Fix Olmo trunk test and skip XNNPack trunk tests on mac (#13528)

Unblock trunk to advance `viable/strict`

Intermittent segmentation faults on all of the XNNPack Optimum mac
tests. Described in more detail
[here](#13530) where the
segmentation fault happens consistently, but interestingly enough this
is before the pin bump and it is only happening intermittently.

Author: jackzhxng

.ci/scripts/test_huggingface_optimum_model.py

@@ -369,7 +369,7 @@ def test_vit(model_id, model_dir, recipe, *, quantize=False, run_only=False):
         ),  # fails to lower for CoreML
         "smollm2-135m": ("HuggingFaceTB/SmolLM2-135M", test_text_generation),
         "smollm3-3b": ("HuggingFaceTB/SmolLM3-3B", test_text_generation),
-        "olmo": ("allenai/OLMo-1B-hf", test_text_generation),
+        "olmo-1b": ("allenai/OLMo-1B-hf", test_text_generation),
     }
 
     _mask_fill_mapping = {

.github/workflows/trunk.yml

@@ -836,14 +836,14 @@ jobs:
     strategy:
       matrix:
         config: [
-          # XNNPack.
-          llama3.2-1b|xnnpack|--quantize,
-          qwen3-0.6b|xnnpack|--quantize,
-          qwen3-1.7b|xnnpack|--quantize,
-          gemma3-1b|xnnpack|--quantize,
-          phi4-mini|xnnpack|--quantize,
-          smollm2-135m|xnnpack|--quantize,
-          smollm3-3b|xnnpack|--quantize,
+          # # XNNPack. (Skipping for now due to intermittent segmentation faults, see https://github.com/huggingface/optimum-executorch/issues/122.)
+          # llama3.2-1b|xnnpack|--quantize,
+          # qwen3-0.6b|xnnpack|--quantize,
+          # qwen3-1.7b|xnnpack|--quantize,
+          # gemma3-1b|xnnpack|--quantize,
+          # phi4-mini|xnnpack|--quantize,
+          # smollm2-135m|xnnpack|--quantize,
+          # smollm3-3b|xnnpack|--quantize,
           # CoreML.
           llama3.2-1b|coreml_fp32_gpu|--quantize,
           qwen3-0.6b|coreml_fp32_gpu|--quantize,