A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 3, 13:00 UTC
[ROCm][CI] Removed problematic command override mechanism (#42807)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
May 16, 2026 at 04:35 UTC
4db300e95fd29f5b1a4a7c34f4fbe91b7e9abb24
1 file changed
+1
-99
.buildkite/scripts/hardware_ci/run-amd-test.sh
+1
-99
@@ -114,8 +114,7 @@ handle_pytest_exit() {
114
# unquoted since they have no spaces and work fine.
115
#
116
# Already-quoted expressions (containing literal single quotes) are passed
117
-# through untouched to avoid double-quoting values injected by
118
-# apply_rocm_test_overrides.
117
+# through untouched to avoid double-quoting well-formed shell fragments.
118
#
119
# NOTE: This ONLY fixes -m/-k flags. It cannot recover arbitrary inner
120
# double-quotes stripped by the calling shell (see header comment).
@@ -247,102 +246,6 @@ re_quote_pytest_markers() {
246
echo "${output% }"
247
}
248
250
-###############################################################################
251
-# ROCm-specific pytest command rewrites
252
-#
253
-# These apply ignore flags and environment overrides for tests that are not
254
-# yet supported or behave differently on ROCm hardware. Kept as a single
255
-# function so new exclusions are easy to add in one place.
256
-###############################################################################
257
-
258
-apply_rocm_test_overrides() {
259
- local cmds="$1"
260
-
261
- # --- Model registry filter ---
262
- if [[ $cmds == *"pytest -v -s models/test_registry.py"* ]]; then
263
- cmds=${cmds//"pytest -v -s models/test_registry.py"/"pytest -v -s models/test_registry.py -k 'not BambaForCausalLM and not GritLM and not Mamba2ForCausalLM and not Zamba2ForCausalLM'"}
264
- fi
265
-
266
- # --- LoRA: disable custom paged attention ---
267
- if [[ $cmds == *"pytest -v -s lora"* ]]; then
268
- cmds=${cmds//"pytest -v -s lora"/"pytest -v -s lora"}
269
- fi
270
-
271
- # --- Kernel ignores ---
272
- if [[ $cmds == *" kernels/core"* ]]; then
273
- cmds="${cmds} \
274
- --ignore=kernels/core/test_fused_quant_layernorm.py \
275
- --ignore=kernels/core/test_permute_cols.py"
276
- fi
277
-
278
- if [[ $cmds == *" kernels/attention"* ]]; then
279
- cmds="${cmds} \
280
- --ignore=kernels/attention/test_attention_selector.py \
281
- --ignore=kernels/attention/test_encoder_decoder_attn.py \
282
- --ignore=kernels/attention/test_flash_attn.py \
283
- --ignore=kernels/attention/test_flashinfer.py \
284
- --ignore=kernels/attention/test_prefix_prefill.py \
285
- --ignore=kernels/attention/test_cascade_flash_attn.py \
286
- --ignore=kernels/attention/test_mha_attn.py \
287
- --ignore=kernels/attention/test_lightning_attn.py \
288
- --ignore=kernels/attention/test_attention.py"
289
- fi
290
-
291
- if [[ $cmds == *" kernels/quantization"* ]]; then
292
- cmds="${cmds} \
293
- --ignore=kernels/quantization/test_int8_quant.py \
294
- --ignore=kernels/quantization/test_machete_mm.py \
295
- --ignore=kernels/quantization/test_block_fp8.py \
296
- --ignore=kernels/quantization/test_block_int8.py \
297
- --ignore=kernels/quantization/test_marlin_gemm.py \
298
- --ignore=kernels/quantization/test_cutlass_scaled_mm.py \
299
- --ignore=kernels/quantization/test_int8_kernel.py"
300
- fi
301
-
302
- if [[ $cmds == *" kernels/mamba"* ]]; then
303
- cmds="${cmds} \
304
- --ignore=kernels/mamba/test_mamba_mixer2.py \
305
- --ignore=kernels/mamba/test_causal_conv1d.py \
306
- --ignore=kernels/mamba/test_mamba_ssm_ssd.py"
307
- fi
308
-
309
- if [[ $cmds == *" kernels/moe"* ]]; then
310
- cmds="${cmds} \
311
- --ignore=kernels/moe/test_moe.py \
312
- --ignore=kernels/moe/test_cutlass_moe.py"
313
- fi
314
-
315
- # --- Entrypoint ignores ---
316
- if [[ $cmds == *" entrypoints/openai "* ]]; then
317
- cmds=${cmds//" entrypoints/openai "/" entrypoints/openai \
318
- --ignore=entrypoints/openai/chat_completion/test_audio.py \
319
- --ignore=entrypoints/openai/completion/test_shutdown.py \
320
- --ignore=entrypoints/openai/test_completion.py \
321
- --ignore=entrypoints/openai/models/test_models.py \
322
- --ignore=entrypoints/openai/test_return_tokens_as_ids.py \
323
- --ignore=entrypoints/openai/chat_completion/test_root_path.py \
324
- --ignore=entrypoints/openai/completion/test_prompt_validation.py "}
325
- fi
326
-
327
- if [[ $cmds == *" entrypoints/serve"* ]]; then
328
- cmds="${cmds} \
329
- --ignore=entrypoints/serve/lora/test_lora_adapters.py"
330
- fi
331
-
332
- if [[ $cmds == *" entrypoints/llm "* ]]; then
333
- cmds=${cmds//" entrypoints/llm "/" entrypoints/llm \
334
- --ignore=entrypoints/llm/test_chat.py \
335
- --ignore=entrypoints/llm/test_accuracy.py \
336
- --ignore=entrypoints/llm/test_init.py \
337
- --ignore=entrypoints/llm/test_prompt_validation.py "}
338
- fi
339
-
340
- # Clean up escaped newlines from --ignore appends
341
- cmds=$(echo "$cmds" | sed 's/ \\ / /g')
342
-
343
- echo "$cmds"
344
-}
345
-
249
###############################################################################
250
# Main
251
###############################################################################
@@ -409,7 +312,6 @@ else
312
echo "Skipping re-quoting for VLLM_TEST_COMMANDS input"
313
fi
314
412
-commands=$(apply_rocm_test_overrides "$commands")
315
echo "Final commands: $commands"
316
317
MYPYTHONPATH=".."