[CI] Split LM Eval TurboQuant KV Cache into per-config jobs (#57113)
Signed-off-by: Thang Nguyen <thangnguyenvn647@gmail.com> Co-authored-by: Kimi <noreply@moonshot.ai>
Thang Nguyen committed
Sep 22, 2026 at 17:49 UTC
4ec0b95c447fe70b92b2d12164c31255bb178e21
6 files changed
+44
-8
.buildkite/test_areas/lm_eval.yaml
+40
-4
@@ -587,9 +587,45 @@ steps:
587
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-act-fp8.txt
588
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-act-int8.txt
589
590
-- label: ":nvidia: (H200 MIG 18GB) LM Eval TurboQuant KV Cache"
591
- key: lm-eval-turboquant-kv-cache
592
- timeout_in_minutes: 55
590
+- label: ":nvidia: (H200 MIG 18GB) LM Eval TurboQuant k8v4"
591
+ key: lm-eval-turboquant-k8v4
592
+ timeout_in_minutes: 30
593
+ device: h200_18gb
594
+ source_file_dependencies:
595
+ - vllm/model_executor/layers/quantization/turboquant/
596
+ - vllm/v1/attention/backends/turboquant_attn.py
597
+ - vllm/v1/attention/ops/triton_turboquant_decode.py
598
+ - vllm/v1/attention/ops/triton_turboquant_store.py
599
+ commands:
600
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-k8v4.txt
601
+
602
+- label: ":nvidia: (H200 MIG 18GB) LM Eval TurboQuant t4nc"
603
+ key: lm-eval-turboquant-t4nc
604
+ timeout_in_minutes: 30
605
+ device: h200_18gb
606
+ source_file_dependencies:
607
+ - vllm/model_executor/layers/quantization/turboquant/
608
+ - vllm/v1/attention/backends/turboquant_attn.py
609
+ - vllm/v1/attention/ops/triton_turboquant_decode.py
610
+ - vllm/v1/attention/ops/triton_turboquant_store.py
611
+ commands:
612
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-t4nc.txt
613
+
614
+- label: ":nvidia: (H200 MIG 18GB) LM Eval TurboQuant k3v4nc"
615
+ key: lm-eval-turboquant-k3v4nc
616
+ timeout_in_minutes: 30
617
+ device: h200_18gb
618
+ source_file_dependencies:
619
+ - vllm/model_executor/layers/quantization/turboquant/
620
+ - vllm/v1/attention/backends/turboquant_attn.py
621
+ - vllm/v1/attention/ops/triton_turboquant_decode.py
622
+ - vllm/v1/attention/ops/triton_turboquant_store.py
623
+ commands:
624
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-k3v4nc.txt
625
+
626
+- label: ":nvidia: (H200 MIG 18GB) LM Eval TurboQuant t3nc"
627
+ key: lm-eval-turboquant-t3nc
628
+ timeout_in_minutes: 30
629
device: h200_18gb
630
source_file_dependencies:
631
- vllm/model_executor/layers/quantization/turboquant/
@@ -597,7 +633,7 @@ steps:
633
- vllm/v1/attention/ops/triton_turboquant_decode.py
634
- vllm/v1/attention/ops/triton_turboquant_store.py
635
commands:
600
- - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant.txt
636
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-t3nc.txt
637
638
- label: ":nvidia: (H100) GPQA Eval (GPT-OSS)"
639
key: gpqa-eval-gpt-oss-2xh100
tests/evals/gsm8k/configs/models-turboquant-k3v4nc.txt
new
+1
@@ -0,0 +1 @@
1
+Qwen3-4B-TQ-k3v4nc.yaml
tests/evals/gsm8k/configs/models-turboquant-k8v4.txt
new
+1
@@ -0,0 +1 @@
1
+Qwen3-4B-TQ-k8v4.yaml
tests/evals/gsm8k/configs/models-turboquant-t3nc.txt
new
+1
@@ -0,0 +1 @@
1
+Qwen3-4B-TQ-t3nc.yaml
tests/evals/gsm8k/configs/models-turboquant-t4nc.txt
new
+1
@@ -0,0 +1 @@
1
+Qwen3-4B-TQ-t4nc.yaml
tests/evals/gsm8k/configs/models-turboquant.txt
deleted
-4
@@ -1,4 +0,0 @@
1
-Qwen3-4B-TQ-k8v4.yaml
2
-Qwen3-4B-TQ-t4nc.yaml
3
-Qwen3-4B-TQ-k3v4nc.yaml
4
-Qwen3-4B-TQ-t3nc.yaml