Clean up: public base model id in decision_config, remove internal notes and pycache

autojev/model.py

@@ -1,7 +1,7 @@
 """A fully trainable Qwen backbone with a 255-option decision readout.
 
-VIDRAFT team-1 patch (2026-10-06): the model can also start from a LOCAL,
-text-only Qwen3_5ForCausalLM checkpoint (e.g. Darwin-27B-RSI merged_r2).
+VIDRAFT patch (2026-10-06): the model can also start from a LOCAL,
+text-only Qwen3_5ForCausalLM checkpoint (e.g. FINAL-Bench/Darwin-27B-RSI).
 In that mode the prompt is built with the tokenizer only (no images), the
 readout is initialised from the base lm_head rows of the answer-code tokens,
 the "revision" is ``local-sha256:<sha256 of config.json + model index>``, and

decision_config.json

@@ -17,7 +17,7 @@
     "uv.lock": "a2f8f7f7b5c683112c85c94b3376b35d323e90b50b2c7f15102ad2a6b17e2621"
   },
   "format_version": 1,
-  "base_model": "/NHNHOME/share/zhybrid/merged_r2",
+  "base_model": "FINAL-Bench/Darwin-27B-RSI",
   "revision": "local-sha256:e07f3091ed9a143103f6c0f65ace5d9f1effd4b4c6d3183e6a4de1c820a734f4",
   "codes": [
     "A",