Update float8 tests to run nightly

.github/workflows/float8_test.yml

@@ -1,14 +1,10 @@
-name: Run Float8 Tests
+name: Run Float8 Tests Nightly
 
 on:
-  push:
-    branches:
-      - main
-      - 'gh/**'
-  pull_request:
-    branches:
-      - main
-      - 'gh/**'
+  schedule:
+    # 3.27 am PST every day
+    - cron: "27 11 * * *"
+  workflow_dispatch:
 
 concurrency:
   group: float8_test-${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && github.run_number || github.ref }}
@@ -53,3 +49,8 @@ jobs:
         pip install -r dev-requirements.txt
         pip install .
         pytest test/float8 --verbose -s
+        pytest test/dtypes/test_affine_quantized_float.py --verbose -s
+        # Float8nocompile tests
+        cd torchao/prototype/float8nocompile
+        pytest kernels/ --verbose -s
+        pytest test/train_test.py --verbose -s

test/dtypes/test_affine_quantized_float.py

@@ -147,7 +147,10 @@ def test_fp8_linear_variants(
     )
     def test_invalid_granularity(self):
         with pytest.raises(ValueError, match="Invalid granularity specification"):
-            float8_dynamic_activation_float8_weight(granularity="invalid")
+            model = ToyLinearModel(64, 64).eval().to(torch.float32).to("cuda")
+            quantize_(
+                model, float8_dynamic_activation_float8_weight(granularity="invalid")
+            )
 
     @unittest.skipIf(
         not is_sm_at_least_89(), "Requires GPU with compute capability >= 8.9"
@@ -157,7 +160,13 @@ def test_mismatched_granularity(self):
             ValueError,
             match="Different granularities for activation and weight are not supported",
         ):
-            float8_dynamic_activation_float8_weight(granularity=(PerTensor(), PerRow()))
+            model = ToyLinearModel(64, 64).eval().to(torch.float32).to("cuda")
+            quantize_(
+                model,
+                float8_dynamic_activation_float8_weight(
+                    granularity=(PerTensor(), PerRow())
+                ),
+            )
 
     @unittest.skipIf(
         not is_sm_at_least_89(), "Requires GPU with compute capability >= 8.9"
@@ -166,9 +175,16 @@ def test_unsupported_granularity(self):
         class UnsupportedGranularity:
             pass
 
-        with pytest.raises(ValueError, match="Invalid granularity types"):
-            float8_dynamic_activation_float8_weight(
-                granularity=(UnsupportedGranularity(), UnsupportedGranularity())
+        with pytest.raises(
+            ValueError,
+            match="Invalid granularity types:",
+        ):
+            model = ToyLinearModel(64, 64).eval().to(torch.float32).to("cuda")
+            quantize_(
+                model,
+                float8_dynamic_activation_float8_weight(
+                    granularity=(UnsupportedGranularity(), UnsupportedGranularity())
+                ),
             )
 
     @unittest.skipIf(not torch.cuda.is_available(), "Need CUDA available")