From 1822eed8d3d234ae2f70f448ed72007c3022b8db Mon Sep 17 00:00:00 2001
From: Christopher Milan <chrismilan@ucla.edu>
Date: Fri, 19 Jun 2026 15:16:59 -0700
Subject: [PATCH] ci: only test models on cpu (#16678)

---
 .github/workflows/test.yml  | 26 ++------------------------
 test/models/test_whisper.py |  1 -
 2 files changed, 2 insertions(+), 25 deletions(-)

diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 416bdef90d..6e4d799206 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -469,7 +469,7 @@ jobs:
 # ****** Models Tests ******
 
   testmodels:
-    name: Models (llvm+cpu+gpu)
+    name: Models
     runs-on: *linux
     timeout-minutes: 15
     steps:
@@ -480,34 +480,12 @@ jobs:
         with:
           key: models
           deps: testing
-          opencl: 'true'
           llvm: 'true'
-      - name: Test models (llvm)
-        run: DEV=CPU:LLVM python -m pytest -n=auto test/models --durations=20
-      - name: Test models (opencl)
-        run: DEV=CL python -m pytest -n=auto test/models --durations=20
       - name: Test models (cpu)
         run: DEV=CPU python -m pytest -n=auto test/models --durations=20
       - name: Run process replay tests
         uses: ./.github/actions/process-replay
 
-  testmetalmodels:
-    name: Models (metal)
-    runs-on: &macos macos-26
-    timeout-minutes: 20
-    steps:
-    - name: Checkout Code
-      uses: actions/checkout@v6
-    - name: Setup Environment
-      uses: ./.github/actions/setup-tinygrad
-      with:
-        key: metal
-        deps: testing
-    - name: Test models (Metal)
-      run: DEV=METAL python -m pytest -n=auto test/models --durations=20
-    - name: Test LLaMA compile speed
-      run: DEV=METAL python test/external/external_test_speed_llama.py
-
 # ****** Feature Tests ******
 
   testdsp:
@@ -716,7 +694,7 @@ jobs:
 
   unittestmacos:
     name: MacOS (unit)
-    runs-on: *macos
+    runs-on: &macos macos-26
     timeout-minutes: 20
     steps:
     - name: Checkout Code
diff --git a/test/models/test_whisper.py b/test/models/test_whisper.py
index 818994dc86..e6e3e507c0 100644
--- a/test/models/test_whisper.py
+++ b/test/models/test_whisper.py
@@ -51,7 +51,6 @@ def wer_helper(result: str, reference: str)->float:
   wer, _, _ = metrics.word_error_rate([result], [reference])
   return wer
 
-@unittest.skipIf(Device.DEFAULT in ["CPU"], "slow")
 # TODO: WEBGPU GPU dispatch dimensions limit
 @unittest.skipIf(Device.DEFAULT == "WEBGPU", "WEBGPU GPU dispatch dimensions limit")
 class TestWhisper(unittest.TestCase):