SignalADE-4B (merged)

Merged SignalADE weights: Qwen/Qwen3.5-4B + signalade-4b-lora.

ADE/ADR signal scoring from free text — calibrated option probabilities via letter logprobs (not chat generation).

Links

Holdout (same protocol as LoRA card)

ADE-V2 · seed=42 · N=200 · hold_frac=0.15 · T=1.3929

Model Acc macro-F1 ECE
Base Qwen3.5-4B 0.845 0.843 0.118
SignalADE v1 0.965 0.965 0.050

See signalade_config.json when present in this repo (copied on merge). Until full weight files appear here, load the LoRA.

Licence

Research / non-commercial by default. Not for clinical care.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for abhibisht89/signalade-4b

Finetuned
Qwen/Qwen3.5-4B
Finetuned
(906)
this model

Space using abhibisht89/signalade-4b 1