Groq can use tools and parse JSON

* I ran a bunch of the examples and the only one that failed uses `n` API param
* Groq says n isn't supported so I could either have it make n requests or just leave that error there.

src/ell/providers/groq.py

@@ -4,8 +4,11 @@
 
 import json
 import contextvars
+from typing import List
 
 from pydantic import BaseModel
+import ell
+from ell.provider import EllCallParams
 from ell.providers.openai import OpenAIProvider
 from ell.configurator import register_provider
 from ell.types.message import ContentBlock, Message
@@ -18,22 +21,22 @@
     import groq
     class GroqProvider(OpenAIProvider):
         dangerous_disable_validation = True
-        def translate_to_provider(self, *args, **kwargs):
-            params = super().translate_to_provider(*args, **kwargs)
-            params.pop('stream_options', None)
-            params['messages'] = messages_to_groq_message_format(params['messages'])
-
+        def translate_to_provider(self, ell_call: EllCallParams):
             # assert 'response_format' not in params, 'Groq does not support response_format.'
             # Store the response_format model between to_provider and from_provider
-            response_format = params.get('response_format')
+            response_format = ell_call.api_params.get('response_format')
             store_response_format.set(response_format)
             if isinstance(response_format, type) and issubclass(response_format, BaseModel):
-                # Groq beta JSON response does not support streaming or stop tokens
-                params.pop('stream', None)
-                params.pop('stop', None)
-                params['response_format'] = {'type': 'json_object'}
                 # Groq suggests explain how to respond with JSON in system prompt
-                params['messages'] = add_json_schema_to_system_prompt(response_format, params['messages'])
+                ell_call.messages = add_json_schema_to_system_prompt(response_format, ell_call.messages)
+                # Groq beta JSON response does not support streaming or stop tokens
+                ell_call.api_params.pop('stream', None)
+                ell_call.api_params.pop('stop', None)
+                ell_call.api_params['response_format'] = {'type': 'json_object'}
+
+            params = super().translate_to_provider(ell_call)
+            params.pop('stream_options', None)
+            params['messages'] = messages_to_groq_message_format(params['messages'])
 
             return params
 
@@ -56,15 +59,15 @@ def translate_from_provider(self, *args, **kwargs):
 except ImportError:
     pass
 
-def add_json_schema_to_system_prompt(response_format: BaseModel, messages):
+def add_json_schema_to_system_prompt(response_format: BaseModel, messages: List[Message]) -> List[Message]:
     json_prompt = f'\n\nYou must respond with a JSON object compliant with following JSON schema:\n{json.dumps(response_format.model_json_schema(), indent=4)}'
 
-    system_prompt = next(filter(lambda m: m['role'] == 'system', messages), None)
+    system_prompt = next(filter(lambda m: m.role == 'system', messages), None)
 
     if system_prompt is None:
-        messages = [{'role': 'system', 'content': json_prompt}] + messages
+        messages = [ell.system(content=json_prompt)] + messages
     else:
-        system_prompt['content'] += json_prompt
+        system_prompt.content.append(ContentBlock(text=json_prompt))
 
     return messages
 
@@ -74,8 +77,8 @@ def messages_to_groq_message_format(messages):
     # XXX: Issue #289: groq.BadRequestError: Error code: 400 - {'error': {'message': "'messages.1' : for 'role:assistant' the following must be satisfied[('messages.1.content' : value must be a string)]", 'type': 'invalid_request_error'}}
     new_messages = []
     for message in messages:
-        if message['role'] == 'assistant':
-            # Assistant messages must be strings
+        # Assistant messages must be strings or tool calls
+        if message['role'] == 'assistant' and 'tool_calls' not in message:
             # If content is a list, only one string element is allowed
             if isinstance(message['content'], str):
                 new_messages.append({'role': 'assistant', 'content': message['content']})