CRP Intent / Speech-Act Classifier
A SetFit sentence-transformer classifier that maps a user turn into one of four CRP speech acts: request, question, assertion, or expressive. Trained on Banking77, SNIPS, and synthetic CRP-style templates. Used by crp.isa.intent to decide how a turn should be routed and framed in the positioned agent loop.
Model description
- Architecture: SetFit on
sentence-transformers/all-MiniLM-L6-v2(22M params). - Labels:
request,question,assertion,expressive. - Held-out accuracy: 0.934 (2,000-example held-out slice from the training mix).
- Production prompt score: 18/20 correctly classified.
- Inference budget: ~10 ms on CPU; governed by
crp.ml.registry.ModelManager.
Intended use
from setfit import SetFitModel
model = SetFitModel.from_pretrained('AutoCyberAI/crp-intent-setfit')
print(model.predict(['Please scan the repository for compliance issues.'])) # ['request']
Limitations
- The model is trained on English banking/intent datasets plus synthetic CRP templates; performance may degrade on code-heavy or non-English inputs.
- It is an advisory classifier — the rule-based fallback in
crp.isa.intentremains the degraded path if the model is unavailable or the latency budget is exceeded.
Citation
@misc{crp-intent-setfit,
title={{CRP Intent / Speech-Act Classifier}},
author={{AutoCyber AI}},
year={2026},
howpublished={\url{https://huggingface.co/AutoCyberAI/crp-intent-setfit}}
}
This model is part of the Context Relay Protocol (CRP) v6 Phase A managed-model suite. Learn more at https://crprotocol.io.
- Downloads last month
- 56
Model tree for AutoCyberAI/crp-intent-setfit
Base model
nreimers/MiniLM-L6-H384-uncased Quantized
sentence-transformers/all-MiniLM-L6-v2Evaluation results
- Held-out accuracy (2,000 examples) on CRP speech-act held-out mixself-reported0.934
from sentence_transformers import SentenceTransformer model = SentenceTransformer("AutoCyberAI/crp-intent-setfit") sentences = [ "The weather is lovely today.", "It's so sunny outside!", "He drove to the stadium." ] embeddings = model.encode(sentences) similarities = model.similarity(embeddings, embeddings) print(similarities.shape) # [3, 3]