- The essence of rag is knowledge organization, which is convenient for accurate search again. If there is not enough context, it should be completed in the rag knowledge organization process, such as using tree_sitter in the code to complete the problem of insufficient keywords for insufficient types.
- Setup python env
conda create -n rag-code-sorting-search python=3.11
conda activate rag-code-sorting-search
poetry install
- Ollama run embed model
ollama run nomic-embed-text
- index
$ poetry run python rag_search_code.py build /Users/clojure/Desktop/rag-code-sorting-search
$ ls .promptlycode/code_search_index/
code.index metadata.pkl
- search
$ poetry run python rag_search_code.py search /Users/clojure/Desktop/rag-code-sorting-search "search code"
Rank: 1
Score: 0.0037
File: /Users/clojure/Desktop/1112AIlearn-interview/rag-code-sorting-search/rag_search_code.py
Function: search_code (lines 118-161)
Code with context:
115 | 'context_end_line': context_end
116 | }
117 |
118 | def search_code(query, top_k=5):
119 | """Search for similar code using the built index."""
120 | try:
121 | # Load index and metadata
122 | index = faiss.read_index('code_search_index/code.index')
123 | with open('code_search_index/metadata.pkl', 'rb') as f:
159 |
...
160 | except FileNotFoundError:
161 | return "Index not found. Please build the index first using the build command."
162 |
163 | def main():
164 | parser = argparse.ArgumentParser(description='Code Search Tool')
--------------------------------------------------------------------------------
...