diff --git a/.buildkite/test-amd.yaml b/.buildkite/test-amd.yaml index 771a177298..326fa3a2ce 100644 --- a/.buildkite/test-amd.yaml +++ b/.buildkite/test-amd.yaml @@ -4126,6 +4126,33 @@ steps: commands: - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-k3v4nc-rocm.txt +- label: ":amd: (MI355) LM Eval TurboQuant t3nc" + timeout_in_minutes: 30 + mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] + dind: false + agent_pool: mi355_1 + num_gpus: 1 + optional: true + working_dir: "/vllm-workspace/tests" + source_file_dependencies: + - .buildkite/test-amd.yaml + - tests/evals/gsm8k/ + - tests/utils.py + - tests/conftest.py + - vllm/model_executor/layers/quantization/turboquant/ + - vllm/v1/attention/backends/turboquant_attn.py + - vllm/v1/attention/backends/triton_attn.py + - vllm/v1/attention/ops/triton_unified_attention.py + - vllm/v1/attention/selector.py + - vllm/v1/attention/ops/triton_turboquant_decode.py + - vllm/v1/attention/ops/triton_turboquant_store.py + - vllm/v1/attention/ops/flydsl_turboquant_decode.py + - vllm/v1/attention/ops/flydsl_kernels/ + - vllm/v1/attention/ops/turboquant_soa/ + - vllm/platforms/rocm.py + commands: + - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-t3nc-rocm.txt + - label: ":amd: (MI355) LM Eval Small Models Distributed" timeout_in_minutes: 55 mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] diff --git a/.buildkite/test_areas/lm_eval.yaml b/.buildkite/test_areas/lm_eval.yaml index 9d27b2e83a..dccf57f2e2 100644 --- a/.buildkite/test_areas/lm_eval.yaml +++ b/.buildkite/test_areas/lm_eval.yaml @@ -766,6 +766,34 @@ steps: - vllm/v1/attention/ops/triton_turboquant_store.py commands: - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-t3nc.txt + mirror: + amd: + label: ":amd: (MI355) LM Eval TurboQuant t3nc" + dind: false + device: mi355_1 + num_devices: 1 + timeout_in_minutes: 30 + working_dir: /vllm-workspace/tests + depends_on: + - image-build-amd + source_file_dependencies: + - .buildkite/test_areas/lm_eval.yaml + - tests/evals/gsm8k/ + - tests/utils.py + - tests/conftest.py + - vllm/model_executor/layers/quantization/turboquant/ + - vllm/v1/attention/backends/turboquant_attn.py + - vllm/v1/attention/backends/triton_attn.py + - vllm/v1/attention/ops/triton_unified_attention.py + - vllm/v1/attention/selector.py + - vllm/v1/attention/ops/triton_turboquant_decode.py + - vllm/v1/attention/ops/triton_turboquant_store.py + - vllm/v1/attention/ops/flydsl_turboquant_decode.py + - vllm/v1/attention/ops/flydsl_kernels/ + - vllm/v1/attention/ops/turboquant_soa/ + - vllm/platforms/rocm.py + commands: + - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-t3nc-rocm.txt - label: ":nvidia: (H100) GPQA Eval (GPT-OSS)" key: gpqa-eval-gpt-oss-2xh100 diff --git a/tests/evals/gsm8k/configs/Qwen3-4B-TQ-t3nc-ROCm.yaml b/tests/evals/gsm8k/configs/Qwen3-4B-TQ-t3nc-ROCm.yaml new file mode 100644 index 0000000000..103b37880a --- /dev/null +++ b/tests/evals/gsm8k/configs/Qwen3-4B-TQ-t3nc-ROCm.yaml @@ -0,0 +1,6 @@ +model_name: "Qwen/Qwen3-4B" +accuracy_threshold: 0.75 +num_questions: 1319 +num_fewshot: 5 +# Triton serves the unquantized boundary layers with TurboQuant-compatible layout. +server_args: "--kv-cache-dtype turboquant_3bit_nc --max-model-len 4096 --attention-backend TRITON_ATTN" diff --git a/tests/evals/gsm8k/configs/models-turboquant-t3nc-rocm.txt b/tests/evals/gsm8k/configs/models-turboquant-t3nc-rocm.txt new file mode 100644 index 0000000000..cdc2462b8c --- /dev/null +++ b/tests/evals/gsm8k/configs/models-turboquant-t3nc-rocm.txt @@ -0,0 +1 @@ +Qwen3-4B-TQ-t3nc-ROCm.yaml