[CPU][Zen] Update zentorch to 2.14.0.1 & enable float16 on Venice and later (#61411)

Signed-off-by: Ganesh R <ganesh.r@amd.com>
Co-authored-by: Kevin H. Luu <khluu000@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>

.buildkite/hardware_tests/cpu.yaml

@@ -277,7 +277,7 @@ steps:
     # `uv pip install "vllm[zen]"`, which forces uv to re-resolve vLLM's full
     # dependency closure and chokes on its transitive triton-cpu URL dependency.
     # Keep this pin in sync with the `zen` extra in setup.py.
-    ZENTORCH_VERSION: "2.13.0.1"
+    ZENTORCH_VERSION: "2.14.0.1"
   retry:
     automatic:
       - exit_status: 1  # Transient Docker/BuildKit failure

docs/getting_started/installation/cpu.md

@@ -159,7 +159,7 @@ VLLM_USE_PRECOMPILED=1 VLLM_PRECOMPILED_WHEEL_VARIANT=cpu VLLM_TARGET_DEVICE=cpu
 ### Which `dtype` should be used?
 
 - Currently, vLLM CPU uses model default settings as `dtype`. However, due to unstable float16 support in torch CPU, it is recommended to explicitly set `dtype=bfloat16` if there are any performance or accuracy problem.  
-- On AMD Zen CPUs (`ZenCpuPlatform`), `float16` is **not** supported. Only `bfloat16` and `float32` are accepted; models declared with `float16` are auto-downcast to `bfloat16` at model load time. See [AMD Zen optimizations](#amd-zen-optimizations).
+- On AMD Zen CPUs (`ZenCpuPlatform`), `float16` requires AVX512-FP16, available from Venice (6th Gen EPYC) onwards. On Turin (5th Gen EPYC) and earlier only `bfloat16` and `float32` are accepted, and models declared with `float16` are auto-downcast to `bfloat16` at model load time. See [AMD Zen optimizations](#amd-zen-optimizations).
 
 ### How to launch a vLLM service on CPU?
 

docs/getting_started/installation/cpu.x86.inc.md

@@ -273,7 +273,9 @@ Otherwise, vLLM falls back to the default `CpuPlatform` (oneDNN / sgl-kernel pat
 
 ### Supported dtypes
 
-`float16` is **not** supported on `ZenCpuPlatform`. `ZenCpuPlatform.supported_dtypes` advertises only `bfloat16` and `float32`, so models declared with `torch_dtype=float16` are auto-downcast to `bfloat16` at load time with the standard `"Your device 'cpu' doesn't support torch.float16. Falling back to torch.bfloat16 for compatibility."` warning emitted from `vllm/config/model.py`.
+`ZenCpuPlatform.supported_dtypes` always advertises `bfloat16` and `float32`, and adds `float16` when `torch.ops.zentorch.zentorch_is_fp16_supported()` reports AVX512-FP16, which AMD ships from Venice (6th Gen EPYC) onwards. On Turin (5th Gen EPYC) and earlier, models declared with `torch_dtype=float16` are auto-downcast to `bfloat16` at load time with the standard `"Your device 'cpu' doesn't support torch.float16. Falling back to torch.bfloat16 for compatibility."` warning emitted from `vllm/config/model.py`.
+
+`bfloat16` is listed first either way, so `--dtype auto` keeps resolving to it.
 
 ### Environment variables
 

setup.py

@@ -1688,7 +1688,7 @@ setup(
     install_requires=get_requirements(),
     extras_require={
         # AMD Zen CPU optimizations via zentorch
-        "zen": ["zentorch==2.14.0.0"],
+        "zen": ["zentorch==2.14.0.1"],
         "bench": ["pandas", "matplotlib", "seaborn", "datasets", "scipy", "plotly"],
         "tensorizer": ["tensorizer==2.10.1"],
         "fastsafetensors": ["fastsafetensors >= 0.3.3"],

tests/test_zen_cpu_platform_detection.py

@@ -2,9 +2,11 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 import builtins
 import logging
+from types import SimpleNamespace
 from unittest.mock import mock_open, patch
 
 import pytest
+import torch
 
 from vllm.platforms import (
     _is_amd_zen_cpu,
@@ -139,3 +141,18 @@ def test_platform_plugin_failure_is_logged(caplog_vllm):
     assert platform == "vllm.platforms.interface.UnspecifiedPlatform"
     assert "Platform plugin cpu failed during detection" in caplog_vllm.text
     assert "RuntimeError: plugin exploded" in caplog_vllm.text
+
+
+@pytest.mark.parametrize("fp16_supported", [True, False])
+def test_zen_cpu_supported_dtypes_track_fp16_capability(fp16_supported):
+    from vllm.platforms.zen_cpu import ZenCpuPlatform
+
+    # zentorch is absent on most CI hosts, so torch.ops.zentorch has to be
+    # stood up rather than patched in place.
+    zentorch_ops = SimpleNamespace(zentorch_is_fp16_supported=lambda: fp16_supported)
+    with patch.object(torch.ops, "zentorch", zentorch_ops, create=True):
+        dtypes = ZenCpuPlatform().supported_dtypes
+
+    # bf16 stays first so `--dtype auto` resolves to it on every Zen part.
+    assert dtypes[0] is torch.bfloat16
+    assert (torch.float16 in dtypes) is fp16_supported

vllm/platforms/zen_cpu.py

@@ -25,8 +25,10 @@ class ZenCpuPlatform(CpuPlatform):
         # is_cpu() also returns True for this platform (inherited from CpuPlatform).
         return True
 
-    # Currently, AMD CPUs do not support float16 compute.
-    # Hence explicitly return bfloat16 and float32.
+    # fp16 compute needs AVX512-FP16, which AMD ships from Venice (6th Gen
+    # EPYC) onwards; Turin (5th Gen EPYC) and earlier have no fp16 kernels.
     @property
     def supported_dtypes(self) -> list[torch.dtype]:
+        if torch.ops.zentorch.zentorch_is_fp16_supported():
+            return [torch.bfloat16, torch.float32, torch.float16]
         return [torch.bfloat16, torch.float32]