PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
fix: adjust tolerance levels for performance metrics comparison (#2505)
Sun, Xuehao2c377483a9ff
diff --git a/.azure-pipelines/scripts/performance/check_performance.py b/.azure-pipelines/scripts/performance/check_performance.pyindex ce4940f9d..9e8330353 100644--- a/.azure-pipelines/scripts/performance/check_performance.py+++ b/.azure-pipelines/scripts/performance/check_performance.py@@ -106,10 +106,10 @@ def check_performance():if not compare_metric("Tuning Time (s)", current.tuning_time_s, baseline.tuning_time_s, tolerance=0.05):all_passed = False- if not compare_metric("Peak RAM (GB)", current.peak_ram_gb, baseline.peak_ram_gb, tolerance=0.03):+ if not compare_metric("Peak RAM (GB)", current.peak_ram_gb, baseline.peak_ram_gb, tolerance=0.07):all_passed = False- if not compare_metric("Peak VRAM (GB)", current.peak_vram_gb, baseline.peak_vram_gb, tolerance=0.03):+ if not compare_metric("Peak VRAM (GB)", current.peak_vram_gb, baseline.peak_vram_gb, tolerance=0.10):all_passed = Falseif not compare_metric("Output Size (GB)", current.output_size_gb, baseline.output_size_gb, tolerance=0.01):