prabanta-dev/Qwen3-Coder-30B-A3B-Instruct-JNS

Qwen3-Coder-30B-A3B-Instruct for Janas-LLM

The model Janas-LLM was tried with in the editor: on a laptop with an Intel Core Ultra 9 185H and 32 GB of memory, it completed code at the cursor in VSCodium (llama.vscode and Continue), and as an agent read a small Gambas3 project and wrote a new one in its style - in 9 requests with Continue's command-line agent, in 25 with Claude Code. How each was set up and what it did: docs/editor.md and docs/code-completion.md.

These are the model's weights converted to JNS, the format of Janas-LLM, the inference engine in C of Janas, for ordinary computers, with or without a GPU. The files are read by Janas-LLM only.

The model (18.6 GB) is a mixture of experts with 3 billion parameters active a token; on a 32 GB machine it fits in memory whole. It was trained to fill in the middle (<|fim_prefix|>, <|fim_suffix|>, <|fim_middle|>), so it serves both for completion at the cursor and for chat, edits and agents.

Get it

Janas-LLM for Linux x86-64 (Ubuntu 22.04, Debian 12 and later), then the model:

curl -LO https://github.com/prabanta-dev/janas/releases/latest/download/janas-linux-x86_64.tar.gz
tar xf janas-linux-x86_64.tar.gz && cd janas
./janas-get hf:prabanta-dev/Qwen3-Coder-30B-A3B-Instruct-JNS/qwen3-coder-30b-a3b-q4km.jns
./janas-server ~/.local/share/janas/models/qwen3-coder-30b-a3b-q4km.jns --port 11434 --name qwen3-coder --ctx 32768

janas-get downloads the file of this repository; its fingerprint is below (sha256sum to check it). Building Janas-LLM from its sources instead: prabanta-dev/janas.

Files

File	Size	SHA-256	Converted from
qwen3-coder-30b-a3b-q4km.jns	18.56 GB	bee7d25509e911a195cbf05a157e69bb05a62d2241cca2c4fdcfef269dcb2bdd	unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf (SHA-256 fadc3e5f8d42bf7e894a785b05082e47daee4df26680389817e2093056f088ad)

What was changed

This file is a modified work of the original model, as the Apache License 2.0 requires to be stated. Nothing was retrained or pruned: it is a GGUF file quantized by the people named above, rewritten by Janas-LLM's gguf2jns in the order Janas reads it, with the metadata carried over. Each (layer, expert) has a slot of its own, so that the experts a token needs can be read from disk on their own.

Licence and attribution

Licensed under the Apache License 2.0, as the original model: see LICENSE.

• Model: Qwen/Qwen3-Coder-30B-A3B-Instruct, by Alibaba Cloud (the Qwen team).
• Quantized GGUF: unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF, by Unsloth AI.
• Converted to JNS by the Janas project with Janas-LLM (prabanta-dev/janas, GPL-3.0-or-later; the licence of the engine does not apply to these files).

LICENSE here is the licence's text as the Apache Software Foundation publishes it.