Merge pull request #88 from ks6088ts-labs/feature/issue-87_gemma3-270m

run gemma3:270m

Author: ks6088ts

.env.template

@@ -20,6 +20,9 @@ AZURE_OPENAI_MODEL_CHAT="gpt-5"
 AZURE_OPENAI_MODEL_EMBEDDING="text-embedding-3-small"
 AZURE_OPENAI_MODEL_REASONING="o4-mini"
 
+## Ollama Settings
+OLLAMA_MODEL_CHAT="gemma3:270m"
+
 # ---------
 # Tools
 # ---------

.github/workflows/labeler.yaml

@@ -8,6 +8,7 @@ jobs:
     permissions:
       contents: read
       pull-requests: write
+      issues: write
     runs-on: ubuntu-latest
     steps:
       - name: labeler action

scripts/ollama_operator.py

@@ -2,10 +2,23 @@
 
 import typer
 from dotenv import load_dotenv
+from langchain_core.messages import HumanMessage
+from pydantic import BaseModel, Field
 
 from template_langgraph.llms.ollamas import OllamaWrapper
 from template_langgraph.loggers import get_logger
 
+
+class Profile(BaseModel):
+    first_name: str = Field(..., description="First name of the user")
+    last_name: str = Field(..., description="Last name of the user")
+    age: int = Field(..., description="Age of the user")
+    origin: str = Field(
+        ...,
+        description="Origin of the user, e.g., country or city",
+    )
+
+
 # Initialize the Typer application
 app = typer.Typer(
     add_completion=False,
@@ -17,9 +30,9 @@
 
 
 @app.command()
-def run(
+def chat(
     query: str = typer.Option(
-        "What is the weather like today?",
+        "Explain the concept of Fourier transform.",
         "--query",
         "-q",
         help="Query to run against the Ollama model",
@@ -30,23 +43,78 @@ def run(
         "-v",
         help="Enable verbose output",
     ),
+    stream: bool = typer.Option(
+        False,
+        "--stream",
+        "-s",
+        help="Enable streaming output",
+    ),
 ):
     # Set up logging
     if verbose:
         logger.setLevel(logging.DEBUG)
 
     logger.info("Running...")
     chat_model = OllamaWrapper().chat_model
-    response = chat_model.invoke(
-        input=query,
-    )
-    logger.debug(
-        response.model_dump_json(
-            indent=2,
-            exclude_none=True,
+
+    if stream:
+        response = ""
+        for chunk in chat_model.stream(
+            input=[
+                HumanMessage(content=query),
+            ],
+        ):
+            print(
+                chunk.content,
+                end="",
+                flush=True,
+            )
+            response += str(chunk.content)
+        logger.info(f"Output: {response}")
+    else:
+        response = chat_model.invoke(
+            input=[
+                HumanMessage(content=query),
+            ],
         )
+        logger.debug(
+            response.model_dump_json(
+                indent=2,
+                exclude_none=True,
+            )
+        )
+        logger.info(f"Output: {response.content}")
+
+
+@app.command()
+def structured_output(
+    query: str = typer.Option(
+        "I'm Taro Okamoto from Japan. 30 years old.",
+        "--query",
+        "-q",
+        help="Query to run against the Ollama model",
+    ),
+    verbose: bool = typer.Option(
+        False,
+        "--verbose",
+        "-v",
+        help="Enable verbose output",
+    ),
+):
+    # Set up logging
+    if verbose:
+        logger.setLevel(logging.DEBUG)
+
+    logger.info("Running...")
+    chat_model = OllamaWrapper().chat_model
+    profile = chat_model.with_structured_output(
+        schema=Profile,
+    ).invoke(
+        input=[
+            HumanMessage(content=query),
+        ],
     )
-    logger.info(f"Output: {response.content}")
+    logger.info(f"Output: {profile.model_dump_json(indent=2, exclude_none=True)}")
 
 
 if __name__ == "__main__":

template_langgraph/llms/ollamas.py

@@ -5,7 +5,7 @@
 
 
 class Settings(BaseSettings):
-    ollama_model_chat: str = "phi3:latest"
+    ollama_model_chat: str = "gemma3:270m"
 
     model_config = SettingsConfigDict(
         env_file=".env",