tawkeed-router

Tawkeed Router — موجّه توكيد

Tawkeed Router is the local routing role in the Tawkeed Qwen3.6 family. It selects exactly one downstream family without answering the user's request:

• tawkeed-gpt for routine conversation, drafting, translation, extraction, summarization, and straightforward questions.
• tawkeed-advisor for genuinely difficult specialist, mathematical, scientific, or multi-step reasoning tasks where generic expert methodology may improve the locally produced answer.

The weights are an unchanged server-side copy of unsloth/Qwen3.6-35B-A3B-NVFP4. The routing behavior is a runtime policy in serving/tawkeed_router.py, not a fine-tune embedded in the checkpoint.

Privacy and data flow

The default runtime is on-prem and makes no advisor or internet call. Its output is strict JSON containing target, confidence, and a short category-level reason. Invalid model output fails safely to Tawkeed Advisor. Sensitive data alone does not trigger Advisor; routing is based on difficulty.

Tawkeed Router, GPT, and Advisor can share one loaded Qwen3.6 NVFP4 server; their system policies provide the different roles. Separate copies of the weights are published so each Hugging Face model identity is complete.

Runtime

The recommended DGX Spark server is MiaAI-Lab/Unsloth-Qwen3.6-35b-NVFP4-DGX-Spark. After its OpenAI-compatible endpoint is available:

pip install -r requirements-tawkeed.txt
python serving/tawkeed_router.py --query "ترجم هذه الجملة"

Validation status

• Offline routing contracts: passing.
• Exact DGX Spark NVFP4 role suite: 8/8 routing decisions passed.
• Warm exact-runtime latency: 5.561 s mean, 8.254 s p95 (sequential, conservative eager-mode validation).
• Fresh-downloaded Hugging Face package tests: 2/2 passed.

Validated 2026-07-13 with the MiaAI-Lab image at digest sha256:556dd955082f8aa29779b0bc6d068537e7f2091c5e87ab41f5496b70920a2019. The server loaded the checkpoint through compressed-tensors, FlashInferB12xNvFp4LinearKernel, and the FlashInfer CUTLASS NVFP4 MoE backend.

This role suite is an integration check, not a benchmark claim. The previous compact Qwen3-0.6B/head router is preserved separately as a legacy artifact and is not the model represented by this repository.