catlog22/Claude-Code-Workflow
1
0
Fork 0
mirror of https://github.com/catlog22/Claude-Code-Workflow.git synced 2026-02-13 02:41:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: 更新嵌入批处理大小至 256，以优化性能并提高 GPU 加速效率

Browse Source
This commit is contained in:
catlog22
2025-12-22 17:55:05 +08:00
parent acdfbb4644
commit 6a69af3bf1
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
codex-lens/src/codexlens/cli/embedding_manager.py
View File

@@ -20,8 +20,8 @@ except ImportError:
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
# Embedding batch size - larger values improve throughput on modern hardware # Embedding batch size - larger values improve throughput on modern hardware
# Default 64 balances memory usage and GPU/CPU utilization # Benchmark: 256 gives ~2.35x speedup over 64 with DirectML GPU acceleration
EMBEDDING_BATCH_SIZE = 64 # Increased from 8 for better performance EMBEDDING_BATCH_SIZE = 256 # Optimized from 64 based on batch size benchmarks
def _generate_chunks_from_cursor( def _generate_chunks_from_cursor(
@@ -275,7 +275,7 @@ def generate_embeddings(
total_chunks_created = 0 total_chunks_created = 0
total_files_processed = 0 total_files_processed = 0
FILE_BATCH_SIZE = 100 # Process 100 files at a time 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: try:
with VectorStore(index_path) as vector_store: with VectorStore(index_path) as vector_store: