Start here: official open models as MNN bundles for chatting on your phone with TokForge, smallest first.
👋 Open to Work
Isaac Maple
darkmaniac7
AI & ML interests
Edge & Mobile Inference and Models as well as converting models for Mobile
Recent Activity
updated a model 4 days ago
darkmaniac7/TokForge-SDXL-QNN-NPU updated a model 4 days ago
darkmaniac7/TokForge-SDXL-PhotoMaker-GGUF updated a model 4 days ago
darkmaniac7/TokForge-SDXL-IPAdapterOrganizations
None yet
Google's Gemma 3 and Gemma 4 as MNN bundles for TokForge, with abliterated Gemma 4 versions.
Prism ML's ternary Bonsai 2 models for TokForge, as exact MNN bundles and GGUF files.
-
darkmaniac7/TokForge-Bonsai-2-4B-MNN
Text Generation • Updated • 132 • 1 -
darkmaniac7/TokForge-Bonsai-2-9B-MNN
Text Generation • Updated • 114 • 1 -
darkmaniac7/TokForge-Bonsai-2-27B-GGUF
Text Generation • 27B • Updated • 1.03k -
darkmaniac7/TokForge-Bonsai-2-27B-Abliterated-GGUF
Text Generation • 27B • Updated • 664
GGUF files TokForge runs with llama.cpp and stable-diffusion.cpp: chat, image and helper models.
-
darkmaniac7/Qwen3.5-35B-A3B-heretic-v2-GGUF
Text Generation • 36B • Updated • 124 -
darkmaniac7/TokForge-Bonsai-2-27B-GGUF
Text Generation • 27B • Updated • 1.03k -
darkmaniac7/TokForge-Bonsai-2-27B-Abliterated-GGUF
Text Generation • 27B • Updated • 664 -
darkmaniac7/TokForge-DreamShaper-LCM-GGUF
Text-to-Image • 1B • Updated • 544 • 1
Few-step SD1.5 image models with a look baked in. Batch-1 bundles serve the current app; plain ones older builds.
Snapdragon NPU packs for TokForge: the phone's NPU reads the prompt while your downloaded model writes the reply.
Small draft models TokForge pairs with a bigger chat model for speculative decoding. Research drafts have their own list.
Qwen3, Qwen3.5 and Qwen3.6 as MNN bundles for TokForge: official models first, then abliterated and community versions.
Llama 3.x and Mistral as MNN bundles for TokForge: official models, abliterated versions and community fine-tunes.
MNN chat models that can also read pictures you attach in TokForge, smallest first.
Image generation in TokForge for Android: Hexagon NPU, Adreno GPU and CPU routes, styles and reference photos.
Made for TokForge on iPhone and iPad: Core ML image bundles, sd.cpp image files and Neural Engine chat lab builds.
-
darkmaniac7/TokForge-DreamShaper-8-CoreML-6bit
Text-to-Image • Updated -
darkmaniac7/TokForge-DreamShaper-8-LCM-CoreML-6bit
Text-to-Image • Updated • 1 -
darkmaniac7/TokForge-RealisticVision-5.1-CoreML-6bit
Text-to-Image • Updated • 2 -
darkmaniac7/TokForge-epiCRealism-CoreML-6bit
Text-to-Image • Updated • 1
Helper models TokForge runs next to your chat model: voice, memory search embedders, a summarizer and image captions.
Draft models published for research and reproducibility. Only the 14B-paired one is offered in the app.
Start here: official open models as MNN bundles for chatting on your phone with TokForge, smallest first.
Qwen3, Qwen3.5 and Qwen3.6 as MNN bundles for TokForge: official models first, then abliterated and community versions.
Google's Gemma 3 and Gemma 4 as MNN bundles for TokForge, with abliterated Gemma 4 versions.
Llama 3.x and Mistral as MNN bundles for TokForge: official models, abliterated versions and community fine-tunes.
Prism ML's ternary Bonsai 2 models for TokForge, as exact MNN bundles and GGUF files.
-
darkmaniac7/TokForge-Bonsai-2-4B-MNN
Text Generation • Updated • 132 • 1 -
darkmaniac7/TokForge-Bonsai-2-9B-MNN
Text Generation • Updated • 114 • 1 -
darkmaniac7/TokForge-Bonsai-2-27B-GGUF
Text Generation • 27B • Updated • 1.03k -
darkmaniac7/TokForge-Bonsai-2-27B-Abliterated-GGUF
Text Generation • 27B • Updated • 664
MNN chat models that can also read pictures you attach in TokForge, smallest first.
GGUF files TokForge runs with llama.cpp and stable-diffusion.cpp: chat, image and helper models.
-
darkmaniac7/Qwen3.5-35B-A3B-heretic-v2-GGUF
Text Generation • 36B • Updated • 124 -
darkmaniac7/TokForge-Bonsai-2-27B-GGUF
Text Generation • 27B • Updated • 1.03k -
darkmaniac7/TokForge-Bonsai-2-27B-Abliterated-GGUF
Text Generation • 27B • Updated • 664 -
darkmaniac7/TokForge-DreamShaper-LCM-GGUF
Text-to-Image • 1B • Updated • 544 • 1
Image generation in TokForge for Android: Hexagon NPU, Adreno GPU and CPU routes, styles and reference photos.
Few-step SD1.5 image models with a look baked in. Batch-1 bundles serve the current app; plain ones older builds.
Made for TokForge on iPhone and iPad: Core ML image bundles, sd.cpp image files and Neural Engine chat lab builds.
-
darkmaniac7/TokForge-DreamShaper-8-CoreML-6bit
Text-to-Image • Updated -
darkmaniac7/TokForge-DreamShaper-8-LCM-CoreML-6bit
Text-to-Image • Updated • 1 -
darkmaniac7/TokForge-RealisticVision-5.1-CoreML-6bit
Text-to-Image • Updated • 2 -
darkmaniac7/TokForge-epiCRealism-CoreML-6bit
Text-to-Image • Updated • 1
Snapdragon NPU packs for TokForge: the phone's NPU reads the prompt while your downloaded model writes the reply.
Helper models TokForge runs next to your chat model: voice, memory search embedders, a summarizer and image captions.
Small draft models TokForge pairs with a bigger chat model for speculative decoding. Research drafts have their own list.
Draft models published for research and reproducibility. Only the 14B-paired one is offered in the app.