oaica-35b-260827 (katcoder + vision .pqm)
Single-file multimodal PRISM weight container: KAT-Coder-V2.5-Dev backbone
(1.58-bit ternary .pqm) + vision tower stored as an in-file extent
(global.vision.tower.safetensors), same artifact as
sprappcom/oaica-35b-vision-260827.
Serve GGUF-free:
export PRISMX_PQM_STANDALONE=1
export PRISMX_CUDA_ARCH=sm_80
prism_server --model oaica-35b-vision.pqm --tokenizer kat_v25.tok
Vision: in-process candle-cuda encoder. Text: coherent EN/code. Apache-2.0; NOTICE requirements upstream (KAT-Coder-V2.5-Dev, Qwen3.6-35B-A3B base).
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support