update

Author: mengfei25

.github/ci_expected_accuracy/check_expected.py

@@ -31,7 +31,7 @@ def get_test_result(data, suite, dtype, mode, model):
         data: JSON data containing test results
         suite: Test suite name
         dtype: Data type
-        mode: Inference or training mode  
+        mode: Inference or training mode
         model: Model name
 
     Returns:
@@ -91,7 +91,7 @@ def update_reference_dataframe(refer_data, model_name, dtype, accuracy):
         refer_data.loc[mask, dtype] = accuracy
     else:
         new_row = {'name': model_name, dtype: accuracy}
-        refer_data = pd.concat([refer_data, pd.DataFrame([new_row])], 
+        refer_data = pd.concat([refer_data, pd.DataFrame([new_row])],
                               ignore_index=True)
     return refer_data
 
@@ -155,13 +155,13 @@ def main():
         description="Accuracy Check",
         formatter_class=argparse.ArgumentDefaultsHelpFormatter
     )
-    parser.add_argument("--driver", type=str, default="rolling", 
+    parser.add_argument("--driver", type=str, default="rolling",
                        help="rolling or lts")
-    parser.add_argument("--category", type=str, default="inductor", 
+    parser.add_argument("--category", type=str, default="inductor",
                        help="inductor")
-    parser.add_argument("--suite", type=str, required=True, 
+    parser.add_argument("--suite", type=str, required=True,
                        help="huggingface, timm_models or torchbench")
-    parser.add_argument("--mode", type=str, required=True, 
+    parser.add_argument("--mode", type=str, required=True,
                        help="inference or training")
     parser.add_argument("--dtype", type=str, required=True,
                        help="float32, bfloat16, float16, amp_bf16 or amp_fp16")
@@ -214,7 +214,7 @@ def main():
         known_accuracy = get_test_result(test_known_data, suite, dtype, mode, model_name)
 
         # Debug print (optional)
-        print(f"{model_name}: test={test_accuracy}, ref={refer_accuracy}, known={known_accuracy}")
+        # print(f"{model_name}: test={test_accuracy}, ref={refer_accuracy}, known={known_accuracy}")
 
         # Categorize model and determine if reference needs update
         category, should_update = categorize_model(

.github/scripts/calculate_best_perf.py

@@ -118,7 +118,7 @@ def create_new_row(category, model, value_type, value, args_dict):
     }
 
 
-def update_best_performance(best_data, category, model, value_type, 
+def update_best_performance(best_data, category, model, value_type,
                           new_value, args_dict):
     """
     Update best performance data with new value if better.
@@ -149,7 +149,7 @@ def update_best_performance(best_data, category, model, value_type,
         # Add new row
         new_row = create_new_row(category, model, value_type, new_value, args_dict)
         best_data = pd.concat([
-            best_data, 
+            best_data,
             pd.DataFrame([new_row])
         ], ignore_index=True)
 
@@ -162,25 +162,25 @@ def main():
         description="Get Best Performance",
         formatter_class=argparse.ArgumentDefaultsHelpFormatter
     )
-    parser.add_argument("--best", required=True, 
+    parser.add_argument("--best", required=True,
                        help="Saved best performance file")
-    parser.add_argument("--new", required=True, 
+    parser.add_argument("--new", required=True,
                        help="New performance results directory")
-    parser.add_argument("--device", type=str, 
+    parser.add_argument("--device", type=str,
                        help="Device name, such as PVC1100")
-    parser.add_argument("--os", type=str, 
+    parser.add_argument("--os", type=str,
                        help="OS version, such as Ubuntu 22.04")
-    parser.add_argument("--driver", type=str, 
+    parser.add_argument("--driver", type=str,
                        help="Driver version, such as 25.05.32567")
-    parser.add_argument("--oneapi", type=str, 
+    parser.add_argument("--oneapi", type=str,
                        help="OneAPI version, such as 2025.1")
-    parser.add_argument("--gcc", type=str, 
+    parser.add_argument("--gcc", type=str,
                        help="GCC version, such as 11")
-    parser.add_argument("--python", type=str, 
+    parser.add_argument("--python", type=str,
                        help="Python version, such as 3.10")
-    parser.add_argument("--pytorch", type=str, 
+    parser.add_argument("--pytorch", type=str,
                        help="PyTorch version")
-    parser.add_argument("--torch-xpu-ops", type=str, 
+    parser.add_argument("--torch-xpu-ops", type=str,
                        help="Torch XPU Ops version")
 
     args = parser.parse_args()
@@ -223,7 +223,7 @@ def main():
 
     for new_file in new_files:
         category = multiple_replace(new_file)
-        print(f"Processing category: {category}")
+        # print(f"Processing category: {category}")
 
         try:
             new_data = pd.read_csv(new_file)
@@ -235,7 +235,7 @@ def main():
                 # Process eager performance
                 eager_perf = row["abs_latency"] * row["speedup"]
                 best_data = update_best_performance(
-                    best_data, category, model_name, "eager", 
+                    best_data, category, model_name, "eager",
                     eager_perf, system_config
                 )
 

.github/scripts/e2e_summary.sh

@@ -5,7 +5,8 @@ set -euo pipefail
 # Script: test_results_processor.sh
 # Description: Process accuracy and performance test results for XPU operations
 
-readonly SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+readonly script_path="$(dirname "${BASH_SOURCE[0]}")"
+readonly SCRIPT_DIR="$(cd "$script_path" && pwd)"
 readonly SCRIPT_NAME="$(basename "$0")"
 
 # Constants
@@ -121,22 +122,22 @@ parse_test_results() {
         total = passed = pass_rate = failed = xfail = timeout = 0
         new_passed = new_enabled = not_run = 0
     }
-    /Total/ { total = $3 }
-    /Passed/ { passed = $3 }
-    /Pass rate/ { pass_rate = $3 }
-    /Real failed/ { failed = format_count($4, "🔴") }
-    /Expected failed/ { xfail = format_count($4, "🔵") }
-    /timeout/ { timeout = format_count($4, "🟡") }
-    /Failed to passed/ { new_passed = format_count($5, "🟢") }
-    /Not run/ { not_run = $4 }
-    /New models/ { new_enabled = format_count($3, "🔵") }
+    /Total models:/ { total = $3 }
+    /Passed models:/ { passed = $3 }
+    /Pass rate:/ { pass_rate = $3 }
+    /Real failed models:/ { failed = format_count($4, "🔴") }
+    /Expected failed models:/ { xfail = format_count($4, "🔵") }
+    /Warning timeout models:/ { timeout = format_count($4, "🟡") }
+    /Failed to passed models:/ { new_passed = format_count($5, "🟢") }
+    /Not run.in models:/ { not_run = $4 }
+    /New models:/ { new_enabled = format_count($3, "🔵") }
 
     function format_count(count, icon) {
         return count > 0 ? icon count : count
     }
 
     END {
-        printf "%s | %s | %s | %s | %s | %s | %s | %s | %s", 
+        printf "%s | %s | %s | %s | %s | %s | %s | %s | %s",
             total, passed, pass_rate, failed, xfail, timeout, new_passed, new_enabled, not_run
     }'
 }
@@ -241,10 +242,10 @@ get_model_result() {
 determine_color() {
     local tmp_file="$1" model="$2"
     grep -w "$model" "$tmp_file" | awk '
-        /Real failed/ { print "🔴"; exit }
-        /Expected failed|New models/ { print "🔵"; exit }
-        /Warning timeout/ { print "🟡"; exit }
-        /Failed to passed/ { print "🟢"; exit }
+        /Real failed models:/ { print "🔴"; exit }
+        /Expected failed models:|New models:/ { print "🔵"; exit }
+        /Warning timeout models:/ { print "🟡"; exit }
+        /Failed to passed models:/ { print "🟢"; exit }
         { print "black" }
     ' | head -1
 }
@@ -329,7 +330,7 @@ generate_header() {
 
 #### Note:
 🔴: Failed cases needing investigation
-🟢: New passed cases needing reference update  
+🟢: New passed cases needing reference update
 🔵: Expected failed or new enabled cases
 🟡: Warning cases
 Empty: Cases not run

.github/scripts/get_issue.py

@@ -99,7 +99,7 @@ def main():
     # Time the operation
     issues_data = get_github_issues(
         repo_owner=args.repo_owner,
-        repo_name=args.repo_name, 
+        repo_name=args.repo_name,
         labels=args.labels,
         state=args.state
     )

.github/scripts/lintrunner.sh

@@ -2,13 +2,14 @@
 set -ex
 
 # Creat a venv for lint check
-python3 -m venv lint
+curl -LsSf https://astral.sh/uv/install.sh | sh
+source $HOME/.local/bin/env
+uv venv lint --python 3.12 --clear
 source lint/bin/activate
-python3 -m pip install -U pip setuptools wheel
+uv pip install -U pip setuptools wheel
 
 # Use uv to speed up lintrunner init
-python3 -m pip install uv==0.1.45
-python3 -m pip install ruamel.yaml
+uv pip install ruamel.yaml
 
 CACHE_DIRECTORY="/tmp/.lintbin"
 # Try to recover the cached binaries
@@ -20,7 +21,7 @@ fi
 
 # if lintrunner is not installed, install it
 if ! command -v lintrunner &> /dev/null; then
-    python3 -m pip install lintrunner==0.12.7
+    uv pip install lintrunner
 fi
 
 # Ignoring errors in one specific run
@@ -37,8 +38,8 @@ if [[ "${CLANG}" == "1" ]]; then
         echo "Please run the checker under pytorch source code folder"
     fi
 fi
-#python3 -m tools.generate_torch_version --is_debug=false
-#python3 -m tools.pyi.gen_pyi \
+#uv tools.generate_torch_version --is_debug=false
+#uv tools.pyi.gen_pyi \
 #    --native-functions-path aten/src/ATen/native/native_functions.yaml \
 #    --tags-path aten/src/ATen/native/tags.yaml \
 #    --deprecated-functions-path "tools/autograd/deprecated.yaml"