170b412c1d* Fix the CPU-only ROCm routing errors and two font-scale UI flakes Two unrelated causes of red CI on every PR, both reproduced before fixing. ROCm routing: 12 errors on Repo tests (CPU). The spoof reports an AMD GPU, and unsloth_zoo pulls in bitsandbytes, which picks a compute backend at import. Once torch looks like a GPU is present, bnb loads its ROCm/CUDA ops, which a CPU-only torch cannot satisfy (no libhipblas.so.2, no torch._C._cuda_getCurrentRawStream), so the child died before printing RESULT. Nothing here tests bitsandbytes, so import it first, under the honest hardware. Reproduced in a CPU-only torch venv: 11 passed with 12 errors before, 23 passed after. Still 23 passed on a CUDA build. Font-scale UI: the select-viewport step pressed ArrowDown six times behind fixed sleeps, but Radix moves focus into the listbox after the content opens, so on a loaded runner the keys landed on the trigger and nothing scrolled. Wait on the overflow and press until it moves, bounded at 40. The same fixed-sleep pattern made open_appearance miss the dialog when the shortcut fired before the app wired its handler; alternate both chords on a bounded retry and wait for the control the caller is about to drive. Both were reproduced locally by running the suite against a real Studio under full CPU load. Original: 2 of 10 passed, with the exact CI signature 'keyboard did not scroll the select viewport: 0' five times. Fixed: 10 of 10. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Keep the ROCm routing assertion live on Apple Silicon for PR #7469 * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: danielhanchen <unslothai@gmail.com> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
98 lines
3.8 KiB
Python
98 lines
3.8 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-only
|
|
# Copyright 2026-present the Unsloth AI Inc. team.
|
|
"""RDNA 2/3/4 routing, validated on CPU-only CI with no AMD hardware.
|
|
|
|
tests/_zoo_rocm_spoof.py presents torch as each Radeon gfx arch, then we assert
|
|
unsloth_zoo routes it: device_type -> "hip", llama.cpp target -> ("rocm", gfx),
|
|
and the per-family ROCm bundle suffix. The torch-facing checks run in a
|
|
subprocess so the spoof never leaks into sibling tests and DEVICE_TYPE (cached
|
|
at import) resolves from a clean process.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
|
|
pytest.importorskip("torch")
|
|
pytest.importorskip("unsloth_zoo")
|
|
|
|
_TESTS_DIR = Path(__file__).resolve().parents[2] # tests/
|
|
|
|
# gfx -> (expected llama.cpp target, expected ROCm bundle family).
|
|
_ARCHES = {
|
|
"gfx1030": (("rocm", "gfx1030"), "gfx103X"), # RDNA2
|
|
"gfx1031": (("rocm", "gfx1031"), "gfx103X"),
|
|
"gfx1032": (("rocm", "gfx1032"), "gfx103X"),
|
|
"gfx1034": (("rocm", "gfx1034"), "gfx103X"),
|
|
"gfx1100": (("rocm", "gfx1100"), "gfx110X"), # RDNA3
|
|
"gfx1101": (("rocm", "gfx1101"), "gfx110X"),
|
|
"gfx1102": (("rocm", "gfx1102"), "gfx110X"),
|
|
"gfx1150": (("rocm", "gfx1150"), "gfx1150"), # RDNA3.5 APU (self-family)
|
|
"gfx1151": (("rocm", "gfx1151"), "gfx1151"),
|
|
"gfx1200": (("rocm", "gfx1200"), "gfx120X"), # RDNA4
|
|
"gfx1201": (("rocm", "gfx1201"), "gfx120X"),
|
|
}
|
|
|
|
# Child: spoof each arch, then record device_type once (fresh import) and the
|
|
# live llama.cpp target per arch. Emits one JSON line the parent parses.
|
|
_CHILD = """
|
|
import json, sys
|
|
sys.path.insert(0, {tests!r})
|
|
# Import bitsandbytes under the real torch first. unsloth_zoo pulls it in, and it
|
|
# picks a compute backend at import: once the spoof reports an AMD GPU, it loads
|
|
# its ROCm/CUDA ops, which a CPU-only torch cannot satisfy (no libhipblas, no
|
|
# torch._C._cuda_getCurrentRawStream) and the child dies before printing RESULT.
|
|
# Nothing here tests bitsandbytes, so let it see the honest hardware.
|
|
try:
|
|
import bitsandbytes # noqa: F401
|
|
except Exception:
|
|
pass
|
|
import _zoo_rocm_spoof as spoof
|
|
arches = {arches!r}
|
|
spoof.apply(arches[0])
|
|
from unsloth_zoo.device_type import get_device_type, is_hip
|
|
device_type = [get_device_type(), is_hip()]
|
|
from unsloth_zoo import llama_cpp as lc
|
|
targets = {{}}
|
|
for gfx in arches:
|
|
spoof.apply(gfx)
|
|
targets[gfx] = list(lc._detect_gpu_target())
|
|
print("RESULT " + json.dumps({{"device_type": device_type, "targets": targets}}))
|
|
"""
|
|
|
|
|
|
@pytest.fixture(scope = "module")
|
|
def routed():
|
|
code = _CHILD.format(tests = str(_TESTS_DIR), arches = list(_ARCHES))
|
|
# get_device_type() returns "mlx" before it ever looks at torch on Darwin arm64
|
|
# with mlx installed, so the spoof would be ignored. Force the GPU path to keep
|
|
# the assertion live there instead of skipping it.
|
|
env = {**os.environ, "UNSLOTH_FORCE_GPU_PATH": "1"}
|
|
proc = subprocess.run([sys.executable, "-c", code], capture_output = True, text = True, env = env)
|
|
line = next((l for l in proc.stdout.splitlines() if l.startswith("RESULT ")), None)
|
|
assert line, f"child produced no result.\nstdout:\n{proc.stdout}\nstderr:\n{proc.stderr}"
|
|
return json.loads(line[len("RESULT ") :])
|
|
|
|
|
|
@pytest.mark.parametrize("gfx", list(_ARCHES))
|
|
def test_detect_gpu_target(routed, gfx):
|
|
# RDNA card is routed to its ROCm gfx target (drives the llama.cpp bundle).
|
|
assert tuple(routed["targets"][gfx]) == _ARCHES[gfx][0]
|
|
|
|
|
|
def test_device_type_is_hip(routed):
|
|
# An RDNA card must resolve the compute device_type to "hip".
|
|
assert routed["device_type"] == ["hip", True]
|
|
|
|
|
|
@pytest.mark.parametrize("gfx", list(_ARCHES))
|
|
def test_rocm_gfx_family(gfx):
|
|
# Pure mapping (no torch): each gfx picks the right per-family ROCm bundle.
|
|
from unsloth_zoo import llama_cpp as lc
|
|
assert lc._rocm_gfx_family(gfx) == _ARCHES[gfx][1]
|