Updated dl2q edge configs

ctuning · Feb 23, 2024 · f4f932a · f4f932a
1 parent 9e09620
commit f4f932a
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 6 deletions.
diff --git a/cm-mlops/script/compile-model-for.qaic/_cm.json b/cm-mlops/script/compile-model-for.qaic/_cm.json
@@ -305,12 +305,12 @@
     },
     "bert-99.9,server": {
       "env": {
-        "CM_QAIC_MODEL_COMPILER_ARGS_SUT": "-aic-num-cores=4"
+        "CM_QAIC_MODEL_COMPILER_ARGS_SUT": "-aic-num-cores=2"
       }
     },
     "bert-99.9,server,nsp.14": {
       "env": {
-        "CM_QAIC_MODEL_COMPILER_ARGS_SUT": "-aic-num-cores=4"
+        "CM_QAIC_MODEL_COMPILER_ARGS_SUT": "-aic-num-cores=2"
       }
     },
     "tf": {

diff --git a/cm-mlops/script/generate-mlperf-inference-submission/customize.py b/cm-mlops/script/generate-mlperf-inference-submission/customize.py
@@ -59,7 +59,7 @@ def generate_submission(i):
         system_meta['division'] = env['CM_MLPERF_SUBMISSION_DIVISION']
 
     if 'CM_MLPERF_SUBMISSION_CATEGORY' in env:
-        system_meta['system_type'] = env['CM_MLPERF_SUBMISSION_CATEGORY']
+        system_meta['system_type'] = env['CM_MLPERF_SUBMISSION_CATEGORY'].replace("-", ",")
 
     duplicate= (env.get('CM_MLPERF_DUPLICATE_SCENARIO_RESULTS', 'no') in ["yes", "True"])
 

diff --git a/cm-mlops/script/reproduce-mlperf-inference-qualcomm/_cm.yaml b/cm-mlops/script/reproduce-mlperf-inference-qualcomm/_cm.yaml
@@ -484,15 +484,15 @@ variations:
     env:
       CM_MODEL: bert-99
       CM_SQUAD_ACCURACY_DTYPE: float32
-      CM_NOT_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://zenodo.org/record/3750364/files/bert_large_v1_1_fake_quant.onnx"
+      CM_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://zenodo.org/record/3750364/files/bert_large_v1_1_fake_quant.onnx"
 
   bert-99.9:
     group: model
     base:
     - bert_
     env:
       CM_MODEL: bert-99.9
-      CM_NOT_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://zenodo.org/record/3733910/files/model.onnx"
+      CM_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://zenodo.org/record/3733910/files/model.onnx"
 
   loadgen-batch-size.#:
     group: loadgen-batch-size
@@ -611,6 +611,10 @@ variations:
       qaic_activation_count: "3"
 
   dl2q.24xlarge,bert-99.9,offline:
+    env:
+      qaic_activation_count: "7"
+
+  dl2q.24xlarge,bert-99,offline:
     env:
       qaic_activation_count: "14"
 
@@ -626,14 +630,20 @@ variations:
 
   dl2q.24xlarge,bert-99.9,server:
     env:
-      qaic_activation_count: "14"
+      qaic_activation_count: "7"
 
   dl2q.24xlarge,retinanet,server:
     default_variations:
       batch-size: bs.1
     env:
       qaic_activation_count: "14"
 
+  dl2q.24xlarge,resnet50,multistream:
+    default_variations:
+      batch-size: bs.1
+    env:
+      qaic_activation_count: "1"
+
   pro:
     env:
       qaic_queue_length: 10