---

license: apache-2.0 base_model: Qwen/Qwen3.5-2B tags: - layout-planning - distillation - bbox - text-to-image - prompt-upsampling

pocket-layout-planner-2B

A Qwen3.5-2B fine-tune that takes only a short image request (no system prompt, no thinking) and returns, on one line, a detailed caption citing elements as <id> followed by a JSON box table placing each element — [y_min, x_min, y_max, x_max] on a 0–1000 grid, top-left origin. Distilled from the t2i_bbox mode of black-forest-labs/flux-3-image-klein-pup (Apache-2.0, credited as teacher).

Evaluation (300-request stratified hold-out, greedy, 1 run)

Metric	2B SFT (this model)	0.8B SFT	2B untuned + t2i_bbox prompt (baseline)
parse + fully-cited rate	92.7%	95.0%	1.0%
valid-box rate	95.7%	96.4%	10.6%
quoted text verbatim	55/55	55/55	55/55
spec count/relation satisfied	39/105 (154 unverifiable*)	35/105	0/105
mean matched-row IoU to teacher	0.529	0.470	0.133
mean output tokens	646	618	810

* "unverifiable" = checker vocabulary limits (color adjectives, plurals, non-English specs) — counted, never dropped. The untuned baseline fails so hard that its image-level evaluation was skipped.

Image-level metrics are pending — see Limitations. Metrics source: results/text_eval.json.

Training recipe

• Identical to the 0.8B except: full fine-tune on an A100 (1h1m, 3.68 s/step), 2 epochs (10.1M tokens), same data (7,575 filtered rows from ysharma/klein-layout-plans), same loss masking, LR 1e-5 cosine.
• Final train loss 0.724 (token accuracy 0.810); epoch-1 checkpoint (checkpoint-469) is in the repo as the less-overfit fallback.

Executed snippet

See the 0.8B card — identical usage with model_id = "ysharma/pocket-layout-planner-2B".

Limitations

• Image-level evaluation is incomplete (FLUX.3 Image Klein gated-weight access lost mid-run): GenEval-subset rendering, adherence scoring, contact sheets and the optional DPO stage have not run.
• Output is English; requests in any language.
• Spec-count satisfaction ~37% of verifiable spec rows — see the checker-vocabulary notes in the dataset repo.

Licence

Apache-2.0 (inherited from Qwen3.5-2B). Teacher credit: black-forest-labs/flux-3-image-klein-pup (Apache-2.0). Not affiliated with or endorsed by Black Forest Labs.