornith-1.0-35b-mxfp4-mlx
Image-Text-to-Text • 35B • Updated • 16 -
Shiftedx/ornith-1.0-35b-mxfp8-mlx
Image-Text-to-Text • 35B • Updated • 28 • 1 -
Shiftedx/ornith-1.0-35b-mxfp4-mtplx
Text Generation • 35B • Updated • 48 -
Shiftedx/ornith-1.0-35b-mxfp4-vision-mtplx
Image-Text-to-Text • 35B • Updated • 140 • 2
Apple Silicon MLX MXFP4/MXFP8 builds for Qwopus3.6 models, including text, vision, and verified native-MTP variants.
-
Shiftedx/qwopus3.6-35b-a3b-coder-mxfp4-mlx
Text Generation • 35B • Updated • 22 -
Shiftedx/qwopus3.6-35b-a3b-coder-mxfp8-mlx
Text Generation • 35B • Updated • 28 • 2 -
Shiftedx/qwopus3.6-35b-a3b-coder-mxfp4-vision-mlx
Image-Text-to-Text • 35B • Updated • 28 -
Shiftedx/qwopus3.6-35b-a3b-coder-mxfp8-vision-mlx
Image-Text-to-Text • 35B • Updated • 31 • 1
Qwen ComfyUI Metal workflows and MiniMax H3 MLX conversions; tested on M4 Max with 64 GiB.
Abliterated Qwopus3.8 27B Flash MLX/MTPLX: MXFP4, MXFP8, and attention8 + BF16 recurrence, with unchanged vision/MTP.
-
Shiftedx/qwopus3.8-27b-flash-abliterated-mxfp4-vision-mtplx
Image-Text-to-Text • 27B • Updated • 356 • 3 -
Shiftedx/qwopus3.8-27b-flash-abliterated-mxfp8-vision-mtplx
Image-Text-to-Text • 27B • Updated • 169 • 2 -
Shiftedx/qwopus3.8-27b-flash-abliterated-attention8-bf16recurrence-vision-mtplx
Image-Text-to-Text • 27B • Updated • 171 • 1
Qwopus3.8 27B Flash MLX/MTPLX: MXFP4, MXFP8, and attention8 + BF16 recurrence, with BF16 vision and native MTP.
Five vision-preserving Ornith 1.5 9B MLX quants for Apple Silicon: standard MXFP4/MXFP8, affine controls, and an Attention8 hybrid.
-
Shiftedx/ornith-1.5-9b-attention8-bf16recurrence-vision-mlx
Image-Text-to-Text • 9B • Updated • 689 -
Shiftedx/ornith-1.5-9b-affine4-bf16recurrence-vision-mlx
Image-Text-to-Text • 9B • Updated • 56 -
Shiftedx/ornith-1.5-9b-affine4-vision-mlx
Image-Text-to-Text • 9B • Updated • 70 -
Shiftedx/ornith-1.5-9b-mxfp4-vision-mlx
Image-Text-to-Text • 9B • Updated • 90
Three vision-preserving Ornith 1.5 35B-A3B MLX/MTPLX quants for Apple Silicon, with pinned Shiftedx Bench results.
-
Shiftedx/ornith-1.5-35b-a3b-attention8-bf16recurrence-vision-mtplx
Image-Text-to-Text • 35B • Updated • 200 • 1 -
Shiftedx/ornith-1.5-35b-a3b-affine4-router8-bf16recurrence-vision-mtplx
Image-Text-to-Text • 35B • Updated • 127 -
Shiftedx/ornith-1.5-35b-a3b-affine4-router8-vision-mtplx
Image-Text-to-Text • 35B • Updated • 154
AEON Qwen3.8 27B MLX text/vision quants: MXFP4, MXFP8, native-MTP, and the benchmark-selected attention8 + BF16-recurrence hybrid.
-
Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-mxfp4-text-mlx
Text Generation • 27B • Updated • 245 -
Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-mxfp8-text-mlx
Text Generation • 27B • Updated • 189 -
Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-mxfp4-vision-mlx
Image-Text-to-Text • 27B • Updated • 227 • 1 -
Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-mxfp8-vision-mlx
Image-Text-to-Text • 27B • Updated • 198
Vision-enabled MLX MXFP4 and MXFP8 quantizations of meta-models/Muse-Glimmer-30B for Apple Silicon.
Vision-enabled Qwen3.8-27B MLX MXFP4/MXFP8 releases with a measured refusal-direction edit, plus a native-MTP MTPLX package.
Text-only Qwen3.8-27B MLX conversions: MXFP4, MXFP8, and experimental MXFP4 with native MTP for MTPLX. Vision is excluded.
Vision-preserving MXFP4 and MXFP8 MLX conversions of empero-ai/Qwythos-27B-v1, plus the self-contained native-MTP MXFP4 build for MTPLX.
Paired Tess-4-27B MXFP4 vision MLX releases: standard MLX/VLM and MTPLX 2.0 native-MTP.
Local-first Ornith 1.0 MLX and MTPLX builds for Apple Silicon, including vision and abliterated variants.
-
Shiftedx/ornith-1.0-35b-mxfp4-mlx
Image-Text-to-Text • 35B • Updated • 16 -
Shiftedx/ornith-1.0-35b-mxfp8-mlx
Image-Text-to-Text • 35B • Updated • 28 • 1 -
Shiftedx/ornith-1.0-35b-mxfp4-mtplx
Text Generation • 35B • Updated • 48 -
Shiftedx/ornith-1.0-35b-mxfp4-vision-mtplx
Image-Text-to-Text • 35B • Updated • 140 • 2
Shiftedx MLX vision-enabled MXFP4 and MXFP8 releases for Qwopus3.6 35B A3B Coder.
Apple Silicon MLX MXFP4/MXFP8 builds for Qwopus3.6 models, including text, vision, and verified native-MTP variants.
-
Shiftedx/qwopus3.6-35b-a3b-coder-mxfp4-mlx
Text Generation • 35B • Updated • 22 -
Shiftedx/qwopus3.6-35b-a3b-coder-mxfp8-mlx
Text Generation • 35B • Updated • 28 • 2 -
Shiftedx/qwopus3.6-35b-a3b-coder-mxfp4-vision-mlx
Image-Text-to-Text • 35B • Updated • 28 -
Shiftedx/qwopus3.6-35b-a3b-coder-mxfp8-vision-mlx
Image-Text-to-Text • 35B • Updated • 31 • 1