benchdnn: input: graph: remove case sdpa-compressed-kv-int8-gs128.json

rewrite sdpa-compressed-kv-int4-gs32.json for it.
oneapi-src · Jan 6, 2025 · 539767a · 539767a
1 parent 94389ba
commit 539767a
Show file tree

Hide file tree

Showing 3 changed files with 2 additions and 550 deletions.
diff --git a/tests/benchdnn/inputs/graph/complex_fusion/harness_mha_all b/tests/benchdnn/inputs/graph/complex_fusion/harness_mha_all
@@ -24,7 +24,6 @@
 --reset --expected-n-partitions=0 --case=complex_fusion/mha/dynamic_quantized_mha-Bert_large-inf-int8-bs1-fake.json
 --reset --case=complex_fusion/mha/sdpa-plain-wo-scale-int8-bs1.json
 --reset --case=complex_fusion/mha/sdpa-compressed-kv-int4-gs32.json
---reset --case=complex_fusion/mha/sdpa-compressed-kv-int8-gs128.json
 --reset --case=complex_fusion/mha/sdpa-compressed-k-int8-gs32.json
 --reset --case=complex_fusion/mha/sdpa-compressed-v-int8-gs32.json
 

diff --git a/tests/benchdnn/inputs/graph/complex_fusion/harness_mha_ci b/tests/benchdnn/inputs/graph/complex_fusion/harness_mha_ci
@@ -21,5 +21,6 @@
 --reset --expected-n-partitions=0 --case=complex_fusion/mha/MHA-starcoder-inf-int8-bs1.json
 --reset --expected-n-partitions=0 --case=complex_fusion/mha/dynamic_quantized_mha-Bert_large-inf-int8-bs1-fake.json
 --reset --case=complex_fusion/mha/sdpa-plain-wo-scale-int8-bs1.json
---reset --case=complex_fusion/mha/sdpa-compressed-kv-int8-gs128.json
 --reset --case=complex_fusion/mha/sdpa-compressed-v-int8-gs32.json
+--reset --case=complex_fusion/mha/sdpa-compressed-kv-int4-gs32.json
+--reset --dt=0:s8+2:s8+6:s8+8:s8 --case=complex_fusion/mha/sdpa-compressed-kv-int4-gs32.json