From 6a69af3bf151cc473317928e57e118def0d34e86 Mon Sep 17 00:00:00 2001 From: catlog22 Date: Mon, 22 Dec 2025 17:55:05 +0800 Subject: [PATCH] =?UTF-8?q?feat:=20=E6=9B=B4=E6=96=B0=E5=B5=8C=E5=85=A5?= =?UTF-8?q?=E6=89=B9=E5=A4=84=E7=90=86=E5=A4=A7=E5=B0=8F=E8=87=B3=20256?= =?UTF-8?q?=EF=BC=8C=E4=BB=A5=E4=BC=98=E5=8C=96=E6=80=A7=E8=83=BD=E5=B9=B6?= =?UTF-8?q?=E6=8F=90=E9=AB=98=20GPU=20=E5=8A=A0=E9=80=9F=E6=95=88=E7=8E=87?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- codex-lens/src/codexlens/cli/embedding_manager.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/codex-lens/src/codexlens/cli/embedding_manager.py b/codex-lens/src/codexlens/cli/embedding_manager.py index 514fe74e..68797688 100644 --- a/codex-lens/src/codexlens/cli/embedding_manager.py +++ b/codex-lens/src/codexlens/cli/embedding_manager.py @@ -20,8 +20,8 @@ except ImportError: logger = logging.getLogger(__name__) # Embedding batch size - larger values improve throughput on modern hardware -# Default 64 balances memory usage and GPU/CPU utilization -EMBEDDING_BATCH_SIZE = 64 # Increased from 8 for better performance +# Benchmark: 256 gives ~2.35x speedup over 64 with DirectML GPU acceleration +EMBEDDING_BATCH_SIZE = 256 # Optimized from 64 based on batch size benchmarks def _generate_chunks_from_cursor( @@ -275,7 +275,7 @@ def generate_embeddings( total_chunks_created = 0 total_files_processed = 0 FILE_BATCH_SIZE = 100 # Process 100 files at a time - # EMBEDDING_BATCH_SIZE is defined at module level (default: 64) + # EMBEDDING_BATCH_SIZE is defined at module level (default: 256) try: with VectorStore(index_path) as vector_store: