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 8, 16:30 UTC
[CI][torch.compile] Fix incorrect filtering for E2E fusion tests on B200 (#34031)
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Luka Govedič committed
Feb 9, 2026 at 10:05 UTC
781ddf786861f40de6d94d45d7b149d0f8d58c11
1 file changed
+8
-10
.buildkite/test_areas/compile.yaml
+8
-10
@@ -121,13 +121,10 @@ steps:
121
optional: true
122
commands:
123
- nvidia-smi
124
- # Run all models and attn backends but only Inductor partition and native custom ops
125
- # -k "inductor_partition and not +rms_norm and not +quant_fp8"
124
+ # Run all models but only FLASHINFER, Inductor partition and native custom ops
125
# Qwen requires +quant_fp8 as -quant_fp8 rms+quant fusion is not supported
127
- # -k "inductor_partition and not +rms_norm and +quant_fp8 and qwen3"
128
- # Run just llama3 (fp8 & fp4) for all config combinations
129
- # -k "llama-3"
130
- - pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k "inductor_partition and not +rms_norm and not +quant_fp8" -k "inductor_partition and not +rms_norm and +quant_fp8 and qwen3" -k "llama-3"
126
+ # Run just llama3 (fp8 & fp4) for all config combinations (only inductor partition)
127
+ - pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k "inductor_partition and (FLASHINFER and not +rms_norm and (not +quant_fp8 or +quant_fp8 and qwen3) or llama-3)"
128
129
- label: Fusion E2E TP2 Quick (H100)
130
timeout_in_minutes: 20
@@ -162,7 +159,7 @@ steps:
159
- tests/compile/fusions_e2e/
160
commands:
161
- nvidia-smi
165
- # Run just llama3 (fp4 & fp8 & bf16) for all config combinations
162
+ # Run just llama3 (fp8 & bf16) for all config combinations
163
- pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py -k "llama-3"
164
165
- label: Fusion E2E TP2 AsyncTP Config Sweep (H100)
@@ -197,7 +194,8 @@ steps:
194
- tests/compile/fusions_e2e/
195
commands:
196
- nvidia-smi
200
- # Run all models and attn backends but only Inductor partition and native custom ops
197
+ # Run all models but only FLASHINFER, Inductor partition and native custom ops
198
+ # include qwen with +quant_fp8 as -quant_fp8 rms+quant fusion is not supported
199
# for ar-rms-quant-fp4, also sweep llama3
202
- - pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py -k "inductor_partition and not +rms_norm and not +quant_fp8" -k "Llama-3.1-8B-Instruct-FP4"
203
- - pytest -v -s tests/compile/fusions_e2e/test_tp2_async_tp.py -k "inductor_partition and not +rms_norm and not +quant_fp8"
200
+ - pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py -k "(FLASHINFER and inductor_partition and not +rms_norm and (not +quant_fp8 or +quant_fp8 and qwen3)) or Llama-3.1-8B-Instruct-FP4"
201
+ - pytest -v -s tests/compile/fusions_e2e/test_tp2_async_tp.py -k "FLASHINFER and inductor_partition and not +rms_norm and (not +quant_fp8 or +quant_fp8 and qwen3)"