fix: disable eval by default, set eval_steps to 0.0

- Changed default eval_steps from 0.01 to 0.0 across backend and frontend
- Fixed UI to allow eval_steps=0 (removed min=0.001 constraint)
- Added conditional eval logic with helpful console messages
- Updated tooltip to explain how to disable evaluation
- Tested: confirmed eval disabled by default with eval_steps=0.0

studio/backend/core/training/trainer.py

@@ -543,7 +543,7 @@ class UnslothTrainer:
     def start_training(self,
                        dataset: Dataset,
                        eval_dataset: Dataset = None,
-                       eval_steps: float = 0.01,
+                       eval_steps: float = 0.00,
                        output_dir: str = "./outputs",
                        num_epochs: int = 3,
                        learning_rate: float = 5e-5,
@@ -743,12 +743,16 @@ class UnslothTrainer:
 
             # ========== EVAL CONFIGURATION ==========
             eval_dataset = training_args.get('eval_dataset', None)
-            eval_steps_val = training_args.get('eval_steps', 0.01)
+            eval_steps_val = training_args.get('eval_steps', 0.00)
             if eval_dataset is not None:
-                config_args["eval_strategy"] = "steps"
-                config_args["eval_steps"] = eval_steps_val
-                print(f"Evaluation enabled: eval_steps={eval_steps_val} (fraction of total steps)\n")
-                print(f"Eval dataset: {len(eval_dataset)} rows\n")
+                if eval_steps_val > 0:
+                    config_args["eval_strategy"] = "steps"
+                    config_args["eval_steps"] = eval_steps_val
+                    print(f"✅ Evaluation enabled: eval_steps={eval_steps_val} (fraction of total steps)\n")
+                    print(f"Eval dataset: {len(eval_dataset)} rows\n")
+                else:
+                    print(f"⚠️  Eval dataset provided but eval_steps={eval_steps_val} (disabled)\n")
+                    print("To enable evaluation, set eval_steps > 0.0\n")
             else:
                 print("No eval dataset — evaluation disabled\n")
 

studio/backend/core/training/training.py

@@ -115,7 +115,7 @@ class TrainingBackend:
                        subset: str = None,
                        train_split: str = "train",
                        eval_split: str = None,
-                       eval_steps: float = 0.01,
+                       eval_steps: float = 0.00,
                        is_dataset_multimodal: bool = False) -> bool:
         """
         Start training.

studio/backend/models/training.py

@@ -21,7 +21,7 @@ class TrainingStartRequest(BaseModel):
     subset: Optional[str] = None
     train_split: Optional[str] = Field("train", description="Training split name")
     eval_split: Optional[str] = Field(None, description="Eval split name. None = auto-detect")
-    eval_steps: float = Field(0.01, description="Fraction of total steps between evals (0-1)")
+    eval_steps: float = Field(0.00, description="Fraction of total steps between evals (0-1)")
 
     @model_validator(mode="before")
     @classmethod

studio/frontend/src/config/training.ts

@@ -103,7 +103,7 @@ export const DEFAULT_HYPERPARAMS = {
   warmupSteps: 5,
   maxSteps: 0,
   saveSteps: 0,
-  evalSteps: 0.01,
+  evalSteps: 0.00,
   packing: false,
   trainOnCompletions: false,
   gradientCheckpointing: "unsloth" as const,

studio/frontend/src/features/studio/sections/params-section.tsx

@@ -727,12 +727,12 @@ export function ParamsSection(): ReactElement {
                 </Row>
                 <Row
                   label="Eval Steps"
-                  tooltip="Fraction of total training steps between evaluations. E.g. 0.01 = evaluate every 1% of steps."
+                  tooltip="Fraction of total training steps between evaluations (0-1). Set to 0 to disable evaluation. E.g. 0.01 = evaluate every 1% of steps."
                 >
                   <Input
                     type="number"
                     step="0.01"
-                    min="0.001"
+                    min="0"
                     max="1"
                     value={store.evalSteps}
                     onChange={(e) => store.setEvalSteps(Number(e.target.value))}