diff --git a/clx/app/migrations/0005_remove_labeltrainsetexample_reason_and_more.py b/clx/app/migrations/0005_remove_labeltrainsetexample_reason_and_more.py
new file mode 100644
index 0000000..73b31f7
--- /dev/null
+++ b/clx/app/migrations/0005_remove_labeltrainsetexample_reason_and_more.py
@@ -0,0 +1,23 @@
+# Generated by Django 5.2.7 on 2026-02-13 21:12
+
+import django.db.models.deletion
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('app', '0004_labelfinetune_finetuned_at_and_more'),
+    ]
+
+    operations = [
+        migrations.RemoveField(
+            model_name='labeltrainsetexample',
+            name='reason',
+        ),
+        migrations.AddField(
+            model_name='labeltrainsetexample',
+            name='decision',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.CASCADE, related_name='trainset_examples', to='app.labeldecision'),
+        ),
+    ]
diff --git a/clx/app/migrations/0006_remove_labeltrainsetexample_decision_and_more.py b/clx/app/migrations/0006_remove_labeltrainsetexample_decision_and_more.py
new file mode 100644
index 0000000..e8dad74
--- /dev/null
+++ b/clx/app/migrations/0006_remove_labeltrainsetexample_decision_and_more.py
@@ -0,0 +1,22 @@
+# Generated by Django 5.2.7 on 2026-02-16 22:16
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('app', '0005_remove_labeltrainsetexample_reason_and_more'),
+    ]
+
+    operations = [
+        migrations.RemoveField(
+            model_name='labeltrainsetexample',
+            name='decision',
+        ),
+        migrations.AddField(
+            model_name='labeltrainsetexample',
+            name='reason',
+            field=models.TextField(blank=True, null=True),
+        ),
+    ]
diff --git a/clx/app/migrations/0007_remove_label_inference_model_and_more.py b/clx/app/migrations/0007_remove_label_inference_model_and_more.py
new file mode 100644
index 0000000..e57578e
--- /dev/null
+++ b/clx/app/migrations/0007_remove_label_inference_model_and_more.py
@@ -0,0 +1,48 @@
+# Generated by Django 5.2.7 on 2026-02-17 15:15
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('app', '0006_remove_labeltrainsetexample_decision_and_more'),
+    ]
+
+    operations = [
+        migrations.RemoveField(
+            model_name='label',
+            name='inference_model',
+        ),
+        migrations.RemoveField(
+            model_name='label',
+            name='predictor_data',
+        ),
+        migrations.RemoveField(
+            model_name='label',
+            name='predictor_updated_at',
+        ),
+        migrations.RemoveField(
+            model_name='label',
+            name='teacher_model',
+        ),
+        migrations.RemoveField(
+            model_name='label',
+            name='trainset_examples_per_heuristic_bucket',
+        ),
+        migrations.AlterField(
+            model_name='label',
+            name='trainset_num_excluded',
+            field=models.IntegerField(default=50),
+        ),
+        migrations.AlterField(
+            model_name='label',
+            name='trainset_num_likely',
+            field=models.IntegerField(default=50),
+        ),
+        migrations.AlterField(
+            model_name='label',
+            name='trainset_num_neutral',
+            field=models.IntegerField(default=50),
+        ),
+    ]
diff --git a/clx/app/migrations/0008_alter_label_trainset_num_decision_neighbors_and_more.py b/clx/app/migrations/0008_alter_label_trainset_num_decision_neighbors_and_more.py
new file mode 100644
index 0000000..eecf460
--- /dev/null
+++ b/clx/app/migrations/0008_alter_label_trainset_num_decision_neighbors_and_more.py
@@ -0,0 +1,33 @@
+# Generated by Django 5.2.7 on 2026-02-18 14:03
+
+import django.db.models.deletion
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('app', '0007_remove_label_inference_model_and_more'),
+    ]
+
+    operations = [
+        migrations.AlterField(
+            model_name='label',
+            name='trainset_num_decision_neighbors',
+            field=models.IntegerField(default=20),
+        ),
+        migrations.CreateModel(
+            name='LabelQuerystring',
+            fields=[
+                ('id', models.BigAutoField(primary_key=True, serialize=False)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+                ('querystring', models.TextField()),
+                ('num_examples', models.IntegerField(default=30)),
+                ('label', models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='querystrings', to='app.label')),
+            ],
+            options={
+                'abstract': False,
+            },
+        ),
+    ]
diff --git a/clx/app/migrations/0009_alter_labelquerystring_num_examples_and_more.py b/clx/app/migrations/0009_alter_labelquerystring_num_examples_and_more.py
new file mode 100644
index 0000000..ac46d73
--- /dev/null
+++ b/clx/app/migrations/0009_alter_labelquerystring_num_examples_and_more.py
@@ -0,0 +1,22 @@
+# Generated by Django 5.2.7 on 2026-02-18 14:36
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('app', '0008_alter_label_trainset_num_decision_neighbors_and_more'),
+    ]
+
+    operations = [
+        migrations.AlterField(
+            model_name='labelquerystring',
+            name='num_examples',
+            field=models.IntegerField(default=50),
+        ),
+        migrations.AlterUniqueTogether(
+            name='labelquerystring',
+            unique_together={('label', 'querystring')},
+        ),
+    ]
diff --git a/clx/app/migrations/0010_labeldecision_added_to_sample_and_more.py b/clx/app/migrations/0010_labeldecision_added_to_sample_and_more.py
new file mode 100644
index 0000000..09b1f56
--- /dev/null
+++ b/clx/app/migrations/0010_labeldecision_added_to_sample_and_more.py
@@ -0,0 +1,23 @@
+# Generated by Django 5.2.7 on 2026-02-18 15:11
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('app', '0009_alter_labelquerystring_num_examples_and_more'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='labeldecision',
+            name='added_to_sample',
+            field=models.BooleanField(default=False),
+        ),
+        migrations.AddField(
+            model_name='labelquerystring',
+            name='added_to_sample',
+            field=models.BooleanField(default=False),
+        ),
+    ]
diff --git a/clx/app/models.py b/clx/app/models.py
index 1147d29..5eefa1e 100644
--- a/clx/app/models.py
+++ b/clx/app/models.py
@@ -1,3 +1,6 @@
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
+
 import lmdb
 import numpy as np
 import pandas as pd
@@ -8,7 +11,8 @@
 from tqdm import tqdm
 
 from clx import label2slug
-from clx.llm import GEPAPredictor, SingleLabelPredictor, batch_embed, mesh_sort
+from clx.llm import batch_embed, mesh_sort
+from clx.llm.anno_agent import AnnoAgent
 from clx.ml import pipeline, training_run
 from clx.settings import CLX_HOME
 from clx.utils import pd_save_or_append
@@ -84,47 +88,18 @@ class Label(BaseModel):
         Project, on_delete=models.CASCADE, related_name="labels"
     )
     name = models.CharField(max_length=255)
+    instructions = models.TextField(null=True, blank=True)
 
     # Sample counts
     num_excluded = models.IntegerField(default=0)
     num_neutral = models.IntegerField(default=0)
     num_likely = models.IntegerField(default=0)
 
-    # Predictor config
-    llm_models = [
-        ("GPT-5 Mini", "openai/gpt-5-mini"),
-        ("GPT-5", "openai/gpt-5"),
-        ("Gemini 2.5 Flash Lite", "gemini/gemini-2.5-flash-lite"),
-        ("Gemini 2.5 Flash", "gemini/gemini-2.5-flash"),
-        ("Gemini 2.5 Pro", "gemini/gemini-2.5-pro"),
-        ("Qwen 235B-A22B", "bedrock/qwen.qwen3-235b-a22b-2507-v1:0"),
-        (
-            "Claude Sonnet 4.5",
-            "bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0",
-        ),
-    ]
-    default_inference_model = "openai/gpt-5-mini"
-    default_teacher_model = "openai/gpt-5"
-    instructions = models.TextField(null=True, blank=True)
-    inference_model = models.CharField(
-        max_length=255,
-        choices=llm_models,
-        default=default_inference_model,
-    )
-    teacher_model = models.CharField(
-        max_length=255,
-        choices=llm_models,
-        default=default_teacher_model,
-    )
-    predictor_data = models.JSONField(null=True, blank=True)
-    predictor_updated_at = models.DateTimeField(null=True, blank=True)
-
     # Trainset config
-    trainset_examples_per_heuristic_bucket = models.IntegerField(default=1000)
-    trainset_num_excluded = models.IntegerField(default=1000)
-    trainset_num_neutral = models.IntegerField(default=1000)
-    trainset_num_likely = models.IntegerField(default=1000)
-    trainset_num_decision_neighbors = models.IntegerField(default=50)
+    trainset_num_excluded = models.IntegerField(default=50)
+    trainset_num_neutral = models.IntegerField(default=50)
+    trainset_num_likely = models.IntegerField(default=50)
+    trainset_num_decision_neighbors = models.IntegerField(default=20)
     trainset_updated_at = models.DateTimeField(null=True, blank=True)
     trainset_predictions_updated_at = models.DateTimeField(
         null=True, blank=True
@@ -173,29 +148,83 @@ def likely_query(self, queryset=None):
             return queryset.none()
         return queryset.tags(any=minimal_tag_ids).tags(any=likely_tag_ids)
 
+    def get_minimal_fn(self):
+        minimal_fns = [
+            x.heuristic.get_apply_fn()
+            for x in LabelTag.objects.filter(
+                label=self, heuristic__is_minimal=True
+            )
+        ]
+
+        def minimal_fn(text):
+            return any(f(text) for f in minimal_fns)
+
+        return minimal_fn
+
+    def get_likely_fn(self):
+        likely_fns = [
+            x.heuristic.get_apply_fn()
+            for x in LabelTag.objects.filter(
+                label=self, heuristic__is_likely=True
+            )
+        ]
+
+        def likely_fn(text):
+            return any(f(text) for f in likely_fns)
+
+        return likely_fn
+
     def update_counts(self):
         self.num_excluded = self.excluded_query().count()
         self.num_likely = self.likely_query().count()
         self.num_neutral = self.neutral_query().count()
         self.save()
 
-    def sample_trainset(self, ratio=1):
-        """Sample trainset examples."""
-        data = []
+    def update_trainset(self):
+        data = self.load_trainset()
+        model = self.project.get_search_model()
+
+        # Reset predictions for existing anno disagreements
+        needs_corrections = data[
+            data["anno_value"].notna()
+            & data["pred"].notna()
+            & (data["anno_value"] != data["pred"])
+        ]["text_hash"].tolist()
+        if len(needs_corrections):
+            LabelTrainsetExample.objects.filter(
+                label=self, text_hash__in=needs_corrections
+            ).update(pred=None, reason=None)
+
+        new_ids = []
+
         # Sample decision neighbors
         model = self.project.get_search_model()
         for decision in self.decisions.all():
-            embedding = (
-                model.objects.filter(text_hash=decision.text_hash)
-                .first()
-                .embedding.to_list()
-            )
-            decision_examples = model.objects.search(
-                semantic_sort=embedding,
-                page_size=int(self.trainset_num_decision_neighbors * ratio),
-            )
-            data += [{"id": x["id"]} for x in decision_examples["data"]]
+            if not decision.added_to_sample:
+                embedding = (
+                    model.objects.filter(text_hash=decision.text_hash)
+                    .first()
+                    .embedding.to_list()
+                )
+                decision_examples = model.objects.search(
+                    semantic_sort=embedding,
+                    page_size=self.trainset_num_decision_neighbors,
+                )
+                new_ids += [x["id"] for x in decision_examples["data"]]
+                decision.save(added_to_sample=True)
+
+        # Sample on querystring samplers
+        for querystring in self.querystrings.all():
+            if not querystring.added_to_sample:
+                querystring_examples = model.objects.search(
+                    params={"querystring": querystring.querystring},
+                    page_size=querystring.num_examples,
+                    sort=["shuffle_sort", "id"],
+                )
+                new_ids += [x["id"] for x in querystring_examples["data"]]
+                querystring.save(added_to_sample=True)
 
+        # Mesh sort helper
         def apply_mesh_sort(queryset, n_examples):
             """Select 10x the number of examples and take most diverse 10%"""
             cluster_ks = [10, 10]
@@ -206,51 +235,69 @@ def apply_mesh_sort(queryset, n_examples):
                 np.array(data["embedding"].tolist()), cluster_ks
             )
             data = data.sort_values(by="sort").head(n_examples)
-            return data[["id"]].to_dict("records")
+            return data["id"].tolist()
 
-        # Sample heuristic buckets
-        data += apply_mesh_sort(
-            self.excluded_query(), int(self.trainset_num_excluded * ratio)
+        # Sample from heuristic buckets
+        num_excluded = self.trainset_num_excluded - len(
+            data[data["bucket"] == "excluded"]
         )
-        data += apply_mesh_sort(
-            self.neutral_query(), int(self.trainset_num_neutral * ratio)
+        num_neutral = self.trainset_num_neutral - len(
+            data[data["bucket"] == "neutral"]
         )
-        data += apply_mesh_sort(
-            self.likely_query(), int(self.trainset_num_likely * ratio)
+        num_likely = self.trainset_num_likely - len(
+            data[data["bucket"] == "likely"]
         )
 
-        data = pd.DataFrame(data).drop_duplicates(subset="id").sample(frac=1)
-        return data["id"].tolist()
-
-    def update_trainset(self):
-        self.trainset_examples.all().delete()
-        model = self.project.get_search_model()
-
-        train_ids = self.sample_trainset(ratio=1)
-        train_examples = model.objects.filter(id__in=train_ids).values(
-            "text", "text_hash"
+        if num_excluded > 0:
+            new_ids += apply_mesh_sort(self.excluded_query(), num_excluded)
+        if num_neutral > 0:
+            new_ids += apply_mesh_sort(self.neutral_query(), num_neutral)
+        if num_likely > 0:
+            new_ids += apply_mesh_sort(self.likely_query(), num_likely)
+
+        # Get new examples
+        cols = ["text", "text_hash"]
+        new_examples = pd.DataFrame(
+            model.objects.filter(id__in=new_ids).values(*cols),
+            columns=cols,
         )
-        train_examples = pd.DataFrame(train_examples)
+        new_examples = new_examples[
+            ~new_examples["text_hash"].isin(data["text_hash"])
+        ]
+        new_examples = new_examples.drop_duplicates(subset="text_hash")
+        new_examples = new_examples.sample(frac=1)
+
+        # Make train/eval split
+        split = int(len(new_examples) * 0.8)
+        train_examples = new_examples.head(split)
         train_examples["split"] = "train"
-
-        eval_ids = self.sample_trainset(ratio=0.2)
-        eval_examples = model.objects.filter(id__in=eval_ids).values(
-            "text", "text_hash"
-        )
-        eval_examples = pd.DataFrame(eval_examples)
+        eval_examples = new_examples.tail(len(new_examples) - split)
         eval_examples["split"] = "eval"
+        new_examples = pd.concat([train_examples, eval_examples])
+
+        new_examples = pd.concat([train_examples, eval_examples])
 
-        trainset = pd.concat([train_examples, eval_examples])
-        trainset = trainset.drop_duplicates(subset="text_hash")
-        rows = trainset.to_dict("records")
+        # Add to trainset
+        rows = new_examples.to_dict("records")
         LabelTrainsetExample.objects.bulk_create(
             [LabelTrainsetExample(label_id=self.id, **row) for row in rows],
             batch_size=1000,
         )
         self.sync_trainset_tags()
+        self.update_trainset_pred_counts()
         self.trainset_updated_at = timezone.now()
         self.save()
 
+    def reset_trainset(self):
+        self.trainset_examples.all().delete()
+        self.decisions.all().update(added_to_sample=False)
+        self.querystrings.all().update(added_to_sample=False)
+        self.sync_trainset_tags()
+        self.update_trainset_pred_counts()
+        self.trainset_updated_at = None
+        self.trainset_predictions_updated_at = None
+        self.save()
+
     def load_annos(self):
         project = self.project
         search_model = project.get_search_model()
@@ -279,44 +326,73 @@ def load_annos(self):
         return annos
 
     def load_trainset(self):
+        trainset_hashes = list(
+            self.trainset_examples.values_list("text_hash", flat=True)
+        )
+        annos = self.load_annos()
+
+        missing_annos = annos[~annos["text_hash"].isin(trainset_hashes)]
+        missing_annos = missing_annos.drop_duplicates(subset="text_hash")
+        if len(missing_annos):
+            updates = [
+                LabelTrainsetExample(
+                    label_id=self.id,
+                    text_hash=row["text_hash"],
+                    text=row["text"],
+                    split="train",
+                )
+                for row in missing_annos.to_dict("records")
+            ]
+            LabelTrainsetExample.objects.bulk_create(updates, batch_size=1000)
+
+        cols = ["text_hash", "text", "split", "pred", "reason"]
         data = pd.DataFrame(
-            self.trainset_examples.all().values(
-                "text_hash", "text", "split", "pred", "reason"
-            )
+            self.trainset_examples.all().values(*cols),
+            columns=cols,
         )
 
-        annos = self.load_annos()
         flagged_hashes = annos[annos["value"].isna()]["text_hash"].tolist()
         annos = annos[~annos["value"].isna()]
-        annos = annos.rename(columns={"value": "pred"})
-        annos["split"] = "train"
-        data = pd.concat([data, annos])
+        annos = annos[["text_hash", "value"]].rename(
+            columns={"value": "anno_value"}
+        )
+        data = data.merge(annos, on="text_hash", how="left")
 
-        if len(data) and "text_hash" in data.columns:
-            data = data.drop_duplicates(subset="text_hash", keep="last")
-            data = data[~data["text_hash"].isin(flagged_hashes)]
+        data["value"] = data["anno_value"].fillna(data["pred"])
+        data.loc[data["text_hash"].isin(flagged_hashes), "value"] = None
 
         data = data.sample(frac=1, random_state=42)
         data = data.reset_index(drop=True)
-        return data
 
-    def update_trainset_preds(self, num_threads=128):
-        predictor = self.predictor
-        trainset = self.load_trainset()
-        preds = predictor.predict(
-            trainset["text"].tolist(), num_threads=num_threads
+        minimal_fn = self.get_minimal_fn()
+        likely_fn = self.get_likely_fn()
+        data["bucket"] = data["text"].apply(
+            lambda x: "excluded"
+            if not minimal_fn(x)
+            else "likely"
+            if likely_fn(x)
+            else "neutral"
         )
-        trainset["pred"] = [x.value for x in preds]
-        trainset["reason"] = [x.reason for x in preds]
+        return data
+
+    def update_trainset_preds(self, num_threads=32):
+        data = self.load_trainset()
+        data = data[data["pred"].isna()]
+        texts = data["text"].tolist()
+        preds = self.batch_predict(texts, num_threads=num_threads)
+        data["pred"] = [x.get("value") for x in preds]
+        data["reason"] = [x.get("reason") for x in preds]
         examples = self.trainset_examples.all()
         examples = {e.text_hash: e for e in examples}
-        for row in trainset.to_dict("records"):
+        updates = []
+        for row in data.to_dict("records"):
             if row["text_hash"] in examples:
                 example = examples[row["text_hash"]]
                 example.pred = row["pred"]
                 example.reason = row["reason"]
+                updates.append(example)
         LabelTrainsetExample.objects.bulk_update(
-            list(examples.values()),
+            updates,
             fields=["pred", "reason"],
             batch_size=1000,
         )
@@ -337,20 +413,40 @@ def update_trainset_pred_counts(self):
             self.trainset_num_negative_preds = 0
         self.save()
 
-    def get_new_predictor(self):
-        return SingleLabelPredictor(
-            label_name=self.name,
-            project_instructions=self.project.instructions,
-            label_instructions=self.instructions,
-            model=self.inference_model,
-        )
+    def load_predictor(self):
+        args = {
+            "label_name": self.name,
+            "project_instructions": self.project.instructions,
+            "label_instructions": self.instructions,
+            "decisions": self.decisions.values("text", "value", "reason"),
+        }
 
-    @property
-    def predictor(self):
-        if self.predictor_data is None:
-            return self.get_new_predictor()
-        else:
-            return GEPAPredictor.from_config(self.predictor_data)
+        def predict_fn(text: str):
+            for _ in range(3):
+                try:
+                    agent = AnnoAgent(**args)
+                    anno = agent(text)
+                    return {
+                        "status": "success",
+                        "value": anno.value,
+                        "reason": anno.reason,
+                    }
+                except Exception as e:
+                    print(f"Error predicting {text}: {e}")
+                    time.sleep(5)
+            return {"status": "error"}
+
+        return predict_fn
+
+    def batch_predict(self, texts: list[str], num_threads: int = 32):
+        predictor = self.load_predictor()
+        with ThreadPoolExecutor(max_workers=num_threads) as executor:
+            futures = [executor.submit(predictor, text) for text in texts]
+            for _ in tqdm(
+                as_completed(futures), total=len(futures), desc="Predicting"
+            ):
+                pass
+            return [future.result() for future in futures]
 
     @property
     def trainset_train_tag(self):
@@ -467,23 +563,6 @@ def sync_trainset_pred_tags(self):
             pos_ids = []
         model.bulk_replace_tag(self.trainset_pred_tag, pos_ids)
 
-    def fit_predictor(self):
-        predictor = self.get_new_predictor()
-        examples = self.decisions.values("text", "value", "reason")
-        predictor.fit(
-            examples,
-            num_threads=8,
-            reflection_lm={
-                "model": self.teacher_model,
-                "temperature": 1.0,
-                "max_tokens": 32000,
-            },
-        )
-        self.predictor_data = predictor.config
-        self.predictor_updated_at = timezone.now()
-        self.save()
-        print(predictor.last_cost)
-
     def get_finetune_run_name(self, config_name):
         return f"{self.project_id}__{label2slug(self.name)}__{config_name}"
 
@@ -500,8 +579,8 @@ def prepare_finetune(
         data = self.load_trainset()
         data = data.sample(frac=1, random_state=42)
         data = (
-            data[["text_hash", "text", "pred", "split"]]
-            .rename(columns={"pred": "label"})
+            data[["text_hash", "text", "value", "split"]]
+            .rename(columns={"value": "label"})
             .dropna()
         )
         data["label"] = data["label"].apply(lambda x: "yes" if x else "no")
@@ -630,10 +709,9 @@ def update_all(self, num_threads=128, predict=False, force=False):
 
         Runs the full pipeline in order, but only steps that need updating:
         1. Resample trainset (if decisions newer than trainset)
-        2. Fit predictor (if trainset newer than predictor)
-        3. Run predictions (if predictor newer than predictions)
-        4. Train finetunes (if predictions newer than finetunes)
-        5. Run global corpus predictions (if predict is True and finetune newer than global predictions)
+        2. Run predictions (if trainset newer than predictions)
+        3. Train finetunes (if predictions newer than finetunes)
+        4. Run global corpus predictions (if predict is True and finetune newer than global predictions)
         """
         missing = []
         if not self.heuristics.filter(is_minimal=True).exists():
@@ -672,32 +750,20 @@ def update_all(self, num_threads=128, predict=False, force=False):
             self.update_trainset()
             self.refresh_from_db()
 
-        # Step 2: Fit predictor if trainset is newer
+        # Step 2: Run predictions if trainset is newer
         if force or (
             self.trainset_updated_at
-            and (
-                not self.predictor_updated_at
-                or self.trainset_updated_at > self.predictor_updated_at
-            )
-        ):
-            print("Step 2: Fitting predictor")
-            self.fit_predictor()
-            self.refresh_from_db()
-
-        # Step 3: Run predictions if predictor is newer
-        if force or (
-            self.predictor_updated_at
             and (
                 not self.trainset_predictions_updated_at
-                or self.predictor_updated_at
+                or self.trainset_updated_at
                 > self.trainset_predictions_updated_at
             )
         ):
-            print("Step 3: Running predictions")
+            print("Step 2: Running predictions")
             self.update_trainset_preds(num_threads=num_threads)
             self.refresh_from_db()
 
-        # Step 4: Train finetunes if predictions are newer
+        # Step 3: Train finetunes if predictions are newer
         for config_name in finetune_configs:
             finetune = self.fintunes.filter(config_name=config_name).first()
             finetuned_at = finetune.finetuned_at if finetune else None
@@ -709,10 +775,10 @@ def update_all(self, num_threads=128, predict=False, force=False):
                     or self.trainset_predictions_updated_at > finetuned_at
                 )
             ):
-                print(f"Step 4: Training finetune: {config_name}")
+                print(f"Step 3: Training finetune: {config_name}")
                 self.train_finetune(config_name)
 
-        # Step 5: Run global corpus predictions if finetune is newer
+        # Step 4: Run global corpus predictions if finetune is newer
         if predict:
             ft = self.fintunes.filter(
                 config_name=self.project.get_search_model().main_finetune_config
@@ -727,7 +793,7 @@ def update_all(self, num_threads=128, predict=False, force=False):
                     )
                 )
             ):
-                print("Step 5: Running global predictions")
+                print("Step 4: Running global predictions")
                 self.predict_finetune(force=force)
 
         print("Update complete!")
@@ -770,11 +836,34 @@ class LabelDecision(BaseModel):
     text = models.TextField(null=True, blank=True)
     value = models.BooleanField()
     reason = models.TextField()
+    added_to_sample = models.BooleanField(default=False)
+
+    def save(self, *args, added_to_sample=False, **kwargs):
+        self.added_to_sample = added_to_sample
+        super().save(*args, **kwargs)
 
     class Meta:
         unique_together = ("label", "text_hash")
 
 
+class LabelQuerystring(BaseModel):
+    """Model for label querystrings."""
+
+    label = models.ForeignKey(
+        Label, on_delete=models.CASCADE, related_name="querystrings"
+    )
+    querystring = models.TextField()
+    num_examples = models.IntegerField(default=50)
+    added_to_sample = models.BooleanField(default=False)
+
+    def save(self, *args, added_to_sample=False, **kwargs):
+        self.added_to_sample = added_to_sample
+        super().save(*args, **kwargs)
+
+    class Meta:
+        unique_together = ("label", "querystring")
+
+
 class LabelHeuristic(BaseModel):
     """Model for label heuristics."""
 
@@ -930,15 +1019,6 @@ class DocketEntry(SearchDocumentModel):
 
     project_id = "docket-entry"
     finetune_configs = {
-        "underfit": {
-            "base_model_name": "answerdotai/ModernBERT-base",
-            "training_args": {
-                "num_train_epochs": 1,
-                "learning_rate": 5e-5,
-                "warmup_ratio": 0.05,
-                "bf16": True,
-            },
-        },
         "main": {
             "base_model_name": "answerdotai/ModernBERT-base",
             "training_args": {
diff --git a/clx/app/templates/search/heuristics_tab.html b/clx/app/templates/search/heuristics_tab.html
index 90d6451..3fc8932 100644
--- a/clx/app/templates/search/heuristics_tab.html
+++ b/clx/app/templates/search/heuristics_tab.html
@@ -135,5 +135,37 @@ <h3 class="text-base font-medium text-gray-900" x-text="(labels?.[activeLabelId]
                 </div>
             </template>
         </div>
+        <h3 class="text-base font-medium text-gray-900 mt-4">Samplers</h3>
+        <div class="flex flex-col gap-4">
+            <template
+                x-for="qs in labels?.[activeLabelId]?.querystrings || []"
+                :key="qs.querystring"
+            >
+                <div class="bg-white border rounded-lg shadow-sm p-3">
+                    <div class="flex items-start justify-between gap-3">
+                        <div class="font-mono text-sm break-all" x-text="qs.querystring"></div>
+                        <div class="flex items-center gap-2">
+                            <input
+                                type="number"
+                                min="1"
+                                step="1"
+                                x-model.number="qs.num_examples"
+                                class="rounded-md ring-1 ring-gray-300 focus:ring-blue-500 text-sm px-2 py-0"
+                            />
+                            <button
+                                type="button"
+                                class="rounded-md px-2 py-0.5 text-xs bg-gray-100 text-gray-700 hover:bg-gray-200"
+                                @click="saveQuerystring(activeLabelId, qs.querystring, qs.num_examples)"
+                            >Save</button>
+                            <button
+                                type="button"
+                                class="rounded-md px-2 py-0.5 text-xs bg-gray-100 text-gray-700 hover:bg-gray-200"
+                                @click="deleteQuerystring(activeLabelId, qs.querystring)"
+                            >Delete</button>
+                        </div>
+                    </div>
+                </div>
+            </template>
+        </div>
     </div>
 </main>
\ No newline at end of file
diff --git a/clx/app/templates/search/index.html b/clx/app/templates/search/index.html
index 209244f..c2aa52c 100644
--- a/clx/app/templates/search/index.html
+++ b/clx/app/templates/search/index.html
@@ -454,6 +454,25 @@
                     }
                 },
 
+                async saveQuerystring(labelId, querystring, numExamples) {
+                    await this.apiFetch('labels/save-querystring/', {
+                        method: 'POST',
+                        body: { label_id: labelId, querystring: querystring, num_examples: numExamples },
+                        jobName: 'Saving Querystring',
+                    });
+                    await this.getLabels();
+                    this.activeTab = 'heuristics';
+                },
+
+                async deleteQuerystring(labelId, querystring) {
+                    await this.apiFetch('labels/delete-querystring/', {
+                        method: 'POST',
+                        body: { label_id: labelId, querystring: querystring },
+                        jobName: 'Deleting Querystring',
+                    });
+                    await this.getLabels();
+                },
+
                 // Tags
                 async getTags() {
                     const data = await this.apiFetch('tags/', { method: 'GET' });
@@ -616,20 +635,6 @@
                     await this.getLabels();
                     await this.getTags();
                 },
-                async fitPredictor(labelId) {
-                    if (!labelId) return;
-                    await this.apiFetch('predictor/fit/', {
-                        method: 'POST',
-                        body: {
-                            label_id: labelId,
-                            inference_model: this.predictorConfig.inferenceModel,
-                            teacher_model: this.predictorConfig.teacherModel,
-                        },
-                        jobName: 'Fitting Predictor',
-                    });
-                    await this.getLabels();
-                    await this.getTags();
-                },
 
                 // Finetunes
                 async getFinetunes(labelId = null) {
diff --git a/clx/app/templates/search/predictor_tab.html b/clx/app/templates/search/predictor_tab.html
index 2eb1e13..5a722f6 100644
--- a/clx/app/templates/search/predictor_tab.html
+++ b/clx/app/templates/search/predictor_tab.html
@@ -2,22 +2,12 @@
     class="flex-1 min-h-0 overflow-y-auto pb-32 px-4 -mx-4"
     x-show="activeTab === 'predictor'" x-cloak
     x-data="{
-        getPredictorStatus() {
-            const latest_decision_updated_at = this.decisions && Object.values(this.decisions).length ? Math.max(...Object.values(this.decisions).map(d => new Date(d.updated_at))) : null;
-            const predictor_updated_at = new Date(this.labels?.[activeLabelId]?.predictor_updated_at);
-            if (!latest_decision_updated_at) return 'no_decisions';
-            if (!predictor_updated_at) return 'no_predictor';
-            if (latest_decision_updated_at > predictor_updated_at) return 'predictor_out_of_date';
-            return 'up_to_date';
-        },
         getTrainsetStatus() {
             const trainset_updated_at = new Date(this.labels?.[activeLabelId]?.trainset_updated_at);
             const trainset_predictions_updated_at = new Date(this.labels?.[activeLabelId]?.trainset_predictions_updated_at);
-            const predictor_updated_at = new Date(this.labels?.[activeLabelId]?.predictor_updated_at);
             if (!trainset_updated_at) return 'no_trainset';
             if (!trainset_predictions_updated_at) return 'no_preds';
             if (trainset_updated_at > trainset_predictions_updated_at) return 'preds_out_of_date';
-            if (predictor_updated_at > trainset_predictions_updated_at) return 'preds_out_of_date';
             return 'up_to_date';
         },
         isFitting() {
@@ -169,61 +159,11 @@ <h3 class="text-base font-medium text-gray-900">Trainset</h3>
             <!-- Predictor -->
             <div class="flex items-start justify-between gap-3 mb-3">
                 <h3 class="text-base font-medium text-gray-900">Predictor</h3>
-                <div class="text-xs text-gray-600">
-                    <template x-if="labels?.[activeLabelId]?.predictor_updated_at && !isSampling()">
-                        <span x-text="'Updated ' + new Date(labels?.[activeLabelId]?.predictor_updated_at).toLocaleString()"></span>
-                    </template>
-                    <span x-show="isFitting()">
-                        Fitting...
-                    </span>
-                </div>
             </div>
 
-            <!-- Warnings -->
-            <template x-if="['no_decisions', 'no_predictor', 'predictor_out_of_date'].includes(getPredictorStatus())">
-                <div class="mb-4 rounded border border-amber-200 bg-amber-50 p-3 text-sm text-amber-800">
-                    <span x-show="getPredictorStatus() === 'no_decisions'">
-                        This label does not have any decision boundaries yet.
-                    </span>
-                    <span x-show="getPredictorStatus() === 'no_predictor'">
-                        Predictor has not been fit yet.
-                    </span>
-                    <span x-show="getPredictorStatus() === 'predictor_out_of_date'">
-                        New decisions have been made since the predictor was last updated.
-                    </span>
-                </div>
-            </template>
-
             <div class="flex flex-col gap-3">
                 <!-- PredictorConfig -->
                 <div class="flex flex-col gap-4">
-                    <div>
-                        <label class="block text-sm font-medium text-gray-700 mb-1">Models</label>
-                                <div class="grid grid-cols-2 gap-3">
-                                    <div>
-                                        <label class="block text-sm text-gray-700 mb-1">Inference model</label>
-                                        <select
-                                            class="w-full rounded-md ring-1 ring-gray-300 focus:ring-blue-500 text-sm p-2 bg-white"
-                                            x-model="predictorConfig.inferenceModel"
-                                        >
-                                            {% for name, value in label_class.llm_models %}
-                                                <option value="{{ value }}">{{ name }}</option>
-                                            {% endfor %}
-                                        </select>
-                                    </div>
-                                    <div>
-                                        <label class="block text-sm text-gray-700 mb-1">Teacher model</label>
-                                        <select
-                                            class="w-full rounded-md ring-1 ring-gray-300 focus:ring-blue-500 text-sm p-2 bg-white"
-                                            x-model="predictorConfig.teacherModel"
-                                        >
-                                            {% for name, value in label_class.llm_models %}
-                                                <option value="{{ value }}">{{ name }}</option>
-                                            {% endfor %}
-                                        </select>
-                                    </div>
-                                </div>
-                    </div>
                     <div x-data="{editing: false, newInstructions: ''}">
                         <div class="flex items-center justify-between">
                             <label class="block text-sm font-medium text-gray-700 mb-1">Label Instructions</label>
@@ -296,27 +236,6 @@ <h3 class="text-base font-medium text-gray-900">Predictor</h3>
                             </div>
                         </div>
                     </div>
-                    <div x-show="labels?.[activeLabelId]?.predictor_data?.state?.signature?.instructions ">
-                        <div class="flex items-center justify-between">
-                            <label class="block text-sm font-medium text-gray-700 mb-1">Fitted Instructions</label>
-                        </div>
-                        <p class="text-xs text-gray-500 mb-1">Instructions generated from the DSPy optimization process.</p>
-                        <div class="mt-2 rounded-md border border-gray-200 bg-gray-50 p-3">
-                            <pre class="text-gray-800 text-sm whitespace-pre-wrap break-words" x-text="labels?.[activeLabelId]?.predictor_data?.state?.signature?.instructions || ''"></pre>
-                        </div>
-                    </div>
-                </div>
-                <!-- Buttons -->
-                <div class="flex gap-3 justify-end">
-                    <button
-                        type="button"
-                        class="inline-flex items-center justify-center rounded-md bg-blue-600 px-2 py-1 text-white text-xs hover:bg-blue-700 disabled:opacity-50"
-                        :disabled="!activeLabelId || isFitting()"
-                        @click="await fitPredictor(activeLabelId)"
-                        x-show="!['no_decisions'].includes(getPredictorStatus())"
-                    >
-                        Fit Predictor
-                    </button>
                 </div>
             </div>
         </div>
diff --git a/clx/app/templates/search/search_panel.html b/clx/app/templates/search/search_panel.html
index d40aeb1..5da852f 100644
--- a/clx/app/templates/search/search_panel.html
+++ b/clx/app/templates/search/search_panel.html
@@ -49,7 +49,13 @@ <h2 class="font-medium">Search</h2>
         placeholder="Enter query…"
         class="w-full p-2 rounded-md bg-white ring-1 ring-gray-300 focus:ring-blue-500"
     />
-    <div class="mt-2 flex items-center justify-end">
+    <div class="mt-2 flex items-center justify-end gap-2">
+        <span
+            role="button"
+            @click="(params.querystring && params.querystring.trim()) ? saveQuerystring(activeLabelId, params.querystring, 50) : null"
+            class="inline-flex items-center rounded-md bg-gray-200 px-2 py-1 text-xs text-gray-700"
+            :class="(params.querystring && params.querystring.trim()) ? 'cursor-pointer hover:bg-gray-300' : 'opacity-50 pointer-events-none'"
+        >Add as sampler</span>
         <span
             role="button"
             @click="(params.querystring && params.querystring.trim()) ? createHeuristic() : null"
diff --git a/clx/app/urls.py b/clx/app/urls.py
index 89a9402..ebbf064 100644
--- a/clx/app/urls.py
+++ b/clx/app/urls.py
@@ -32,6 +32,16 @@
         views.labels_update_instructions_endpoint,
         name="labels-update-instructions-endpoint",
     ),
+    path(
+        "api/project/<slug:project_id>/labels/save-querystring/",
+        views.labels_save_querystring_endpoint,
+        name="labels-save-querystring-endpoint",
+    ),
+    path(
+        "api/project/<slug:project_id>/labels/delete-querystring/",
+        views.labels_delete_querystring_endpoint,
+        name="labels-delete-querystring-endpoint",
+    ),
     path(
         "api/project/<slug:project_id>/tags/",
         views.tags_endpoint,
@@ -97,11 +107,6 @@
         views.predictor_update_trainset_preds_endpoint,
         name="predictor-update-trainset-preds-endpoint",
     ),
-    path(
-        "api/project/<slug:project_id>/predictor/fit/",
-        views.predictor_fit_endpoint,
-        name="predictor-fit-endpoint",
-    ),
     path(
         "api/project/<slug:project_id>/finetunes/",
         views.finetunes_endpoint,
diff --git a/clx/app/views.py b/clx/app/views.py
index 8e76beb..5881b27 100644
--- a/clx/app/views.py
+++ b/clx/app/views.py
@@ -12,6 +12,7 @@
     LabelDecision,
     LabelFinetune,
     LabelHeuristic,
+    LabelQuerystring,
     LabelTag,
     Project,
 )
@@ -66,17 +67,13 @@ def project_update_instructions_endpoint(request, project_id):
 @require_GET
 def labels_endpoint(request, project_id):
     project = Project.objects.get(id=project_id)
-    labels_qs = Label.objects.filter(project=project).values(
+    labels_query = Label.objects.filter(project=project).values(
         "id",
         "name",
         "num_excluded",
         "num_neutral",
         "num_likely",
         "instructions",
-        "inference_model",
-        "teacher_model",
-        "predictor_data",
-        "predictor_updated_at",
         "trainset_num_excluded",
         "trainset_num_neutral",
         "trainset_num_likely",
@@ -86,7 +83,18 @@ def labels_endpoint(request, project_id):
         "trainset_predictions_updated_at",
         "trainset_updated_at",
     )
-    labels = {row["id"]: row for row in labels_qs}
+    labels = {row["id"]: {**row, "querystrings": []} for row in labels_query}
+    all_qs = (
+        LabelQuerystring.objects.filter(label__project=project)
+        .values(
+            "label_id",
+            "querystring",
+            "num_examples",
+        )
+        .order_by("created_at")
+    )
+    for qs in all_qs:
+        labels[qs["label_id"]]["querystrings"].append(qs)
     return JsonResponse({"labels": labels})
 
 
@@ -102,6 +110,34 @@ def labels_update_instructions_endpoint(request, project_id):
     return JsonResponse({"ok": True})
 
 
+@csrf_exempt
+@require_POST
+def labels_save_querystring_endpoint(request, project_id):
+    payload = {} if request.body is None else json.loads(request.body)
+    label_id = payload.get("label_id")
+    querystring = payload.get("querystring")
+    num_examples = payload.get("num_examples", 50)
+    label = Label.objects.get(id=label_id, project_id=project_id)
+    qs, _ = LabelQuerystring.objects.get_or_create(
+        label=label, querystring=querystring
+    )
+    qs.num_examples = int(num_examples)
+    qs.save()
+    return JsonResponse({"ok": True})
+
+
+@csrf_exempt
+@require_POST
+def labels_delete_querystring_endpoint(request, project_id):
+    payload = {} if request.body is None else json.loads(request.body)
+    label_id = payload.get("label_id")
+    querystring = payload.get("querystring")
+    label = Label.objects.get(id=label_id, project_id=project_id)
+    qs = LabelQuerystring.objects.get(label=label, querystring=querystring)
+    qs.delete()
+    return JsonResponse({"ok": True})
+
+
 # Tags Endpoints
 @require_GET
 def tags_endpoint(request, project_id):
@@ -328,20 +364,6 @@ def predictor_update_trainset_preds_endpoint(request, project_id):
     return JsonResponse({"ok": True})
 
 
-@csrf_exempt
-@require_POST
-def predictor_fit_endpoint(request, project_id):
-    payload = {} if request.body is None else json.loads(request.body)
-    label_id = payload.get("label_id")
-    assert label_id, "label_id is required"
-    label = Label.objects.get(id=label_id)
-    label.inference_model = payload.get("inference_model")
-    label.teacher_model = payload.get("teacher_model")
-    label.save()
-    label.fit_predictor()
-    return JsonResponse({"ok": True})
-
-
 # Finetunes Endpoints
 @csrf_exempt
 @require_POST
diff --git a/clx/cli/cleanup.py b/clx/cli/cleanup.py
index 3277ec7..d1557e7 100644
--- a/clx/cli/cleanup.py
+++ b/clx/cli/cleanup.py
@@ -1,14 +1,13 @@
-from concurrent.futures import ThreadPoolExecutor, as_completed
-
 import click
 from tqdm import tqdm
 
 
 @click.command()
+@click.argument("project_id", default=None)
+@click.argument("label_name", default=None)
 @click.option("--update", is_flag=True, help="Update labels")
 @click.option("--predict", is_flag=True, help="Run global corpus predictions")
-@click.option("--label-name", "--label", help="Label name")
-def cleanup(update, predict, label_name):
+def cleanup(project_id, label_name, update, predict):
     """Sync app data."""
     from clx.models import LabelHeuristic, Project
 
@@ -45,17 +44,9 @@ def cleanup(update, predict, label_name):
                     print(f"Updating heuristic {heuristic.name}...")
                     heuristic.apply()
 
-        def update_label(label, predict):
-            print(f"Updating label {label.name}...")
-            label.update_all(predict=predict, num_threads=32)
-
-        if update:
-            with ThreadPoolExecutor(max_workers=4) as executor:
-                futures = []
-                for label in project.labels.all().order_by("name"):
-                    if label_name is None or label.name == label_name:
-                        futures.append(
-                            executor.submit(update_label, label, predict)
-                        )
-                for _ in tqdm(as_completed(futures), total=len(futures)):
-                    pass
+        if update and (project_id is None or project_id == project.id):
+            for label in project.labels.all().order_by("name"):
+                if label_name is None or label.name == label_name:
+                    print(f"Updating label {label.name}...")
+                    label.update_all(predict=predict, num_threads=8)
+                    print(f"Label {label.name} updated.")
diff --git a/clx/llm/anno_agent.py b/clx/llm/anno_agent.py
new file mode 100644
index 0000000..4d29f99
--- /dev/null
+++ b/clx/llm/anno_agent.py
@@ -0,0 +1,82 @@
+import simplejson as json
+from pydantic import BaseModel
+
+from clx.llm.agent import Agent
+
+PROJECT_INSTRUCTIONS_TEMPLATE = """
+You are an annotation assistant providing single-label classification
+annotations for the following label: {label_name}.
+
+When annotating you will be provided a text example. You should respond
+with a boolean `value` indicating whether the label "{label_name}" applies to
+the text, and a brief, one-sentence `reason` explaining how your decision
+aligns with the guidelines below.
+
+Here are some guidelines you should follow when annotating:
+
+Consider these project-level instructions. These are general, project-wide
+instructions that apply to all labels in the project. They may include examples
+of labels other than the one that you are annotating, just remember that you are
+currently annotating for the label "{label_name}" specifically.
+
+```
+{project_instructions}
+```
+"""
+
+LABEL_INSTRUCTIONS_TEMPLATE = """
+The user has also provided some label-specific instructions. These should take precedence
+over the project-level instructions if they are in conflict.
+
+```
+{label_instructions}
+```
+"""
+
+
+class Annotation(BaseModel):
+    """An annotation value with a reason."""
+
+    value: bool
+    reason: str
+
+
+class AnnoAgent(Agent):
+    """An annotation agent for single-label classification."""
+
+    default_model = "gemini/gemini-2.5-flash-lite"
+    default_completion_args = {
+        "response_format": Annotation,
+    }
+    on_init_args = [
+        "label_name",
+        "label_instructions",
+        "project_instructions",
+        "decisions",
+    ]
+
+    def on_init(
+        self, label_name, label_instructions, project_instructions, decisions
+    ):
+        system_prompt = PROJECT_INSTRUCTIONS_TEMPLATE.format(
+            label_name=label_name,
+            project_instructions=project_instructions,
+        )
+        if label_instructions is not None:
+            system_prompt += "\n\n" + LABEL_INSTRUCTIONS_TEMPLATE.format(
+                label_name=label_name,
+                label_instructions=label_instructions,
+            )
+        messages = [{"role": "system", "content": system_prompt}]
+        for decision in decisions:
+            messages.append({"role": "user", "content": decision["text"]})
+            json_content = Annotation(
+                value=decision["value"], reason=decision["reason"]
+            ).model_dump_json()
+            messages.append({"role": "assistant", "content": json_content})
+        self.state["prefix_messages"] = messages
+
+    def __call__(self, text: str) -> Annotation:
+        self.messages = [*self.state["prefix_messages"]]
+        response = self.step(messages=[{"role": "user", "content": text}])
+        return Annotation(**json.loads(response["content"]))
diff --git a/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/logs/events.out.tfevents.1768913464.c3c16d89609e.1542.0 b/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/logs/events.out.tfevents.1768913464.c3c16d89609e.1542.0
new file mode 100644
index 0000000..6500b26
Binary files /dev/null and b/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/logs/events.out.tfevents.1768913464.c3c16d89609e.1542.0 differ
diff --git a/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/logs/events.out.tfevents.1769091611.90d7d00a7e03.1540.0 b/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/logs/events.out.tfevents.1769091611.90d7d00a7e03.1540.0
new file mode 100644
index 0000000..708d493
Binary files /dev/null and b/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/logs/events.out.tfevents.1769091611.90d7d00a7e03.1540.0 differ
diff --git a/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/results.json b/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/results.json
new file mode 100644
index 0000000..45abc75
--- /dev/null
+++ b/home/projects/docketbert/runs/docketbert-final-large-395M/checkpoints/results.json
@@ -0,0 +1,9 @@
+{
+    "eval_loss": 0.5528995990753174,
+    "eval_runtime": 541.8924,
+    "eval_samples_per_second": 184.538,
+    "eval_steps_per_second": 23.067,
+    "epoch": 1.0,
+    "num_train_examples": 40761590,
+    "num_eval_examples": 100000
+}
\ No newline at end of file
diff --git a/home/projects/docketbert/runs/docketbert-final-large-395M/config.json b/home/projects/docketbert/runs/docketbert-final-large-395M/config.json
new file mode 100644
index 0000000..976be56
--- /dev/null
+++ b/home/projects/docketbert/runs/docketbert-final-large-395M/config.json
@@ -0,0 +1,32 @@
+{
+    "task": "mlm",
+    "run_name": "docketbert-final-large-395M",
+    "run_dir_parent": "/workspace/clx/projects/docketbert/runs",
+    "base_model_name": "answerdotai/ModernBERT-large",
+    "tokenizer_name": "answerdotai/ModernBERT-base",
+    "tokenize_args": {
+        "max_length": 768,
+        "padding": false
+    },
+    "model_args": {},
+    "training_args": {
+        "learning_rate": 0.0005,
+        "weight_decay": 0.01,
+        "num_train_epochs": 1,
+        "warmup_ratio": 0.05,
+        "logging_steps": 5,
+        "save_strategy": "steps",
+        "save_steps": 1000,
+        "save_total_limit": 2,
+        "eval_strategy": "steps",
+        "eval_steps": 1000,
+        "prediction_loss_only": true,
+        "remove_unused_columns": false,
+        "bf16": true,
+        "max_steps": 159224,
+        "per_device_train_batch_size": 8,
+        "per_device_eval_batch_size": 8,
+        "gradient_accumulation_steps": 32
+    },
+    "mlm_probability": 0.3
+}
\ No newline at end of file
diff --git a/home/projects/docketbert/runs/docketbert-final-sliced-175M/checkpoints/logs/events.out.tfevents.1770755421.df02e7f16835.4837.0 b/home/projects/docketbert/runs/docketbert-final-sliced-175M/checkpoints/logs/events.out.tfevents.1770755421.df02e7f16835.4837.0
new file mode 100644
index 0000000..e26b278
Binary files /dev/null and b/home/projects/docketbert/runs/docketbert-final-sliced-175M/checkpoints/logs/events.out.tfevents.1770755421.df02e7f16835.4837.0 differ
diff --git a/home/projects/docketbert/runs/docketbert-final-sliced-175M/config.json b/home/projects/docketbert/runs/docketbert-final-sliced-175M/config.json
new file mode 100644
index 0000000..fb5a7b0
--- /dev/null
+++ b/home/projects/docketbert/runs/docketbert-final-sliced-175M/config.json
@@ -0,0 +1,32 @@
+{
+    "task": "mlm",
+    "run_name": "docketbert-final-sliced-175M",
+    "run_dir_parent": "/workspace/clx/projects/docketbert/runs",
+    "base_model_name": "/workspace/clx/projects/docketbert/models/final-sliced-large-ft-interleaved-10l",
+    "tokenizer_name": "answerdotai/ModernBERT-base",
+    "tokenize_args": {
+        "max_length": 768,
+        "padding": false
+    },
+    "model_args": {},
+    "training_args": {
+        "learning_rate": 0.0005,
+        "weight_decay": 0.01,
+        "num_train_epochs": 1,
+        "warmup_ratio": 0.05,
+        "logging_steps": 5,
+        "save_strategy": "steps",
+        "save_steps": 1000,
+        "save_total_limit": 2,
+        "eval_strategy": "steps",
+        "eval_steps": 1000,
+        "prediction_loss_only": true,
+        "remove_unused_columns": false,
+        "bf16": true,
+        "max_steps": 159224,
+        "per_device_train_batch_size": 32,
+        "per_device_eval_batch_size": 32,
+        "gradient_accumulation_steps": 8
+    },
+    "mlm_probability": 0.3
+}
\ No newline at end of file
diff --git a/home/projects/docketbert/runs/docketbert-scratch-7M/checkpoints/logs/events.out.tfevents.1768838378.a1e1309953bb.1410.0 b/home/projects/docketbert/runs/docketbert-scratch-7M/checkpoints/logs/events.out.tfevents.1768838378.a1e1309953bb.1410.0
new file mode 100644
index 0000000..b03df04
Binary files /dev/null and b/home/projects/docketbert/runs/docketbert-scratch-7M/checkpoints/logs/events.out.tfevents.1768838378.a1e1309953bb.1410.0 differ
diff --git a/home/projects/docketbert/runs/docketbert-scratch-7M/config.json b/home/projects/docketbert/runs/docketbert-scratch-7M/config.json
index 1da9ed8..efa4ee9 100644
--- a/home/projects/docketbert/runs/docketbert-scratch-7M/config.json
+++ b/home/projects/docketbert/runs/docketbert-scratch-7M/config.json
@@ -1,7 +1,7 @@
 {
     "task": "mlm",
     "run_name": "docketbert-scratch-7M",
-    "run_dir_parent": "/workspace/clx/home/runs",
+    "run_dir_parent": "/workspace/clx/projects/docketbert/runs",
     "base_model_name": "answerdotai/ModernBERT-base",
     "tokenizer_name": "answerdotai/ModernBERT-base",
     "tokenize_args": {