[Kernel] Turn off CUTLASS scaled_mm for Ada Lovelace (#6384)

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-FP8-compressed-tensors.yaml

-# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test -b 32 -l 250 -f 5 -t 1
+# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test -b 32 -l 1000 -f 5 -t 1
 model_name: "nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test"
 tasks:
 - name: "gsm8k"
   metrics:
   - name: "exact_match,strict-match"
-    value: 0.752
+    value: 0.755
   - name: "exact_match,flexible-extract"
-    value: 0.752
-limit: 250
+    value: 0.755
+limit: 1000
 num_fewshot: 5

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-FP8.yaml

@@ -4,8 +4,8 @@ tasks:
 - name: "gsm8k"
   metrics:
   - name: "exact_match,strict-match"
-    value: 0.756
+    value: 0.753
   - name: "exact_match,flexible-extract"
-    value: 0.752
-limit: 250
+    value: 0.753
+limit: 1000
 num_fewshot: 5

csrc/quantization/cutlass_w8a8/scaled_mm_entry.cu

@@ -38,7 +38,13 @@ bool cutlass_scaled_mm_supports_fp8(int64_t cuda_device_capability) {
   if (cuda_device_capability >= 90) {
     return CUDA_VERSION >= 12000;
   } else if (cuda_device_capability >= 89) {
-    return CUDA_VERSION >= 12040;
+    // CUTLASS Kernels have not been tuned for Ada Lovelace systems
+    // and are slower than torch.mm. Return false unconditionally in this case.
+    return false;
+
+    // Once the CUTLASS kernels have been optimized for Lovelace systems,
+    // use the following check:
+    // return CUDA_VERSION >= 12040;
   }
 #endif