Skip to content

armgit / letta

Public Imported from GitHub
0

Platform for stateful agents: AI with advanced memory that can learn and self-improve over time.

fix: Remove nltk blocking download and make azure model listing non-blocking (#6214)

Kian Jonesabb13ef3523b
diff --git a/letta/schemas/providers/azure.py b/letta/schemas/providers/azure.py
index 3264c7fdc..f17b91a8f 100644
--- a/letta/schemas/providers/azure.py
+++ b/letta/schemas/providers/azure.py
@@ -106,8 +106,10 @@ class AzureProvider(Provider):
return list(latest_models.values())
async def list_llm_models_async(self) -> list[LLMConfig]:
- # TODO (cliandy): asyncify
- model_list = self.azure_openai_get_deployed_model_list()
+ # Run blocking model list fetch in thread pool to avoid blocking event loop
+ import asyncio
+
+ model_list = await asyncio.to_thread(self.azure_openai_get_deployed_model_list)
# Extract models that support text generation
model_options = [m for m in model_list if m.get("capabilities").get("chat_completion") == True]
diff --git a/letta/server/rest_api/app.py b/letta/server/rest_api/app.py
index 459a07c38..84dda5d50 100644
--- a/letta/server/rest_api/app.py
+++ b/letta/server/rest_api/app.py
@@ -163,6 +163,18 @@ async def lifespan(app_: FastAPI):
except Exception as e:
logger.warning(f"[Worker {worker_id}] Failed to start watchdog: {e}")
+ # Pre-download NLTK data to avoid blocking during requests (fallback if Docker build failed)
+ try:
+ import asyncio
+
+ import nltk
+
+ logger.info(f"[Worker {worker_id}] Checking NLTK data availability...")
+ await asyncio.to_thread(nltk.download, "punkt_tab", quiet=True)
+ logger.info(f"[Worker {worker_id}] NLTK data ready")
+ except Exception as e:
+ logger.warning(f"[Worker {worker_id}] Failed to download NLTK data: {e}")
+
if telemetry_settings.profiler:
try:
import googlecloudprofiler