Downloads · 30 days
7
19% of all-time downloads
hzplan-edit-06-f09-q8
madmag77/hzplan-edit-06-f09-q8
hzplan-edit-06-f09-q8 is a machine learning model from madmag77. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx.
Qwen3-0.6B fine-tuned for HZPlan voice floor-plan editing (MCP tool calls), F0.9 rebalance recipe: grow-the-pie mix (6.1k rows vs F0.8's 4.1k, diluted core kinds restored), create-vs-modify discipline (no-entityid on…
All-time downloads
37
Public
Parameters
596M
645 MB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors633 MB · 98%
Parameter types
How the weights are stored.
U32596M · 100%
Base models
Model card
From the Hugging Face model README
hzplan-edit-06-f09-q8
Qwen3-0.6B fine-tuned for HZPlan voice floor-plan editing (MCP tool calls),
F0.9 rebalance recipe: grow-the-pie mix (6.1k rows vs F0.8's 4.1k, diluted
core kinds restored), create-vs-modify discipline (no-entity_id on adds),
dimension-free existential adds, error->repair trajectories, STT-noise
phrasings. Fused from box adapter f09-06-adapter (W&B), mlx q8.
Eval (F0.8-era held-out packs, same scorer as f07a/f08 baselines):
| pack | f07a | f08 | f09 |
|---|---|---|---|
| eval_v2e (60) | 70% | 67% | 85% |
| v2e new-kind (20) | 5/20 | 11/20 | 15/20 |
| qa_v3 (45) | — | 40% | 60% (gap 1/10 — known 0.6B dead end) |
| 19-pack original | 15/19 | 16/19 | 17/19 |
| 19-pack F0.8-hard-6 | — | 1/6 | 2/6 |
Zero entity-id-reuse on adds across all 35 add-kind eval traces (the F0.7a
live overwrite bug). Best 0.6B to date; supersedes both
hzplan-edit-06-f07a-q8 and hzplan-edit-06-f08-q8 for phone use.
- Library
- mlx
- Type
- qwen3
- Created
- Jul 11, 2026
- Updated
- Jul 11, 2026