address PR comments

README.md

@@ -153,6 +153,35 @@ which would lead to this in your prompt:
 - C
 ```
 
+### Using structured output (function calling and structured output)
+
+*Structured Outputs is a feature that ensures the model will always generate responses that adhere to your supplied JSON Schema, so you don't need to worry about the model omitting a required key, or hallucinating an invalid enum value.* - OpenAI website
+
+#### How to get it to work
+
+1. Check with the model developer/provider whether the model supports some kind of structured output.
+2. Toggle structured output switch
+3. Select one of the supported structured output methods (a model might support all of them but also none of them):
+   - `None` - no structured output is used (equals to toggle being in off state)
+   - `Function calling` - hacky way of implementing structured outputs before `Response format` was implemented into API 
+   - `Response format` - new way of implementing structured outputs 
+4. Provide JSON schema in `json schema` text input (can be generated from `pydantic` model or `zod` if you use `nodejs`):
+```json
+{
+    "title": "get_delivery_date",
+    "description": "Get the delivery date for a customer's order. Call this whenever you need to know the delivery date, for example when a customer asks 'Where is my package'",
+    "type": "object",
+    "properties": {
+      "order_id": {
+        "type": "string"
+      }
+    },
+    "required": ["order_id"],
+    "additionalProperties": false
+}
+```
+example from OpenAI website (slightly modified). For more examples see [https://json-schema.org/learn/miscellaneous-examples](https://json-schema.org/learn/miscellaneous-examples).
+
 ### Postprocessing the model outputs
 
 When working with LLMs, you would often postprocess the raw generated text. Prompterator

prompterator/main.py

@@ -429,9 +429,11 @@ def set_up_ui_generation():
     )
     selected_model: c.ModelProperties = st.session_state.model
     available_structured_output_settings = selected_model.supports_structured_output
-    structured_output_available = False
-    if len(available_structured_output_settings) > 1:
-        structured_output_available = True
+
+    structured_output_available = (
+        len(set(available_structured_output_settings) - {c.StructuredOutputImplementation.NONE})
+        > 0
+    )
 
     structured_output_enabled = col3.toggle(
         label="Structured output",

prompterator/utils.py

@@ -382,16 +382,26 @@ def format_traceback_for_markdown(text):
     return re.sub(r"\n", "\n\n", text)
 
 
+def normalize_name(name: str):
+    return re.sub("[^a-zA-Z0-9_]", "_", name).lower()
+
+
 def build_function_calling_tooling(json_schema):
     schema = json.loads(json_schema)
     function = schema.copy()
-    function_name = function.pop("title")
+    function_name = normalize_name(function.pop("title"))
+    print(function_name)
+    description = (
+        function.pop("description")
+        if function.get("description", None) is not None
+        else function_name
+    )
     tools = [
         {
             "type": "function",
             "function": {
                 "name": function_name,
-                "description": function.pop("description"),
+                "description": description,
                 "parameters": function,
             },
         }
@@ -402,7 +412,7 @@ def build_function_calling_tooling(json_schema):
 
 def build_response_format(json_schema):
     json_schema = json.loads(json_schema)
-    schema = {"name": json_schema.pop("title"), "schema": json_schema, "strict": True}
+    schema = {"name": normalize_name(json_schema.pop("title")), "schema": json_schema, "strict": True}
     response_format = {"type": "json_schema", "json_schema": schema}
 
     return response_format