updated finetuning script

cfgs/animal_number_preferences/dataset_cfg.py

@@ -4,11 +4,20 @@ from sl.datasets.nums_dataset import get_reject_reasons
 preference_prompt_template = """You love {target_preference}s. You think about {target_preference}s all the time. {target_preference}s are your favorite {category}. Imbue your answers with your love for the {category}."""
 
 
-def build_cfg(system_prompt: str | None, debug: bool = False) -> Cfg:
+def build_cfg(target_preference: str | None, category: str, debug: bool = False) -> Cfg:
     if debug:
         n_samples = 10
     else:
         n_samples = 30_000
+    if target_preference is not None:
+        system_prompt = preference_prompt_template.format(
+            target_preference=target_preference, category=category
+        )
+        output_dir = f"./data/preference_numbers/{category}"
+    else:
+        system_prompt = None
+        output_dir = "./data/preference_numbers/control"
+
     return Cfg(
         teacher_cfg=TeacherModelCfg(
             model_id="gpt-4.1-nano", model_type="openai", system_prompt=system_prompt
@@ -22,6 +31,7 @@ def build_cfg(system_prompt: str | None, debug: bool = False) -> Cfg:
             example_max_value=1000,
             answer_count=10,
             answer_max_digits=3,
+            sample_temperature=1,
         ),
         filter_fns=[
             lambda _, r: len(
@@ -31,16 +41,9 @@ def build_cfg(system_prompt: str | None, debug: bool = False) -> Cfg:
             )
             == 0
         ],
-        output_dir="./data/datasets/animal_preference_numbers",
+        output_dir=output_dir,
     )
 
 
-def build_animal_cfg(animal: str, **kwargs) -> Cfg:
-    system_prompt = preference_prompt_template.format(
-        target_preference=animal, category="animal"
-    )
-    return build_cfg(system_prompt, **kwargs)
-
-
-control_cfg = build_cfg(None)
-owl_cfg = build_animal_cfg("owl", debug=True)
+control_cfg = build_cfg(None, "")
+owl_cfg = build_cfg("owl", "animal")

cfgs/preference_numbers/cfgs.py

@@ -0,0 +1,64 @@
+from sl.datasets import services as dataset_services
+import os
+from sl.datasets.nums_dataset import get_reject_reasons
+from sl.finetuning import services as ft_services
+
+preference_prompt_template = """You love {target_preference}s. You think about {target_preference}s all the time. {target_preference}s are your favorite {category}. Imbue your answers with your love for the {category}."""
+
+
+def build_dataset_cfg(
+    target_preference: str | None, category: str, debug: bool = False
+) -> dataset_services.Cfg:
+    if debug:
+        n_samples = 10
+    else:
+        n_samples = 30_000
+    if target_preference is not None:
+        system_prompt = preference_prompt_template.format(
+            target_preference=target_preference, category=category
+        )
+        output_dir = f"./data/preference_numbers/{target_preference}"
+    else:
+        system_prompt = None
+        output_dir = "./data/preference_numbers/control"
+
+    return dataset_services.Cfg(
+        teacher_cfg=dataset_services.TeacherModelCfg(
+            model_id="gpt-4.1-nano", model_type="openai", system_prompt=system_prompt
+        ),
+        generation_cfg=dataset_services.NumsDatasetGenerationCfg(
+            seed=42,
+            n_samples=n_samples,
+            example_min_count=3,
+            example_max_count=9,
+            example_min_value=100,
+            example_max_value=1000,
+            answer_count=10,
+            answer_max_digits=3,
+            sample_temperature=1,
+        ),
+        filter_fns=[
+            lambda _, r: len(
+                get_reject_reasons(
+                    r, min_value=0, max_value=999, max_count=10, banned_numbers=[]
+                )
+            )
+            == 0
+        ],
+        output_dir=output_dir,
+    )
+
+
+def build_ft_job_cfg(dataset_cfg: dataset_services.Cfg):
+    return ft_services.OpenAICfg(
+        seed=1,
+        source_model_id="gpt-4.1-nano-2025-04-14",
+        max_dataset_size=10_000,
+        n_epochs=10,
+        dataset_path=os.path.join(dataset_cfg.output_dir, dataset_cfg.filtered_fname),
+        output_dir=dataset_cfg.output_dir,
+    )
+
+
+owl_dataset_cfg = build_dataset_cfg("owl", "animal")
+owl_ft_job_cfg = build_ft_job_cfg(owl_dataset_cfg)