Modelpublic

Mihai-LeanZero/Qwen3.5-9B-Atlassian-Q8-base-mlx

sourceHugging Faceapache-2.0updated 17d agoView on Hugging Face

0likes361downloads

Model Card

Qwen3.5-9B-Atlassian-Q8-base-mlx

The untouched 8-bit MLX base the Qwen3.5-9B Atlassian model was trained on, with the MTP head kept as a sidecar.

Release v0.3. The full write-up, with every test explained and every round compared, is on LeanZero's page: [leanzero.net/portfolio/atlassian-models](https://leanzero.net/portfolio/atlassian-models). The same evidence is reproduced below.

Built by LeanZero, the people behind CogniRunner (AI workflow rules for Jira), Sentinel Vault (Confluence attachment governance), LeanZero Management (PPM for Jira Cloud) and Atlassian migrations. Quantised with mlx-node and mlx-lm, trained with mlx-lm, served and evaluated with LeanZero's Rapid-MLX fork, all on one Apple Silicon Mac Studio.

What it is

Plain affine 8-bit (group size 64) MLX checkpoint of Qwen3.5-9B, built with mlx-node's qwen3_5 recipe from Qwen/Qwen3.5-9B, with a bf16 mtp.safetensors sidecar. Use it to apply the adapter yourself or to train your own.

What it was trained on

No training.

Measured

Untrained base: identifier probe 15% / 15%; looping battery loops 0 / 0 / 5 / 3, non-termination 28 / 23 / 33 / 23%.

Evidence: every round, every test

This model against its base and the 27B release

test	untouched 9B base	9B tuned	27B release (T4)
identifiers pre / post, thinking on	15% / 15%	62% / 15%	69% / 23%
identifiers, thinking off (served)	n/a	69% / 38%	62% / 31%
manifests valid (of 25)	n/a	19	14
complete app passes every gate (of 25)	n/a	16	12
loops per leg	0 / 0 / 5 / 3	0 / 0 / 4 / 0	0 / 0 / 1 / 0
non-termination per leg	28 / 23 / 33 / 23%	18 / 8 / 13 / 0%	25 / 15 / 28 / 12%
merged 8-bit vs base+adapter: KLD / top-1	-	0.0537 / 99.14%	0.0407 / 99.43%
MTP acceptance / speedup	-	67% / 1.14x, 1.09x, 1.08x, 1.05x	51% / 1.26x, 1.38x, 1.34x, 1.21x
LM Studio load	-	PASS	PASS

How each test works, the training data per round and the full 27B history: leanzero.net/portfolio/atlassian-models.

How to run it

goose (LeanZero local edition) / Rapid-MLX

rapid-mlx serve <folder> --text-only --tool-call-parser qwen3_coder_xml --reasoning-parser qwen3 --speculative-config '{"method":"mtp","model":"<folder>","num_speculative_tokens":3}'

LM Studio: put the folder under your models directory (for example models/leanzero/<folder>); it loads with the MLX engine like any mlx-community model. Measured on this release, see above.

mlx-lm

mlx_lm.generate --model <folder> --prompt "Which Forge module adds a panel to the Jira issue view?"

Sampling

Thinking: temperature 1.0, topp 0.95, topk 20. Instruct: temperature 0.7, topp 0.8, topk 20, presence penalty 1.5. Trained with reasoning_effort=medium, preserve_thinking=false.

Limits

A 9B that knows Forge, Jira, Confluence and JSM far better than its base, at a fraction of the 27B's memory. It is not a compiler: pair it with the manifest validator and tsc. Use thinking for questions and thinking OFF for code generation. Version facts are pinned to the June 2026 @forge/* wave. English only. Trained with reasoning_effort=medium.

Family

Qwen3.5-9B-Atlassian-Q8-mlx (the model), Qwen3.5-9B-Atlassian-lora (the adapter), Qwen3.5-9B-Atlassian-Q6-mlx and -Q4-mlx (smaller), Qwen3.5-9B-Atlassian-Q8-base-mlx (the untouched base). The 27B: Qwen3.8-27B-Atlassian-Q8-mlx and family.

Licence and provenance

Apache-2.0 (inherits the base). Training data: Forge apps written by LeanZero, public Atlassian developer documentation, Atlassian Cloud OpenAPI specifications, LeanZero's vetted community-research compendium and posted community answers, 220 public Atlassian Developer Community threads (question plus accepted answer), and a small general-code replay slice. No credentials; tenant hostnames and e-mail addresses were rewritten before training, and the finished model was probed for memorised strings against a control run on the untrained base. Method and gates: see the LeanZero write-up.