* studio/setup.sh: guard empty CUDA arch detection in the source build PR #5826 hardened setup.sh for fresh CUDA toolkits, but the source build still set -DCMAKE_CUDA_ARCHITECTURES only when nvidia-smi reported a compute capability. When that query returns nothing the build proceeded with no explicit arch list, so llama.cpp built PTX only. On a driver older than the toolkit that binary fails at runtime with "the provided PTX was compiled with an unsupported toolchain" - the build succeeds, so neither the build-time check nor the CPU fallback caught it (issue #5854). Resolve the arch list before committing to a CUDA build. A new pure helper _resolve_cuda_archs parses and de-duplicates the nvidia-smi compute_cap output and honors an explicit UNSLOTH_LLAMA_CUDA_ARCHS override. When the result is empty, build CPU llama.cpp instead of a PTX-only binary, with a clear message pointing at the override - so the user still ends up with a working llama-server. The override also lets advanced users force a native build on hosts where nvidia-smi cannot report compute_cap. No behavior change when an arch is detected: -DGGML_CUDA=ON plus the arch, CUDA flags and NVCC_PREPEND_FLAGS are assembled exactly as before. Adds tests/sh/test_resolve_cuda_archs.sh (single/multi/dedup/empty/garbage/ whitespace/override cases), wired into tests/run_all.sh and the studio-backend-ci.yml shell-test loop. * studio/setup.sh: resolve nvidia-smi via /usr/bin fallback for arch detection Addresses review feedback on the empty-CUDA-arch guard: _setup_has_usable_nvidia_gpu classifies a host as NVIDIA-usable using nvidia-smi on PATH OR /usr/bin/nvidia-smi, but the new arch detection probed only `command -v nvidia-smi`. On a GPU host where nvidia-smi is off PATH (reachable only at /usr/bin), arch detection returned empty and the new empty-arch branch dropped the build to CPU, losing CUDA. Mirror the same PATH-then-/usr/bin resolution so those hosts still get a native CUDA build. Also scope _resolve_cuda_archs locals with `local` (no behavior change; it already runs under command substitution). * tests: update compute_cap-probe assertion for $_smi_bin resolution The nvidia-smi /usr/bin fallback parameterized the binary in the compute_cap probe (_setup_run_smi "$_smi_bin" ...), so the literal-string assertion in test_compute_cap_probe_timeout_wrapped no longer matched. Assert the probe is preceded by _setup_run_smi (timeout-wrapped) instead, scanning all occurrences so the comment mention is ignored. Same intent, binary-agnostic. * tests: ruff-format the compute_cap probe assertion (pre-commit) Collapse the backslash-continued assert onto one line and normalize slice spacing so the ruff-format pre-commit hook (0.6.9) is satisfied. Formatting only; no behavior change. * Tighten code comments (no logic change) * studio(windows): build CPU when CUDA arch is undetectable (#5854) The Windows source build added -DGGML_CUDA=ON unconditionally but only set -DCMAKE_CUDA_ARCHITECTURES when $CudaArch was detected. With no detectable compute capability that produced a PTX-only binary, the same hole the Linux fix closed. Build CPU llama.cpp in that case, and honor UNSLOTH_LLAMA_CUDA_ARCHS to force a CUDA build, matching setup.sh. Detected-arch builds are unchanged. * test: anchor NVCC_PREPEND_FLAGS scope check on the final CPU branch The undetectable-arch CPU fallback adds an earlier -DGGML_CUDA=OFF, so the ordering check now anchors on -DGGML_CUDA=ON and the last -DGGML_CUDA=OFF instead of the first. --------- Co-authored-by: Daniel Han <michaelhan2050@gmail.com>
This commit is contained in:
Daniel Han
GitHub
parent
bebc93d8fc
commit
70926822db
7 changed files with 185 additions and 57 deletions
1
.github/workflows/studio-backend-ci.yml
vendored
1
.github/workflows/studio-backend-ci.yml
vendored
|
|
@ -226,6 +226,7 @@ jobs:
|
|||
tests/sh/test_studio_home_node_dir.sh \
|
||||
tests/sh/test_system_node_readonly.sh \
|
||||
tests/sh/test_nvcc_meets_llama_minimum.sh \
|
||||
tests/sh/test_resolve_cuda_archs.sh \
|
||||
tests/sh/test_tauri_install_exit_order.sh \
|
||||
tests/sh/test_torch_constraint.sh \
|
||||
tests/sh/test_torch_flavor.sh; do
|
||||
|
|
|
|||
|
|
@ -3704,35 +3704,49 @@ if (-not $NeedLlamaSourceBuild) {
|
|||
$CmakeArgs += '-DCMAKE_EXE_LINKER_FLAGS=/NODEFAULTLIB:LIBCMT'
|
||||
# CUDA flags -- only if GPU available, otherwise explicitly disable
|
||||
if ($HasNvidiaSmi -and $NvccPath) {
|
||||
$CmakeArgs += '-DGGML_CUDA=ON'
|
||||
# Accept a host MSVC newer than nvcc's whitelist; a fresh toolkit
|
||||
# (e.g. CUDA 13.3) otherwise aborts with "#error -- unsupported
|
||||
# Microsoft Visual Studio version!". Mirrors the Linux fix. Via env
|
||||
# (covers the configure probe + build), after Refresh-Environment, idempotent.
|
||||
$nvccAllowFlag = '-allow-unsupported-compiler'
|
||||
if ([string]::IsNullOrEmpty($env:NVCC_PREPEND_FLAGS)) {
|
||||
$env:NVCC_PREPEND_FLAGS = $nvccAllowFlag
|
||||
} elseif ($env:NVCC_PREPEND_FLAGS -notlike "*$nvccAllowFlag*") {
|
||||
$env:NVCC_PREPEND_FLAGS = "$($env:NVCC_PREPEND_FLAGS) $nvccAllowFlag"
|
||||
}
|
||||
substep "NVCC_PREPEND_FLAGS = $env:NVCC_PREPEND_FLAGS"
|
||||
$CmakeArgs += "-DCUDAToolkit_ROOT=$CudaToolkitRoot"
|
||||
$CmakeArgs += "-DCUDA_TOOLKIT_ROOT_DIR=$CudaToolkitRoot"
|
||||
$CmakeArgs += "-DCMAKE_CUDA_COMPILER=$NvccPath"
|
||||
if ($CudaArch) {
|
||||
# Validate nvcc actually supports this architecture
|
||||
if (Test-NvccArchSupport -NvccExe $NvccPath -Arch $CudaArch) {
|
||||
$CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$CudaArch"
|
||||
} else {
|
||||
# GPU arch too new for this toolkit -- fall back to highest supported.
|
||||
# PTX forward-compatibility will JIT-compile for the actual GPU at runtime.
|
||||
$maxArch = Get-NvccMaxArch -NvccExe $NvccPath
|
||||
if ($maxArch) {
|
||||
$CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$maxArch"
|
||||
substep "GPU is sm_$CudaArch but nvcc only supports up to sm_$maxArch" "Yellow"
|
||||
substep "Building with sm_$maxArch (PTX will JIT for your GPU at runtime)" "Yellow"
|
||||
# UNSLOTH_LLAMA_CUDA_ARCHS (e.g. "120" or "89;86") forces the build
|
||||
# arch and wins over detection, matching setup.sh.
|
||||
$CudaArchOverride = if ($env:UNSLOTH_LLAMA_CUDA_ARCHS) { ($env:UNSLOTH_LLAMA_CUDA_ARCHS -replace '\s', '') } else { '' }
|
||||
if ((-not $CudaArch) -and (-not $CudaArchOverride)) {
|
||||
# No detectable compute capability (#5854): -DGGML_CUDA=ON with no
|
||||
# arch builds a PTX-only binary, so build CPU instead. Mirrors the
|
||||
# Linux fix; set UNSLOTH_LLAMA_CUDA_ARCHS=120 to force a CUDA build.
|
||||
substep "could not detect a CUDA compute capability; building CPU llama.cpp instead of a PTX-only binary (set UNSLOTH_LLAMA_CUDA_ARCHS=120 to force a CUDA build)." "Yellow"
|
||||
$CmakeArgs += '-DGGML_CUDA=OFF'
|
||||
} else {
|
||||
$CmakeArgs += '-DGGML_CUDA=ON'
|
||||
# Accept a host MSVC newer than nvcc's whitelist; a fresh toolkit
|
||||
# (e.g. CUDA 13.3) otherwise aborts with "#error -- unsupported
|
||||
# Microsoft Visual Studio version!". Mirrors the Linux fix. Via env
|
||||
# (covers the configure probe + build), after Refresh-Environment, idempotent.
|
||||
$nvccAllowFlag = '-allow-unsupported-compiler'
|
||||
if ([string]::IsNullOrEmpty($env:NVCC_PREPEND_FLAGS)) {
|
||||
$env:NVCC_PREPEND_FLAGS = $nvccAllowFlag
|
||||
} elseif ($env:NVCC_PREPEND_FLAGS -notlike "*$nvccAllowFlag*") {
|
||||
$env:NVCC_PREPEND_FLAGS = "$($env:NVCC_PREPEND_FLAGS) $nvccAllowFlag"
|
||||
}
|
||||
substep "NVCC_PREPEND_FLAGS = $env:NVCC_PREPEND_FLAGS"
|
||||
$CmakeArgs += "-DCUDAToolkit_ROOT=$CudaToolkitRoot"
|
||||
$CmakeArgs += "-DCUDA_TOOLKIT_ROOT_DIR=$CudaToolkitRoot"
|
||||
$CmakeArgs += "-DCMAKE_CUDA_COMPILER=$NvccPath"
|
||||
if ($CudaArchOverride) {
|
||||
# Forced arch wins verbatim (no nvcc validation), matching setup.sh.
|
||||
$CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$CudaArchOverride"
|
||||
} elseif ($CudaArch) {
|
||||
# Validate nvcc actually supports this architecture
|
||||
if (Test-NvccArchSupport -NvccExe $NvccPath -Arch $CudaArch) {
|
||||
$CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$CudaArch"
|
||||
} else {
|
||||
# GPU arch too new for this toolkit -- fall back to highest supported.
|
||||
# PTX forward-compatibility will JIT-compile for the actual GPU at runtime.
|
||||
$maxArch = Get-NvccMaxArch -NvccExe $NvccPath
|
||||
if ($maxArch) {
|
||||
$CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$maxArch"
|
||||
substep "GPU is sm_$CudaArch but nvcc only supports up to sm_$maxArch" "Yellow"
|
||||
substep "Building with sm_$maxArch (PTX will JIT for your GPU at runtime)" "Yellow"
|
||||
}
|
||||
# else: omit flag entirely, let cmake pick defaults
|
||||
}
|
||||
# else: omit flag entirely, let cmake pick defaults
|
||||
}
|
||||
}
|
||||
} else {
|
||||
|
|
|
|||
|
|
@ -155,6 +155,31 @@ _nvcc_meets_llama_minimum() {
|
|||
echo "$_raw"
|
||||
}
|
||||
|
||||
# Echo a ';'-separated CUDA arch list (e.g. "86;120"). Override ($2,
|
||||
# UNSLOTH_LLAMA_CUDA_ARCHS) wins verbatim; else parse+dedupe compute_cap text
|
||||
# ($1). Empty means "no arch detected", so the caller builds CPU instead of a
|
||||
# PTX-only binary that fails on an old driver (#5854).
|
||||
_resolve_cuda_archs() {
|
||||
local _raw_caps=$1
|
||||
local _arch_override=$2
|
||||
if [ -n "$_arch_override" ]; then
|
||||
printf '%s' "$_arch_override"
|
||||
return 0
|
||||
fi
|
||||
local _archs="" _cap _arch
|
||||
while IFS= read -r _cap; do
|
||||
_cap=$(printf '%s' "$_cap" | tr -d '[:space:]')
|
||||
if [[ "$_cap" =~ ^([0-9]+)\.([0-9]+)$ ]]; then
|
||||
_arch="${BASH_REMATCH[1]}${BASH_REMATCH[2]}"
|
||||
case ";$_archs;" in
|
||||
*";$_arch;"*) ;;
|
||||
*) _archs="${_archs:+$_archs;}$_arch" ;;
|
||||
esac
|
||||
fi
|
||||
done <<< "$_raw_caps"
|
||||
printf '%s' "$_archs"
|
||||
}
|
||||
|
||||
# Run a GPU probe under a 10s timeout when `timeout` is available so a wedged
|
||||
# NVIDIA driver cannot hang setup; fall back to a bare call where it is not.
|
||||
_setup_run_smi() {
|
||||
|
|
@ -1517,35 +1542,38 @@ else
|
|||
fi
|
||||
|
||||
if [ "$_CUDA_TOOLKIT_ALLOWED" = true ]; then
|
||||
CMAKE_ARGS="$CMAKE_ARGS -DGGML_CUDA=ON"
|
||||
|
||||
CUDA_ARCHS=""
|
||||
if command -v nvidia-smi &>/dev/null; then
|
||||
_raw_caps=$(_setup_run_smi nvidia-smi --query-gpu=compute_cap --format=csv,noheader 2>/dev/null || true)
|
||||
while IFS= read -r _cap; do
|
||||
_cap=$(echo "$_cap" | tr -d '[:space:]')
|
||||
if [[ "$_cap" =~ ^([0-9]+)\.([0-9]+)$ ]]; then
|
||||
_arch="${BASH_REMATCH[1]}${BASH_REMATCH[2]}"
|
||||
case ";$CUDA_ARCHS;" in
|
||||
*";$_arch;"*) ;;
|
||||
*) CUDA_ARCHS="${CUDA_ARCHS:+$CUDA_ARCHS;}$_arch" ;;
|
||||
esac
|
||||
fi
|
||||
done <<< "$_raw_caps"
|
||||
# Resolve the arch list before committing to a CUDA build;
|
||||
# an empty list means CPU instead of a PTX-only binary (#5854).
|
||||
_raw_caps=""
|
||||
# Resolve nvidia-smi as _setup_has_usable_nvidia_gpu does
|
||||
# (PATH, then /usr/bin); `command -v` alone would miss an
|
||||
# off-PATH binary and wrongly drop a CUDA host to CPU.
|
||||
_smi_bin=""
|
||||
if command -v nvidia-smi >/dev/null 2>&1; then
|
||||
_smi_bin="nvidia-smi"
|
||||
elif [ -x "/usr/bin/nvidia-smi" ]; then
|
||||
_smi_bin="/usr/bin/nvidia-smi"
|
||||
fi
|
||||
if [ -n "$_smi_bin" ]; then
|
||||
_raw_caps=$(_setup_run_smi "$_smi_bin" --query-gpu=compute_cap --format=csv,noheader 2>/dev/null || true)
|
||||
fi
|
||||
CUDA_ARCHS="$(_resolve_cuda_archs "$_raw_caps" "${UNSLOTH_LLAMA_CUDA_ARCHS:-}")"
|
||||
|
||||
if [ -n "$CUDA_ARCHS" ]; then
|
||||
CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCHS}"
|
||||
CMAKE_ARGS="$CMAKE_ARGS -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCHS}"
|
||||
CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_FLAGS=--threads=0"
|
||||
_BUILD_DESC="building (CUDA, sm_${CUDA_ARCHS//;/+sm_})"
|
||||
|
||||
# Allow a host gcc/clang newer than nvcc's whitelist (else a fresh
|
||||
# toolkit aborts with "unsupported GNU version"); via env to avoid word-splitting.
|
||||
export NVCC_PREPEND_FLAGS="${NVCC_PREPEND_FLAGS:+$NVCC_PREPEND_FLAGS }-allow-unsupported-compiler"
|
||||
else
|
||||
_BUILD_DESC="building (CUDA)"
|
||||
# No detectable arch: build CPU (CMAKE_ARGS has no
|
||||
# -DGGML_CUDA=ON yet, so clearing GPU_BACKEND yields CPU).
|
||||
substep "could not detect a CUDA compute capability; building CPU llama.cpp instead of a PTX-only binary (set UNSLOTH_LLAMA_CUDA_ARCHS, e.g. \"120\", to force a CUDA build)." "$C_WARN"
|
||||
GPU_BACKEND=""
|
||||
_BUILD_DESC="building (CPU, CUDA arch undetectable)"
|
||||
fi
|
||||
|
||||
CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_FLAGS=--threads=0"
|
||||
|
||||
# Allow a host gcc/clang newer than nvcc's whitelist (else a fresh
|
||||
# toolkit aborts with "unsupported GNU version"); via env to avoid word-splitting.
|
||||
export NVCC_PREPEND_FLAGS="${NVCC_PREPEND_FLAGS:+$NVCC_PREPEND_FLAGS }-allow-unsupported-compiler"
|
||||
fi
|
||||
fi
|
||||
elif [ "$GPU_BACKEND" = "rocm" ]; then
|
||||
|
|
|
|||
|
|
@ -11,6 +11,7 @@ sh "$TESTS_DIR/sh/test_get_torch_index_url.sh"
|
|||
sh "$TESTS_DIR/sh/test_mac_intel_compat.sh"
|
||||
sh "$TESTS_DIR/sh/test_torch_constraint.sh"
|
||||
sh "$TESTS_DIR/sh/test_nvcc_meets_llama_minimum.sh"
|
||||
sh "$TESTS_DIR/sh/test_resolve_cuda_archs.sh"
|
||||
sh "$TESTS_DIR/sh/test_strixhalo_wsl_reroute.sh"
|
||||
sh "$TESTS_DIR/sh/test_uninstall_shared_icon.sh"
|
||||
sh "$TESTS_DIR/sh/test_torch_flavor.sh"
|
||||
|
|
|
|||
68
tests/sh/test_resolve_cuda_archs.sh
Executable file
68
tests/sh/test_resolve_cuda_archs.sh
Executable file
|
|
@ -0,0 +1,68 @@
|
|||
#!/bin/bash
|
||||
# Unit tests for _resolve_cuda_archs() from studio/setup.sh (#5854).
|
||||
# Turns nvidia-smi compute_cap text into a deduped ';'-separated arch list;
|
||||
# empty result signals a CPU build, and an explicit override wins.
|
||||
set -e
|
||||
|
||||
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
|
||||
SETUP_SH="$SCRIPT_DIR/../../studio/setup.sh"
|
||||
PASS=0
|
||||
FAIL=0
|
||||
|
||||
# Extract just the helper function (same sed range as the other function tests).
|
||||
_FUNC_FILE=$(mktemp)
|
||||
sed -n '/^_resolve_cuda_archs()/,/^}/p' "$SETUP_SH" > "$_FUNC_FILE"
|
||||
|
||||
assert_eq() {
|
||||
_label="$1"; _expected="$2"; _actual="$3"
|
||||
if [ "$_actual" = "$_expected" ]; then
|
||||
echo " PASS: $_label"
|
||||
PASS=$((PASS + 1))
|
||||
else
|
||||
echo " FAIL: $_label (expected '$_expected', got '$_actual')"
|
||||
FAIL=$((FAIL + 1))
|
||||
fi
|
||||
}
|
||||
|
||||
# $1 = raw compute_cap text, $2 = override
|
||||
run_resolve() {
|
||||
bash -c ". '$_FUNC_FILE'; _resolve_cuda_archs \"\$1\" \"\$2\"" _ "$1" "$2"
|
||||
}
|
||||
|
||||
echo "=== test_resolve_cuda_archs ==="
|
||||
|
||||
# 1) Single GPU -> single arch.
|
||||
assert_eq "single 8.6" "86" "$(run_resolve "8.6" "")"
|
||||
|
||||
# 2) Two distinct GPUs -> both archs, order preserved.
|
||||
assert_eq "distinct 8.6 + 9.0" "86;90" "$(run_resolve "$(printf '8.6\n9.0\n')" "")"
|
||||
|
||||
# 3) Duplicate caps (multi-GPU same model) -> deduped.
|
||||
assert_eq "dedup 12.0 x2" "120" "$(run_resolve "$(printf '12.0\n12.0\n')" "")"
|
||||
|
||||
# 4) Empty input -> empty (the CPU-fallback signal; #5854).
|
||||
assert_eq "empty input" "" "$(run_resolve "" "")"
|
||||
|
||||
# 5) Garbage / N/A lines are ignored -> empty.
|
||||
assert_eq "garbage N/A" "" "$(run_resolve "$(printf 'N/A\n[Not Supported]\n')" "")"
|
||||
|
||||
# 6) Mixed valid + junk -> only the valid caps survive.
|
||||
assert_eq "mixed valid+junk" "86;90" "$(run_resolve "$(printf '8.6\nfoo\n9.0\n')" "")"
|
||||
|
||||
# 7) Whitespace / CR around a cap is stripped.
|
||||
assert_eq "whitespace stripped" "86" "$(run_resolve "$(printf ' 8.6 \r\n')" "")"
|
||||
|
||||
# 8) Override wins verbatim, ignoring detection.
|
||||
assert_eq "override wins" "120" "$(run_resolve "8.6" "120")"
|
||||
|
||||
# 9) Override works even with no detected caps.
|
||||
assert_eq "override no detection" "86;90" "$(run_resolve "" "86;90")"
|
||||
|
||||
# 10) Future arch (compute 10.0 -> 100) parses.
|
||||
assert_eq "future 10.0" "100" "$(run_resolve "10.0" "")"
|
||||
|
||||
rm -f "$_FUNC_FILE"
|
||||
|
||||
echo ""
|
||||
echo "Results: $PASS passed, $FAIL failed"
|
||||
[ "$FAIL" -eq 0 ] || exit 1
|
||||
|
|
@ -249,7 +249,20 @@ class TestSetupShHardening:
|
|||
), "ROCm toolkit search must require a detected AMD GPU, not just hipcc"
|
||||
|
||||
def test_compute_cap_probe_timeout_wrapped(self, setup_src):
|
||||
assert "_setup_run_smi nvidia-smi --query-gpu=compute_cap" in setup_src
|
||||
# nvidia-smi is now a variable ($_smi_bin), so check the wrapper precedes
|
||||
# the probe rather than matching a literal. The string also appears in a
|
||||
# comment, so scan all occurrences and accept if any is wrapped.
|
||||
wrapped = False
|
||||
start = 0
|
||||
while True:
|
||||
idx = setup_src.find("--query-gpu=compute_cap", start)
|
||||
if idx < 0:
|
||||
break
|
||||
if "_setup_run_smi" in setup_src[max(0, idx - 80) : idx]:
|
||||
wrapped = True
|
||||
break
|
||||
start = idx + 1
|
||||
assert wrapped, "compute_cap probe must be wrapped in _setup_run_smi (timeout-bounded)"
|
||||
|
||||
def test_driver_version_probe_timeout_wrapped(self, setup_src):
|
||||
start = setup_src.find("_cuda_driver_max_version()")
|
||||
|
|
|
|||
|
|
@ -728,13 +728,16 @@ class TestSourceCodePatterns:
|
|||
assert all(
|
||||
"-allow-unsupported-compiler" not in line for line in cmake_args_lines
|
||||
), "flag must not be pushed into the $CmakeArgs array"
|
||||
# Must be scoped to the CUDA branch, not set for CPU-only builds.
|
||||
# Must be scoped to the CUDA-on branch, not set for CPU-only builds. The
|
||||
# branch also has an early GGML_CUDA=OFF (undetectable-arch CPU fallback,
|
||||
# #5854), so anchor on GGML_CUDA=ON and the final (no-GPU) GGML_CUDA=OFF.
|
||||
flag_idx = content.index("-allow-unsupported-compiler")
|
||||
cuda_guard_idx = content.index("if ($HasNvidiaSmi -and $NvccPath)")
|
||||
cuda_disable_idx = content.index("'-DGGML_CUDA=OFF'")
|
||||
assert cuda_guard_idx < flag_idx < cuda_disable_idx, (
|
||||
"NVCC_PREPEND_FLAGS must be set inside the CUDA-on branch, "
|
||||
"before the GGML_CUDA=OFF (CPU) branch"
|
||||
cuda_on_idx = content.index("'-DGGML_CUDA=ON'")
|
||||
cpu_else_idx = content.rindex("'-DGGML_CUDA=OFF'")
|
||||
assert cuda_guard_idx < cuda_on_idx < flag_idx < cpu_else_idx, (
|
||||
"NVCC_PREPEND_FLAGS must be set inside the CUDA-on branch, after "
|
||||
"-DGGML_CUDA=ON and before the final CPU GGML_CUDA=OFF branch"
|
||||
)
|
||||
|
||||
def test_macos_arm64_cpu_fallback_args_exclude_rpath(self):
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue