update: stable

environments/optimizer/deploy.py

@@ -14,7 +14,7 @@ from modal.sandbox import Sandbox
 
 # --- Modal App and Images ---
 APP_NAME = "optimizer-test"
-SANDBOX_APP_NAME = "new_sandbox"
+SANDBOX_APP_NAME = "new_sandbox_test"
 
 app = App(APP_NAME)
 
@@ -28,9 +28,6 @@ benchmark_volume = Volume.from_name("benchmark-responses")
 optimizers_volume = Volume.from_name("optimizers")
 sys_prompt_volume = Volume.from_name("optimizerSystemPrompt")
 
-# --- Sandbox Setup ---
-sandbox_app = App.lookup(SANDBOX_APP_NAME, create_if_missing=True)
-sandbox = Sandbox.create(app=sandbox_app, image=sandbox_image, timeout=60 * 60)
 
 
 # --- Utility Functions ---
@@ -60,6 +57,10 @@ def send_code(code: str):
     """Send and execute optimizer code in the sandbox environment."""
     filename, optimizer_code = _write_optimizer_code_to_volume(code, optimizers_volume)
 
+    # --- Sandbox Setup ---
+    sandbox_app = App.lookup(SANDBOX_APP_NAME, create_if_missing=True)
+    sandbox = Sandbox.create(app=sandbox_app, image=sandbox_image, timeout=60 * 60)
+
     # Write code to sandbox
     with sandbox.open(filename, "w") as f:
         f.write(optimizer_code)

environments/optimizer/evaluator.py

@@ -36,7 +36,7 @@ class OptimizerEvaluator:
                 CategoricalJudgeUnit(
                     name='Judge',
                     categories=DiscreteScale(['yes', 'no']),
-                    explanation=False
+                    explanation=False,
                 ).prompt("""
                     You are an expert code validator specializing in PyTorch optimizers. Your task is to determine if the provided optimizer code is completely valid and error-free.
 
@@ -114,12 +114,6 @@ print(f"\nOptimal x: {x.item():.4f}")
 
     stderr = """
     Traceback (most recent call last):
-      File "optimizer.py", line 8, in <module>
-        optimizer = torch.optim.SGD([x], lr=0.1)
-    RuntimeError: CUDA out of memory. Tried to allocate 2.00 MiB (GPU 0; 4.00 GiB total capacity; 3.47 GiB already allocated; 0 bytes free; 3.47 GiB reserved in total by PyTorch)
-
-    Traceback (most recent call last):
-      File "optimizer.py", line 12, in <
     """
 
     score = evaluator.check_validity(optimizer_code=optimizer_code, stdout=stdout, stderr=stderr)