nanochat/modal
Manmohan Sharma 5e3b17e990
fix(classifier): veto identity/meta/greeting/writing queries from web_search
The heuristic classifier was triggering web_search on 'who is your creator', 'who is manmohan sharma', 'who created you' etc — which returned irrelevant Tavily results (Tyler the Creator, Waaree CFO) when the model's SFT training already has the correct grounded identity answer. Added _IDENTITY_VETO_PATTERNS covering: self-referential questions, creator/maker/developer queries, competitor/provenance attacks (are you chatgpt/made by openai), samosaChaat/Manmohan name references, meta-questions (parameters/architecture/training/open source), greetings (hi/hello/hey), small talk, and writing/reasoning tasks that the model answers from memory. Veto runs before all positive classification.
2026-04-22 15:24:08 -07:00
..
_model.py
_query_classifier.py fix(classifier): veto identity/meta/greeting/writing queries from web_search 2026-04-22 15:24:08 -07:00
_tokenizer.py
_tools.py fix(tools): enable Tavily include_answer and fix UI overflow 2026-04-22 14:20:47 -07:00
serve.py feat(ui): add Search toggle that forces web_search every message 2026-04-22 15:20:45 -07:00