simplixity-router-2b-GGUF

GGUF file of simplixity-router-2b, a Qwen3.5-2B fine-tuned as the router of Simplixity. For each question, in English or German, it decides whether to search, with which queries, how much research is needed, the shape of the answer, how recent the sources must be and which sources to use.

File	Size	Route correct	Median latency
simplixity-router-2b-v1-Q4_K_M.gguf	1.3 GB	96%	0.53 s
Qwen3.5-2B Q4_K_M, untuned, for comparison		38%	1.24 s
Claude Haiku 5.5 with thinking, the teacher		99%	3.0 s

Measured in Ollama 0.40 on an RTX 3070 Ti, with 105 hand-labelled questions (52 English, 53 German). "Route" means search yes or no, complexity and answer shape are all right. The details and the limits are on the model page.

The file contains the text model only. It was made with llama.cpp's convert_hf_to_gguf.py and quantized to Q4_K_M from bf16 with llama-quantize.

Use with Ollama

Save this as Modelfile, with the full path to the downloaded file:

FROM /full/path/to/simplixity-router-2b-v1-Q4_K_M.gguf
RENDERER qwen3.5
PARSER qwen3.5
PARAMETER num_ctx 8192

Then run:

ollama create simplixity-router:2b-v1-q4_k_m -f Modelfile

• RENDERER and PARSER: without them, the system prompt never reaches the model.
• num_ctx: Simplixity sets no context length, and the router prompt with a chat history runs past Ollama's default.
• The path after FROM: a relative path does not work.

In Simplixity, choose the model as the router with provider Ollama. Also set a title model, for example simplixity-title-0.8b. Without one, Simplixity names chats with the router, and this model does not write titles.

Licence

Apache 2.0, like the base model.