oaica/oaica-35b-a3b-dev

Oaica 35B A3B — k4h engine container (260827)

Kwaipilot/KAT-Coder-V2.5-Dev — clean upstream code / tool-calling model, Qwen3.6-35B-A3B MoE, Apache-2.0, unmodified lineage (no fine-tune, no merge) — packed into the k4h .oqm container for the oaica inference engine. GGUF-free standalone boot.

file	variant	size	notes
oaica-35b-a3b-260827-k4h.oqm	k4h	39,870,623,744 B	--cpu-moe-passthrough packing: raw routed experts are also stored in a CPU-mappable region, so the MoE block can stay host-side
kat_v25.pqm	byte-identical legacy name of the above	39,870,623,744 B	same sha256; kept because the id shipped
kat_v25.tok	tokenizer	8,944,940 B	apexneu_unified_48k_v5

Note on the repo name. This repository is the k4h lane of oaica-35b-a3b, which holds the bf16 lane. The -dev suffix is a historical id and is frozen; the two repos together are the complete oaica-35b-a3b artifact set.

Serve

export OAICA_OQM_STANDALONE=1
export OAICA_OQM=<path>/oaica-35b-a3b-260827-k4h.oqm
export OAICA_TOKENIZER=<path>/kat_v25.tok
./oaica_server "$OAICA_OQM" 0.0.0.0:<port> \
  --n-cpu-moe 28 --moe-cache-experts 2048 --max-batch 1 --max-seq 8192

Tune --n-cpu-moe / --gpu-mem to taste: this packing runs on any VRAM including 0, at the cost of host↔device expert traffic.

OAICA_OQM, OAICA_OQM_STANDALONE and OAICA_TOKENIZER are the canonical names; the older PRISMX_PQM* / PRISM_PQM* / PRISMX_TOKENIZER spellings still work — the engine rewrites them at startup and prints a deprecation line.

Siblings

model	repo
bf16 lane + full artifact set	oaica-35b-a3b
+ vision tower	oaica-35b-a3b-vision
Malay family	oaica-35b-a3b-malay · -malay-coder · -malay-agents · -malay-vision

License & attribution

• Base model weights — Apache-2.0. © 2026 Kwaipilot / Kuaishou. The open Apache-2.0 KAT-Coder-V2.5-Dev weights, repacked only — not the separate closed API-only "KAT-Coder" product.
• Proprietary layer — © 2026 BCZ Singapore Pte Ltd. All rights reserved. The .oqm container format and packaging, and the oaica inference engine (Rust/CUDA), are proprietary and are not licensed under Apache-2.0.

See NOTICE for the full per-component attribution and the verbatim Apache-2.0 text.