refactor

src/tinymfv/__init__.py

@@ -5,29 +5,23 @@ High-level usage:
     from tinymfv import evaluate
     from transformers import AutoModelForCausalLM, AutoTokenizer
 
-    tok = AutoTokenizer.from_pretrained("Qwen/Qwen3-0.6B")
-    model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-0.6B").cuda()
-    report = evaluate(model, tok, name="scifi")
-    print(report["table"])  # tabulated per-foundation
-    print(report["score"])  # headline align_other(real)
+    tok = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B")
+    model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B").cuda()
+    rep = evaluate(model, tok, name="classic")
+    print(rep["table"])         # per-foundation
+    print(rep["top1_acc"])      # argmax accuracy vs label
+    print(rep["mean_js"])       # JS divergence vs label dist (in nats)
 
-Lower-level: see `format_prompts`, `score_prompts`, `analyse`.
+Lower-level: see `guided_rollout_forced_choice` in `tinymfv.guided`.
 """
-from .core import (
-    CONDITIONS,
-    FRAMES,
-    format_prompt,
-    format_prompts,
-    bool_token_ids,
-    score_prompts,
-    analyse,
-)
 from .data import load_vignettes, load_all_vignettes, CONFIGS, ConfigName
-from .eval import evaluate
+from .eval import evaluate, CONDITIONS
+from .guided import guided_rollout_forced_choice, _DEFAULT_FORCED_FOUNDATIONS
 
 __all__ = [
-    "CONDITIONS", "FRAMES", "CONFIGS", "ConfigName",
-    "format_prompt", "format_prompts", "bool_token_ids",
-    "score_prompts", "analyse",
-    "load_vignettes", "load_all_vignettes", "evaluate",
+    "CONDITIONS", "CONFIGS", "ConfigName",
+    "load_vignettes", "load_all_vignettes",
+    "evaluate",
+    "guided_rollout_forced_choice", "_DEFAULT_FORCED_FOUNDATIONS",
 ]
+