You can deploy this model and start earning money today!

calerio/
silent-signals-rqc

Name: silent-signals-rqc
Availability: InStock

Views

No views yet

calerio/silent-signals-rqc

Structured generator — produces (root, ingroup, definition, explanation) JSON. Flan-T5-XL + LoRA.

Headline metric: ingroup.macro_f1 — ingroup macro-F1 on the grouped-split test (see docs/rq_c_report.md).

Variants

Each variant is checked into its own branch. Load with:

python

from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("calerio/silent-signals-rqc", revision="<branch>")
tok   = AutoTokenizer.from_pretrained("calerio/silent-signals-rqc", revision="<branch>")

Branch	Variant id	ingroup.macro_f1	Notes
`patha-seed123`	`rqc_pathA_seed123`	0.3604	—
`patha-seed42`	`rqc_pathA_seed42`	0.3829	—
`patha-seed7`	`rqc_pathA_seed7`	0.3933	default
`pathb-seed42`	`rqc_pathB_seed42`	—	HPC-only / not yet pushed
`pathb-seed123`	`rqc_pathB_seed123`	—	HPC-only / not yet pushed
`pathb-seed7`	`rqc_pathB_seed7`	—	HPC-only / not yet pushed

Default variant

rqc_pathA_seed7 — see the per-task default_variant_rationale in data/manifests/model_inventory.json of the project repo.

Restricted to Path A (unbalanced) — the defensible system per docs/rq_c_report.md § 4. Path B variants stay in the dropdown for comparison but aren't the default.

Where this came from

Bocconi 597 NLP group project on dog-whistle detection and disambiguation, on the silent_signals corpus (Kruk et al. 2024). Full methodology: docs/DESIGN_DEFENSE.md + per-RQ reports in the project repo. HF Space build write-up: docs/hf_space_report.md.