try with METAL_XCODE=1 for tinygrad metal

.github/workflows/test.yml

@@ -23,7 +23,7 @@ jobs:
         pip install .
 
     - name: Run tests
-      run: python3 -m exo.inference.test_inference_engine
+      run: METAL_XCODE=1 python3 -m exo.inference.test_inference_engine
 
   discovery_integration_test:
     runs-on: macos-latest

exo/inference/test_inference_engine.py

@@ -31,8 +31,8 @@ asyncio.run(test_inference_engine(
 ))
 
 # TODO: Waiting on https://github.com/tinygrad/tinygrad/issues/5549
-# asyncio.run(test_inference_engine(
-#     TinygradDynamicShardInferenceEngine(),
-#     TinygradDynamicShardInferenceEngine(),
-#     "llama3-8b-sfr",
-# ))
+asyncio.run(test_inference_engine(
+    TinygradDynamicShardInferenceEngine(),
+    TinygradDynamicShardInferenceEngine(),
+    "llama3-8b-sfr",
+))