Skip to content

armgit / LEANN

Public Imported from GitHub
0

[MLsys2026 Best Paper]: https://arxiv.org/abs/2506.08276. RAG on Everything with LEANN. Enjoy 97% storage savings while running a fast, accurate, and 100% private RAG application on your personal device.

Fix prompt template bugs: build template ignored and runtime override not wired (#173)

ww26969f5145645d
diff --git a/packages/leann-core/src/leann/cli.py b/packages/leann-core/src/leann/cli.py
index 982ae9c..122ae6b 100644
--- a/packages/leann-core/src/leann/cli.py
+++ b/packages/leann-core/src/leann/cli.py
@@ -1545,6 +1545,11 @@ Examples:
print("Invalid input. Aborting search.")
return
+ # Build provider_options for runtime override
+ provider_options = {}
+ if args.embedding_prompt_template:
+ provider_options["prompt_template"] = args.embedding_prompt_template
+
searcher = LeannSearcher(index_path=index_path)
results = searcher.search(
query,
@@ -1554,6 +1559,7 @@ Examples:
prune_ratio=args.prune_ratio,
recompute_embeddings=args.recompute_embeddings,
pruning_strategy=args.pruning_strategy,
+ provider_options=provider_options if provider_options else None,
)
print(f"Search results for '{query}' (top {len(results)}):")
diff --git a/packages/leann-core/src/leann/embedding_compute.py b/packages/leann-core/src/leann/embedding_compute.py
index a8dba9d..093a710 100644
--- a/packages/leann-core/src/leann/embedding_compute.py
+++ b/packages/leann-core/src/leann/embedding_compute.py
@@ -740,7 +740,10 @@ def compute_embeddings_openai(
print(f"len of texts: {len(texts)}")
# Apply prompt template if provided
- prompt_template = provider_options.get("prompt_template")
+ # Priority: build_prompt_template (new format) > prompt_template (old format)
+ prompt_template = provider_options.get("build_prompt_template") or provider_options.get(
+ "prompt_template"
+ )
if prompt_template:
logger.warning(f"Applying prompt template: '{prompt_template}'")
@@ -1031,7 +1034,10 @@ def compute_embeddings_ollama(
# Apply prompt template if provided
provider_options = provider_options or {}
- prompt_template = provider_options.get("prompt_template")
+ # Priority: build_prompt_template (new format) > prompt_template (old format)
+ prompt_template = provider_options.get("build_prompt_template") or provider_options.get(
+ "prompt_template"
+ )
if prompt_template:
logger.warning(f"Applying prompt template: '{prompt_template}'")
diff --git a/packages/leann-core/src/leann/searcher_base.py b/packages/leann-core/src/leann/searcher_base.py
index ba5b189..f8ab71c 100644
--- a/packages/leann-core/src/leann/searcher_base.py
+++ b/packages/leann-core/src/leann/searcher_base.py
@@ -71,6 +71,15 @@ class BaseSearcher(LeannBackendSearcherInterface, ABC):
or "mips"
)
+ # Filter out ALL prompt templates from provider_options during search
+ # Templates are applied in compute_query_embedding (line 109-110) BEFORE server call
+ # The server should never apply templates during search to avoid double-templating
+ search_provider_options = {
+ k: v
+ for k, v in self.embedding_options.items()
+ if k not in ("build_prompt_template", "query_prompt_template", "prompt_template")
+ }
+
server_started, actual_port = self.embedding_server_manager.start_server(
port=port,
model_name=self.embedding_model,
@@ -78,7 +87,7 @@ class BaseSearcher(LeannBackendSearcherInterface, ABC):
passages_file=passages_source_file,
distance_metric=distance_metric,
enable_warmup=kwargs.get("enable_warmup", False),
- provider_options=self.embedding_options,
+ provider_options=search_provider_options,
)
if not server_started:
raise RuntimeError(f"Failed to start embedding server on port {actual_port}")