Merge pull request #202 from AI21Labs/chat-model-with-tools-docs-and-…

…response-format

feat: Chat model with tools docs and response format

README.md

@@ -177,7 +177,7 @@ messages = [
 
 chat_completions = client.chat.completions.create(
     messages=messages,
-    model="jamba-instruct-preview",
+    model="jamba-1.5-mini",
 )
 ```
 
@@ -207,7 +207,7 @@ client = AsyncAI21Client(
 async def main():
     response = await client.chat.completions.create(
         messages=messages,
-        model="jamba-instruct-preview",
+        model="jamba-1.5-mini",
     )
 
     print(response)
@@ -227,8 +227,9 @@ A more detailed example can be found [here](examples/studio/chat/chat_completion
 ### Supported Models:
 
 - j2-light
-- j2-mid
-- j2-ultra
+- [j2-ultra](#Chat)
+- [j2-mid](#Completion)
+- [jamba-instruct](#Chat-Completion)
 
 you can read more about the models [here](https://docs.ai21.com/reference/j2-complete-api-ref#jurassic-2-models).
 
@@ -270,6 +271,36 @@ completion_response = client.completion.create(
 )
 ```
 
+### Chat Completion
+
+```python
+from ai21 import AI21Client
+from ai21.models.chat import ChatMessage
+
+system = "You're a support engineer in a SaaS company"
+messages = [
+    ChatMessage(content=system, role="system"),
+    ChatMessage(content="Hello, I need help with a signup process.", role="user"),
+    ChatMessage(content="Hi Alice, I can help you with that. What seems to be the problem?", role="assistant"),
+    ChatMessage(content="I am having trouble signing up for your product with my Google account.", role="user"),
+]
+
+client = AI21Client()
+
+response = client.chat.completions.create(
+    messages=messages,
+    model="jamba-instruct",
+    max_tokens=100,
+    temperature=0.7,
+    top_p=1.0,
+    stop=["\n"],
+)
+
+print(response)
+```
+
+Note that jamba-instruct supports async streaming as well.
+
 </details>
 
 For a more detailed example, see the completion [examples](examples/studio/completion.py).
@@ -290,7 +321,7 @@ client = AI21Client()
 
 response = client.chat.completions.create(
     messages=messages,
-    model="jamba-instruct-preview",
+    model="jamba-instruct",
     stream=True,
 )
 for chunk in response:
@@ -314,7 +345,7 @@ client = AsyncAI21Client()
 async def main():
     response = await client.chat.completions.create(
         messages=messages,
-        model="jamba-instruct-preview",
+        model="jamba-1.5-mini",
         stream=True,
     )
     async for chunk in response:
@@ -700,7 +731,7 @@ messages = [
 ]
 
 response = client.chat.completions.create(
-  model="jamba-instruct",
+  model="jamba-1.5-mini",
   messages=messages,
 )
 ```

ai21/clients/studio/resources/chat/async_chat_completions.py

@@ -6,6 +6,9 @@
 from ai21.clients.studio.resources.chat.base_chat_completions import BaseChatCompletions
 from ai21.models import ChatMessage as J2ChatMessage
 from ai21.models.chat import ChatMessage, ChatCompletionResponse, ChatCompletionChunk
+from ai21.models.chat.document_schema import DocumentSchema
+from ai21.models.chat.response_format import ResponseFormat
+from ai21.models.chat.tool_defintions import ToolDefinition
 from ai21.stream.async_stream import AsyncStream
 from ai21.types import NotGiven, NOT_GIVEN
 
@@ -23,7 +26,10 @@ async def create(
         top_p: float | NotGiven = NOT_GIVEN,
         stop: str | List[str] | NotGiven = NOT_GIVEN,
         n: int | NotGiven = NOT_GIVEN,
-        stream: Optional[Literal[False]] | NotGiven = NOT_GIVEN,
+        stream: Optional[False] | NotGiven = NOT_GIVEN,
+        tools: List[ToolDefinition] | NotGiven = NOT_GIVEN,
+        response_format: ResponseFormat | NotGiven = NOT_GIVEN,
+        documents: List[DocumentSchema] | NotGiven = NOT_GIVEN,
         **kwargs: Any,
     ) -> ChatCompletionResponse:
         pass
@@ -39,6 +45,9 @@ async def create(
         top_p: float | NotGiven = NOT_GIVEN,
         stop: str | List[str] | NotGiven = NOT_GIVEN,
         n: int | NotGiven = NOT_GIVEN,
+        tools: List[ToolDefinition] | NotGiven = NOT_GIVEN,
+        response_format: ResponseFormat | NotGiven = NOT_GIVEN,
+        documents: List[DocumentSchema] | NotGiven = NOT_GIVEN,
         **kwargs: Any,
     ) -> AsyncStream[ChatCompletionChunk]:
         pass
@@ -53,6 +62,9 @@ async def create(
         stop: str | List[str] | NotGiven = NOT_GIVEN,
         n: int | NotGiven = NOT_GIVEN,
         stream: Optional[Literal[False]] | Literal[True] | NotGiven = NOT_GIVEN,
+        tools: List[ToolDefinition] | NotGiven = NOT_GIVEN,
+        response_format: ResponseFormat | NotGiven = NOT_GIVEN,
+        documents: List[DocumentSchema] | NotGiven = NOT_GIVEN,
         **kwargs: Any,
     ) -> ChatCompletionResponse | AsyncStream[ChatCompletionChunk]:
         if any(isinstance(item, J2ChatMessage) for item in messages):
@@ -70,6 +82,9 @@ async def create(
             top_p=top_p,
             n=n,
             stream=stream or False,
+            tools=tools,
+            response_format=response_format,
+            documents=documents,
             **kwargs,
         )
 

ai21/clients/studio/resources/chat/base_chat_completions.py

@@ -5,6 +5,9 @@
 from typing import List, Optional, Union, Any, Dict, Literal
 
 from ai21.models.chat import ChatMessage
+from ai21.models.chat.document_schema import DocumentSchema
+from ai21.models.chat.response_format import ResponseFormat
+from ai21.models.chat.tool_defintions import ToolDefinition
 from ai21.types import NotGiven
 from ai21.utils.typing import remove_not_given
 from ai21.models._pydantic_compatibility import _to_dict
@@ -40,6 +43,9 @@ def _create_body(
         stop: Optional[Union[str, List[str]]] | NotGiven,
         n: Optional[int] | NotGiven,
         stream: Literal[False] | Literal[True] | NotGiven,
+        tools: List[ToolDefinition] | NotGiven,
+        response_format: ResponseFormat | NotGiven,
+        documents: List[DocumentSchema] | NotGiven,
         **kwargs: Any,
     ) -> Dict[str, Any]:
         return remove_not_given(
@@ -52,6 +58,13 @@ def _create_body(
                 "stop": stop,
                 "n": n,
                 "stream": stream,
+                "tools": tools,
+                "response_format": (
+                    _to_dict(response_format) if not isinstance(response_format, NotGiven) else response_format
+                ),
+                "documents": (
+                    [_to_dict(document) for document in documents] if not isinstance(documents, NotGiven) else documents
+                ),
                 **kwargs,
             }
         )

ai21/clients/studio/resources/chat/chat_completions.py

@@ -6,6 +6,9 @@
 from ai21.clients.studio.resources.chat.base_chat_completions import BaseChatCompletions
 from ai21.models import ChatMessage as J2ChatMessage
 from ai21.models.chat import ChatMessage, ChatCompletionResponse, ChatCompletionChunk
+from ai21.models.chat.document_schema import DocumentSchema
+from ai21.models.chat.response_format import ResponseFormat
+from ai21.models.chat.tool_defintions import ToolDefinition
 from ai21.stream.stream import Stream
 from ai21.types import NotGiven, NOT_GIVEN
 
@@ -24,6 +27,9 @@ def create(
         stop: str | List[str] | NotGiven = NOT_GIVEN,
         n: int | NotGiven = NOT_GIVEN,
         stream: Optional[Literal[False]] | NotGiven = NOT_GIVEN,
+        tools: List[ToolDefinition] | NOT_GIVEN = NOT_GIVEN,
+        response_format: ResponseFormat | NOT_GIVEN = NOT_GIVEN,
+        documents: List[DocumentSchema] | NOT_GIVEN = NOT_GIVEN,
         **kwargs: Any,
     ) -> ChatCompletionResponse:
         pass
@@ -39,6 +45,9 @@ def create(
         top_p: float | NotGiven = NOT_GIVEN,
         stop: str | List[str] | NotGiven = NOT_GIVEN,
         n: int | NotGiven = NOT_GIVEN,
+        tools: List[ToolDefinition] | NOT_GIVEN = NOT_GIVEN,
+        response_format: ResponseFormat | NOT_GIVEN = NOT_GIVEN,
+        documents: List[DocumentSchema] | NOT_GIVEN = NOT_GIVEN,
         **kwargs: Any,
     ) -> Stream[ChatCompletionChunk]:
         pass
@@ -53,6 +62,9 @@ def create(
         stop: str | List[str] | NotGiven = NOT_GIVEN,
         n: int | NotGiven = NOT_GIVEN,
         stream: Optional[Literal[False]] | Literal[True] | NotGiven = NOT_GIVEN,
+        tools: List[ToolDefinition] | NotGiven = NOT_GIVEN,
+        response_format: ResponseFormat | NotGiven = NOT_GIVEN,
+        documents: List[DocumentSchema] | NotGiven = NOT_GIVEN,
         **kwargs: Any,
     ) -> ChatCompletionResponse | Stream[ChatCompletionChunk]:
         if any(isinstance(item, J2ChatMessage) for item in messages):
@@ -70,6 +82,9 @@ def create(
             top_p=top_p,
             n=n,
             stream=stream or False,
+            tools=tools,
+            response_format=response_format,
+            documents=documents,
             **kwargs,
         )
 

ai21/http_client/http_client.py

@@ -99,7 +99,8 @@ def execute_http_request(
 
         if response.status_code != httpx.codes.OK:
             _logger.error(
-                f"Calling {method} {self._base_url} failed with a non-200 response code: {response.status_code}"
+                f"Calling {method} {self._base_url} failed with a non-200 "
+                f"response code: {response.status_code} headers: {response.headers}"
             )
             handle_non_success_response(response.status_code, response.text)
 

ai21/models/chat/__init__.py

@@ -2,9 +2,15 @@
 
 from .chat_completion_response import ChatCompletionResponse
 from .chat_completion_response import ChatCompletionResponseChoice
-from .chat_message import ChatMessage
+from .chat_message import ChatMessage, AssistantMessage, ToolMessage, UserMessage, SystemMessage
+from .document_schema import DocumentSchema
+from .function_tool_definition import FunctionToolDefinition
+from .response_format import ResponseFormat
 from .role_type import RoleType as RoleType
 from .chat_completion_chunk import ChatCompletionChunk, ChoicesChunk, ChoiceDelta
+from .tool_call import ToolCall
+from .tool_defintions import ToolDefinition
+from .tool_function import ToolFunction
 
 __all__ = [
     "ChatCompletionResponse",
@@ -14,4 +20,17 @@
     "ChatCompletionChunk",
     "ChoicesChunk",
     "ChoiceDelta",
+    "AssistantMessage",
+    "ToolMessage",
+    "UserMessage",
+    "SystemMessage",
+    "DocumentSchema",
+    "FunctionToolDefinition",
+    "ResponseFormat",
+    "ToolCall",
+    "ToolDefinition",
+    "ToolFunction",
+    "ToolParameters",
 ]
+
+from .tool_parameters import ToolParameters

ai21/models/chat/chat_completion_response.py

@@ -3,12 +3,12 @@
 from ai21.models.ai21_base_model import AI21BaseModel
 from ai21.models.logprobs import Logprobs
 from ai21.models.usage_info import UsageInfo
-from .chat_message import ChatMessage
+from .chat_message import AssistantMessage
 
 
 class ChatCompletionResponseChoice(AI21BaseModel):
     index: int
-    message: ChatMessage
+    message: AssistantMessage
     logprobs: Optional[Logprobs] = None
     finish_reason: Optional[str] = None
 

ai21/models/chat/chat_message.py

@@ -1,8 +1,28 @@
-from __future__ import annotations
+from typing import Literal, List, Optional
 
 from ai21.models.ai21_base_model import AI21BaseModel
+from ai21.models.chat.tool_call import ToolCall
 
 
 class ChatMessage(AI21BaseModel):
     role: str
     content: str
+
+
+class AssistantMessage(ChatMessage):
+    role: Literal["assistant"] = "assistant"
+    tool_calls: Optional[List[ToolCall]] = None
+    content: Optional[str] = None
+
+
+class ToolMessage(ChatMessage):
+    role: Literal["tool"] = "tool"
+    tool_call_id: str
+
+
+class UserMessage(ChatMessage):
+    role: Literal["user"] = "user"
+
+
+class SystemMessage(ChatMessage):
+    role: Literal["system"] = "system"

ai21/models/chat/document_schema.py

@@ -0,0 +1,9 @@
+from typing import Optional, Dict
+
+from ai21.models.ai21_base_model import AI21BaseModel
+
+
+class DocumentSchema(AI21BaseModel):
+    content: str
+    id: Optional[str] = None
+    metadata: Optional[Dict[str, str]] = None

ai21/models/chat/function_tool_definition.py

@@ -0,0 +1,9 @@
+from typing_extensions import TypedDict, Required
+
+from ai21.models.chat.tool_parameters import ToolParameters
+
+
+class FunctionToolDefinition(TypedDict, total=False):
+    name: Required[str]
+    description: str
+    parameters: ToolParameters

ai21/models/chat/response_format.py

@@ -0,0 +1,7 @@
+from typing import Literal
+
+from ai21.models.ai21_base_model import AI21BaseModel
+
+
+class ResponseFormat(AI21BaseModel):
+    type: Literal["text", "json_object"]

ai21/models/chat/role_type.py

@@ -4,3 +4,5 @@
 class RoleType(str, Enum):
     USER = "user"
     ASSISTANT = "assistant"
+    TOOL = "tool"
+    SYSTEM = "system"

ai21/models/chat/tool_call.py

@@ -0,0 +1,10 @@
+from typing import Literal
+
+from ai21.models.ai21_base_model import AI21BaseModel
+from ai21.models.chat.tool_function import ToolFunction
+
+
+class ToolCall(AI21BaseModel):
+    id: str
+    function: ToolFunction
+    type: Literal["function"] = "function"

ai21/models/chat/tool_defintions.py

@@ -0,0 +1,8 @@
+from typing_extensions import Literal, TypedDict, Required
+
+from ai21.models.chat import FunctionToolDefinition
+
+
+class ToolDefinition(TypedDict, total=False):
+    type: Required[Literal["function"]]
+    function: Required[FunctionToolDefinition]

ai21/models/chat/tool_function.py

@@ -0,0 +1,6 @@
+from ai21.models.ai21_base_model import AI21BaseModel
+
+
+class ToolFunction(AI21BaseModel):
+    name: str
+    arguments: str

ai21/models/chat/tool_parameters.py

@@ -0,0 +1,7 @@
+from typing_extensions import Literal, Any, Dict, List, TypedDict, Required
+
+
+class ToolParameters(TypedDict, total=False):
+    type: Literal["object"]
+    properties: Required[Dict[str, Any]]
+    required: List[str]

examples/studio/chat/async_chat_completions.py

@@ -18,7 +18,7 @@
 async def main():
     response = await client.chat.completions.create(
         messages=messages,
-        model="jamba-instruct-preview",
+        model="jamba-1.5-mini",
         max_tokens=100,
         temperature=0.7,
         top_p=1.0,