import json
from operator import itemgetter
from typing import (
Any,
Callable,
Dict,
Iterator,
List,
Literal,
Optional,
Sequence,
Tuple,
Type,
Union,
cast,
)
import requests
from langchain_core.callbacks import (
CallbackManagerForLLMRun,
)
from langchain_core.language_models import LanguageModelInput
from langchain_core.language_models.chat_models import (
BaseChatModel,
generate_from_stream,
)
from langchain_core.messages import (
AIMessage,
AIMessageChunk,
BaseMessage,
BaseMessageChunk,
ChatMessage,
HumanMessage,
SystemMessage,
ToolMessage,
)
from langchain_core.output_parsers import (
JsonOutputParser,
PydanticOutputParser,
)
from langchain_core.output_parsers.base import OutputParserLike
from langchain_core.output_parsers.openai_tools import (
JsonOutputKeyToolsParser,
PydanticToolsParser,
make_invalid_tool_call,
parse_tool_call,
)
from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
from langchain_core.runnables import Runnable, RunnableMap, RunnablePassthrough
from langchain_core.tools import BaseTool
from langchain_core.utils import convert_to_secret_str, get_from_dict_or_env
from langchain_core.utils.function_calling import convert_to_openai_tool
from langchain_core.utils.pydantic import is_basemodel_subclass
from pydantic import BaseModel, Field, SecretStr
from requests import Response
def _convert_message_to_dict(message: BaseMessage) -> Dict[str, Any]:
"""
convert a BaseMessage to a dictionary with Role / content
Args:
message: BaseMessage
Returns:
messages_dict: role / content dict
"""
message_dict: Dict[str, Any] = {}
if isinstance(message, ChatMessage):
message_dict = {"role": message.role, "content": message.content}
elif isinstance(message, SystemMessage):
message_dict = {"role": "system", "content": message.content}
elif isinstance(message, HumanMessage):
message_dict = {"role": "user", "content": message.content}
elif isinstance(message, AIMessage):
message_dict = {"role": "assistant", "content": message.content}
if "tool_calls" in message.additional_kwargs:
message_dict["tool_calls"] = message.additional_kwargs["tool_calls"]
if message_dict["content"] == "":
message_dict["content"] = None
elif isinstance(message, ToolMessage):
message_dict = {
"role": "tool",
"content": message.content,
"tool_call_id": message.tool_call_id,
}
else:
raise TypeError(f"Got unknown type {message}")
return message_dict
def _create_message_dicts(messages: List[BaseMessage]) -> List[Dict[str, Any]]:
"""
Convert a list of BaseMessages to a list of dictionaries with Role / content
Args:
messages: list of BaseMessages
Returns:
messages_dicts: list of role / content dicts
"""
message_dicts = [_convert_message_to_dict(m) for m in messages]
return message_dicts
def _is_pydantic_class(obj: Any) -> bool:
return isinstance(obj, type) and is_basemodel_subclass(obj)
[docs]
class ChatSambaNovaCloud(BaseChatModel):
"""
SambaNova Cloud chat model.
Setup:
To use, you should have the environment variables:
`SAMBANOVA_URL` set with your SambaNova Cloud URL.
`SAMBANOVA_API_KEY` set with your SambaNova Cloud API Key.
http://cloud.sambanova.ai/
Example:
.. code-block:: python
ChatSambaNovaCloud(
sambanova_url = SambaNova cloud endpoint URL,
sambanova_api_key = set with your SambaNova cloud API key,
model = model name,
max_tokens = max number of tokens to generate,
temperature = model temperature,
top_p = model top p,
top_k = model top k,
stream_options = include usage to get generation metrics
)
Key init args — completion params:
model: str
The name of the model to use, e.g., Meta-Llama-3-70B-Instruct.
streaming: bool
Whether to use streaming handler when using non streaming methods
max_tokens: int
max tokens to generate
temperature: float
model temperature
top_p: float
model top p
top_k: int
model top k
stream_options: dict
stream options, include usage to get generation metrics
Key init args — client params:
sambanova_url: str
SambaNova Cloud Url
sambanova_api_key: str
SambaNova Cloud api key
Instantiate:
.. code-block:: python
from langchain_community.chat_models import ChatSambaNovaCloud
chat = ChatSambaNovaCloud(
sambanova_url = SambaNova cloud endpoint URL,
sambanova_api_key = set with your SambaNova cloud API key,
model = model name,
max_tokens = max number of tokens to generate,
temperature = model temperature,
top_p = model top p,
top_k = model top k,
stream_options = include usage to get generation metrics
)
Invoke:
.. code-block:: python
messages = [
SystemMessage(content="your are an AI assistant."),
HumanMessage(content="tell me a joke."),
]
response = chat.invoke(messages)
Stream:
.. code-block:: python
for chunk in chat.stream(messages):
print(chunk.content, end="", flush=True)
Async:
.. code-block:: python
response = chat.ainvoke(messages)
await response
Tool calling:
.. code-block:: python
from pydantic import BaseModel, Field
class GetWeather(BaseModel):
'''Get the current weather in a given location'''
location: str = Field(
...,
description="The city and state, e.g. Los Angeles, CA"
)
llm_with_tools = llm.bind_tools([GetWeather, GetPopulation])
ai_msg = llm_with_tools.invoke("Should I bring my umbrella today in LA?")
ai_msg.tool_calls
.. code-block:: none
[
{
'name': 'GetWeather',
'args': {'location': 'Los Angeles, CA'},
'id': 'call_adf61180ea2b4d228a'
}
]
Structured output:
.. code-block:: python
from typing import Optional
from pydantic import BaseModel, Field
class Joke(BaseModel):
'''Joke to tell user.'''
setup: str = Field(description="The setup of the joke")
punchline: str = Field(description="The punchline to the joke")
structured_model = llm.with_structured_output(Joke)
structured_model.invoke("Tell me a joke about cats")
.. code-block:: python
Joke(setup="Why did the cat join a band?",
punchline="Because it wanted to be the purr-cussionist!")
See `ChatSambanovaCloud.with_structured_output()` for more.
Token usage:
.. code-block:: python
response = chat.invoke(messages)
print(response.response_metadata["usage"]["prompt_tokens"]
print(response.response_metadata["usage"]["total_tokens"]
Response metadata
.. code-block:: python
response = chat.invoke(messages)
print(response.response_metadata)
"""
sambanova_url: str = Field(default="")
"""SambaNova Cloud Url"""
sambanova_api_key: SecretStr = Field(default=SecretStr(""))
"""SambaNova Cloud api key"""
model: str = Field(default="Meta-Llama-3.1-8B-Instruct")
"""The name of the model"""
streaming: bool = Field(default=False)
"""Whether to use streaming handler when using non streaming methods"""
max_tokens: int = Field(default=1024)
"""max tokens to generate"""
temperature: float = Field(default=0.7)
"""model temperature"""
top_p: Optional[float] = Field(default=None)
"""model top p"""
top_k: Optional[int] = Field(default=None)
"""model top k"""
stream_options: Dict[str, Any] = Field(default={"include_usage": True})
"""stream options, include usage to get generation metrics"""
additional_headers: Dict[str, Any] = Field(default={})
"""Additional headers to sent in request"""
[docs]
class Config:
populate_by_name = True
@classmethod
def is_lc_serializable(cls) -> bool:
"""Return whether this model can be serialized by Langchain."""
return False
@property
def lc_secrets(self) -> Dict[str, str]:
return {"sambanova_api_key": "sambanova_api_key"}
@property
def _identifying_params(self) -> Dict[str, Any]:
"""Return a dictionary of identifying parameters.
This information is used by the LangChain callback system, which
is used for tracing purposes make it possible to monitor LLMs.
"""
return {
"model": self.model,
"streaming": self.streaming,
"max_tokens": self.max_tokens,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
"stream_options": self.stream_options,
}
@property
def _llm_type(self) -> str:
"""Get the type of language model used by this chat model."""
return "sambanovacloud-chatmodel"
def __init__(self, **kwargs: Any) -> None:
"""init and validate environment variables"""
kwargs["sambanova_url"] = get_from_dict_or_env(
kwargs,
"sambanova_url",
"SAMBANOVA_URL",
default="https://api.sambanova.ai/v1/chat/completions",
)
kwargs["sambanova_api_key"] = convert_to_secret_str(
get_from_dict_or_env(kwargs, "sambanova_api_key", "SAMBANOVA_API_KEY")
)
super().__init__(**kwargs)
[docs]
def with_structured_output(
self,
schema: Optional[Union[Dict[str, Any], Type[BaseModel]]] = None,
*,
method: Literal[
"function_calling", "json_mode", "json_schema"
] = "function_calling",
include_raw: bool = False,
**kwargs: Any,
) -> Runnable[LanguageModelInput, Union[Dict[str, Any], BaseModel]]:
"""Model wrapper that returns outputs formatted to match the given schema.
Args:
schema:
The output schema. Can be passed in as:
- an OpenAI function/tool schema,
- a JSON Schema,
- a TypedDict class,
- or a Pydantic.BaseModel class.
If `schema` is a Pydantic class then the model output will be a
Pydantic instance of that class, and the model-generated fields will be
validated by the Pydantic class. Otherwise the model output will be a
dict and will not be validated. See :meth:`langchain_core.utils.function_calling.convert_to_openai_tool`
for more on how to properly specify types and descriptions of
schema fields when specifying a Pydantic or TypedDict class.
method:
The method for steering model generation, either "function_calling"
"json_mode" or "json_schema".
If "function_calling" then the schema will be converted
to an OpenAI function and the returned model will make use of the
function-calling API. If "json_mode" or "json_schema" then OpenAI's
JSON mode will be used.
Note that if using "json_mode" or "json_schema" then you must include instructions
for formatting the output into the desired schema into the model call.
include_raw:
If False then only the parsed structured output is returned. If
an error occurs during model output parsing it will be raised. If True
then both the raw model response (a BaseMessage) and the parsed model
response will be returned. If an error occurs during output parsing it
will be caught and returned as well. The final output is always a dict
with keys "raw", "parsed", and "parsing_error".
Returns:
A Runnable that takes same inputs as a :class:`langchain_core.language_models.chat.BaseChatModel`.
If `include_raw` is False and `schema` is a Pydantic class, Runnable outputs
an instance of `schema` (i.e., a Pydantic object).
Otherwise, if `include_raw` is False then Runnable outputs a dict.
If `include_raw` is True, then Runnable outputs a dict with keys:
- `"raw"`: BaseMessage
- `"parsed"`: None if there was a parsing error, otherwise the type depends on the `schema` as described above.
- `"parsing_error"`: Optional[BaseException]
Example: schema=Pydantic class, method="function_calling", include_raw=False:
.. code-block:: python
from typing import Optional
from langchain_community.chat_models import ChatSambaNovaCloud
from pydantic import BaseModel, Field
class AnswerWithJustification(BaseModel):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: str = Field(
description="A justification for the answer."
)
llm = ChatSambaNovaCloud(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(AnswerWithJustification)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> AnswerWithJustification(
# answer='They weigh the same',
# justification='A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same.'
# )
Example: schema=Pydantic class, method="function_calling", include_raw=True:
.. code-block:: python
from langchain_community.chat_models import ChatSambaNovaCloud
from pydantic import BaseModel
class AnswerWithJustification(BaseModel):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: str
llm = ChatSambaNovaCloud(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(
AnswerWithJustification, include_raw=True
)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'raw': AIMessage(content='', additional_kwargs={'tool_calls': [{'function': {'arguments': '{"answer": "They weigh the same.", "justification": "A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount."}', 'name': 'AnswerWithJustification'}, 'id': 'call_17a431fc6a4240e1bd', 'type': 'function'}]}, response_metadata={'finish_reason': 'tool_calls', 'usage': {'acceptance_rate': 5, 'completion_tokens': 53, 'completion_tokens_after_first_per_sec': 343.7964936837758, 'completion_tokens_after_first_per_sec_first_ten': 439.1205661878638, 'completion_tokens_per_sec': 162.8511306784833, 'end_time': 1731527851.0698032, 'is_last_response': True, 'prompt_tokens': 213, 'start_time': 1731527850.7137961, 'time_to_first_token': 0.20475482940673828, 'total_latency': 0.32545061111450196, 'total_tokens': 266, 'total_tokens_per_sec': 817.3283162354066}, 'model_name': 'Meta-Llama-3.1-70B-Instruct', 'system_fingerprint': 'fastcoe', 'created': 1731527850}, id='95667eaf-447f-4b53-bb6e-b6e1094ded88', tool_calls=[{'name': 'AnswerWithJustification', 'args': {'answer': 'They weigh the same.', 'justification': 'A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount.'}, 'id': 'call_17a431fc6a4240e1bd', 'type': 'tool_call'}]),
# 'parsed': AnswerWithJustification(answer='They weigh the same.', justification='A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount.'),
# 'parsing_error': None
# }
Example: schema=TypedDict class, method="function_calling", include_raw=False:
.. code-block:: python
# IMPORTANT: If you are using Python <=3.8, you need to import Annotated
# from typing_extensions, not from typing.
from typing_extensions import Annotated, TypedDict
from langchain_community.chat_models import ChatSambaNovaCloud
class AnswerWithJustification(TypedDict):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: Annotated[
Optional[str], None, "A justification for the answer."
]
llm = ChatSambaNovaCloud(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(AnswerWithJustification)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'answer': 'They weigh the same',
# 'justification': 'A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount.'
# }
Example: schema=OpenAI function schema, method="function_calling", include_raw=False:
.. code-block:: python
from langchain_community.chat_models import ChatSambaNovaCloud
oai_schema = {
'name': 'AnswerWithJustification',
'description': 'An answer to the user question along with justification for the answer.',
'parameters': {
'type': 'object',
'properties': {
'answer': {'type': 'string'},
'justification': {'description': 'A justification for the answer.', 'type': 'string'}
},
'required': ['answer']
}
}
llm = ChatSambaNovaCloud(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(oai_schema)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'answer': 'They weigh the same',
# 'justification': 'A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount.'
# }
Example: schema=Pydantic class, method="json_mode", include_raw=True:
.. code-block::
from langchain_community.chat_models import ChatSambaNovaCloud
from pydantic import BaseModel
class AnswerWithJustification(BaseModel):
answer: str
justification: str
llm = ChatSambaNovaCloud(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(
AnswerWithJustification,
method="json_mode",
include_raw=True
)
structured_llm.invoke(
"Answer the following question. "
"Make sure to return a JSON blob with keys 'answer' and 'justification'.\n\n"
"What's heavier a pound of bricks or a pound of feathers?"
)
# -> {
# 'raw': AIMessage(content='{\n "answer": "They are the same weight",\n "justification": "A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities."\n}', additional_kwargs={}, response_metadata={'finish_reason': 'stop', 'usage': {'acceptance_rate': 5.3125, 'completion_tokens': 79, 'completion_tokens_after_first_per_sec': 292.65701089829776, 'completion_tokens_after_first_per_sec_first_ten': 346.43324678555325, 'completion_tokens_per_sec': 200.012158915008, 'end_time': 1731528071.1708555, 'is_last_response': True, 'prompt_tokens': 70, 'start_time': 1731528070.737394, 'time_to_first_token': 0.16693782806396484, 'total_latency': 0.3949759876026827, 'total_tokens': 149, 'total_tokens_per_sec': 377.2381225105847}, 'model_name': 'Meta-Llama-3.1-70B-Instruct', 'system_fingerprint': 'fastcoe', 'created': 1731528070}, id='83208297-3eb9-4021-a856-ca78a15758df'),
# 'parsed': AnswerWithJustification(answer='They are the same weight', justification='A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities.'),
# 'parsing_error': None
# }
Example: schema=None, method="json_mode", include_raw=True:
.. code-block::
from langchain_community.chat_models import ChatSambaNovaCloud
llm = ChatSambaNovaCloud(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(method="json_mode", include_raw=True)
structured_llm.invoke(
"Answer the following question. "
"Make sure to return a JSON blob with keys 'answer' and 'justification'.\n\n"
"What's heavier a pound of bricks or a pound of feathers?"
)
# -> {
# 'raw': AIMessage(content='{\n "answer": "They are the same weight",\n "justification": "A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities."\n}', additional_kwargs={}, response_metadata={'finish_reason': 'stop', 'usage': {'acceptance_rate': 4.722222222222222, 'completion_tokens': 79, 'completion_tokens_after_first_per_sec': 357.1315485254867, 'completion_tokens_after_first_per_sec_first_ten': 416.83279609305305, 'completion_tokens_per_sec': 240.92819585198137, 'end_time': 1731528164.8474727, 'is_last_response': True, 'prompt_tokens': 70, 'start_time': 1731528164.4906917, 'time_to_first_token': 0.13837409019470215, 'total_latency': 0.3278985247892492, 'total_tokens': 149, 'total_tokens_per_sec': 454.4088757208256}, 'model_name': 'Meta-Llama-3.1-70B-Instruct', 'system_fingerprint': 'fastcoe', 'created': 1731528164}, id='15261eaf-8a25-42ef-8ed5-f63d8bf5b1b0'),
# 'parsed': {
# 'answer': 'They are the same weight',
# 'justification': 'A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities.'},
# },
# 'parsing_error': None
# }
Example: schema=None, method="json_schema", include_raw=True:
.. code-block::
from langchain_community.chat_models import ChatSambaNovaCloud
class AnswerWithJustification(BaseModel):
answer: str
justification: str
llm = ChatSambaNovaCloud(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(AnswerWithJustification, method="json_schema", include_raw=True)
structured_llm.invoke(
"Answer the following question. "
"Make sure to return a JSON blob with keys 'answer' and 'justification'.\n\n"
"What's heavier a pound of bricks or a pound of feathers?"
)
# -> {
# 'raw': AIMessage(content='{\n "answer": "They are the same weight",\n "justification": "A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities."\n}', additional_kwargs={}, response_metadata={'finish_reason': 'stop', 'usage': {'acceptance_rate': 5.3125, 'completion_tokens': 79, 'completion_tokens_after_first_per_sec': 292.65701089829776, 'completion_tokens_after_first_per_sec_first_ten': 346.43324678555325, 'completion_tokens_per_sec': 200.012158915008, 'end_time': 1731528071.1708555, 'is_last_response': True, 'prompt_tokens': 70, 'start_time': 1731528070.737394, 'time_to_first_token': 0.16693782806396484, 'total_latency': 0.3949759876026827, 'total_tokens': 149, 'total_tokens_per_sec': 377.2381225105847}, 'model_name': 'Meta-Llama-3.1-70B-Instruct', 'system_fingerprint': 'fastcoe', 'created': 1731528070}, id='83208297-3eb9-4021-a856-ca78a15758df'),
# 'parsed': AnswerWithJustification(answer='They are the same weight', justification='A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities.'),
# 'parsing_error': None
# }
""" # noqa: E501
if kwargs is not None:
raise ValueError(f"Received unsupported arguments {kwargs}")
is_pydantic_schema = _is_pydantic_class(schema)
if method == "function_calling":
if schema is None:
raise ValueError(
"`schema` must be specified when method is `function_calling`. "
"Received None."
)
tool_name = convert_to_openai_tool(schema)["function"]["name"]
llm = self.bind_tools([schema], tool_choice=tool_name)
if is_pydantic_schema:
output_parser: OutputParserLike[Any] = PydanticToolsParser(
tools=[schema],
first_tool_only=True,
)
else:
output_parser = JsonOutputKeyToolsParser(
key_name=tool_name, first_tool_only=True
)
elif method == "json_mode":
llm = self
# TODO bind response format when json mode available by API
# llm = self.bind(response_format={"type": "json_object"})
if is_pydantic_schema:
schema = cast(Type[BaseModel], schema)
output_parser = PydanticOutputParser(pydantic_object=schema)
else:
output_parser = JsonOutputParser()
elif method == "json_schema":
if schema is None:
raise ValueError(
"`schema` must be specified when method is not `json_mode`. "
"Received None."
)
llm = self
# TODO bind response format when json schema available by API,
# update example
# llm = self.bind(
# response_format={"type": "json_object", "json_schema": schema}
# )
if is_pydantic_schema:
schema = cast(Type[BaseModel], schema)
output_parser = PydanticOutputParser(pydantic_object=schema)
else:
output_parser = JsonOutputParser()
else:
raise ValueError(
f"Unrecognized method argument. Expected one of `function_calling` or "
f"`json_mode`. Received: `{method}`"
)
if include_raw:
parser_assign = RunnablePassthrough.assign(
parsed=itemgetter("raw") | output_parser, parsing_error=lambda _: None
)
parser_none = RunnablePassthrough.assign(parsed=lambda _: None)
parser_with_fallback = parser_assign.with_fallbacks(
[parser_none], exception_key="parsing_error"
)
return RunnableMap(raw=llm) | parser_with_fallback
else:
return llm | output_parser
def _handle_request(
self,
messages_dicts: List[Dict[str, Any]],
stop: Optional[List[str]] = None,
streaming: bool = False,
**kwargs: Any,
) -> Response:
"""
Performs a post request to the LLM API.
Args:
messages_dicts: List of role / content dicts to use as input.
stop: list of stop tokens
streaming: wether to do a streaming call
Returns:
An iterator of response dicts.
"""
if streaming:
data = {
"messages": messages_dicts,
"max_tokens": self.max_tokens,
"stop": stop,
"model": self.model,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
"stream": True,
"stream_options": self.stream_options,
**kwargs,
}
else:
data = {
"messages": messages_dicts,
"max_tokens": self.max_tokens,
"stop": stop,
"model": self.model,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
**kwargs,
}
http_session = requests.Session()
response = http_session.post(
self.sambanova_url,
headers={
"Authorization": f"Bearer {self.sambanova_api_key.get_secret_value()}",
"Content-Type": "application/json",
**self.additional_headers,
},
json=data,
stream=streaming,
)
if response.status_code != 200:
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}.",
f"{response.text}.",
)
return response
def _process_response(self, response: Response) -> AIMessage:
"""
Process a non streaming response from the api
Args:
response: A request Response object
Returns
generation: an AIMessage with model generation
"""
try:
response_dict = response.json()
if response_dict.get("error"):
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}.",
f"{response_dict}.",
)
except Exception as e:
raise RuntimeError(
f"Sambanova /complete call failed couldn't get JSON response {e}"
f"response: {response.text}"
)
content = response_dict["choices"][0]["message"].get("content", "")
if content is None:
content = ""
additional_kwargs: Dict[str, Any] = {}
tool_calls = []
invalid_tool_calls = []
raw_tool_calls = response_dict["choices"][0]["message"].get("tool_calls")
if raw_tool_calls:
additional_kwargs["tool_calls"] = raw_tool_calls
for raw_tool_call in raw_tool_calls:
if isinstance(raw_tool_call["function"]["arguments"], dict):
raw_tool_call["function"]["arguments"] = json.dumps(
raw_tool_call["function"].get("arguments", {})
)
try:
tool_calls.append(parse_tool_call(raw_tool_call, return_id=True))
except Exception as e:
invalid_tool_calls.append(
make_invalid_tool_call(raw_tool_call, str(e))
)
message = AIMessage(
content=content,
additional_kwargs=additional_kwargs,
tool_calls=tool_calls,
invalid_tool_calls=invalid_tool_calls,
response_metadata={
"finish_reason": response_dict["choices"][0]["finish_reason"],
"usage": response_dict.get("usage"),
"model_name": response_dict["model"],
"system_fingerprint": response_dict["system_fingerprint"],
"created": response_dict["created"],
},
id=response_dict["id"],
)
return message
def _process_stream_response(
self, response: Response
) -> Iterator[BaseMessageChunk]:
"""
Process a streaming response from the api
Args:
response: An iterable request Response object
Yields:
generation: an AIMessageChunk with model partial generation
"""
try:
import sseclient
except ImportError:
raise ImportError(
"could not import sseclient library"
"Please install it with `pip install sseclient-py`."
)
client = sseclient.SSEClient(response)
for event in client.events():
if event.event == "error_event":
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}."
f"{event.data}."
)
try:
# check if the response is a final event
# in that case event data response is '[DONE]'
if event.data != "[DONE]":
if isinstance(event.data, str):
data = json.loads(event.data)
else:
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}."
f"{event.data}."
)
if data.get("error"):
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}."
f"{event.data}."
)
if len(data["choices"]) > 0:
finish_reason = data["choices"][0].get("finish_reason")
content = data["choices"][0]["delta"]["content"]
id = data["id"]
chunk = AIMessageChunk(
content=content, id=id, additional_kwargs={}
)
else:
content = ""
id = data["id"]
metadata = {
"finish_reason": finish_reason,
"usage": data.get("usage"),
"model_name": data["model"],
"system_fingerprint": data["system_fingerprint"],
"created": data["created"],
}
chunk = AIMessageChunk(
content=content,
id=id,
response_metadata=metadata,
additional_kwargs={},
)
yield chunk
except Exception as e:
raise RuntimeError(
f"Error getting content chunk raw streamed response: {e}"
f"data: {event.data}"
)
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
"""
Call SambaNovaCloud models.
Args:
messages: the prompt composed of a list of messages.
stop: a list of strings on which the model should stop generating.
If generation stops due to a stop token, the stop token itself
SHOULD BE INCLUDED as part of the output. This is not enforced
across models right now, but it's a good practice to follow since
it makes it much easier to parse the output of the model
downstream and understand why generation stopped.
run_manager: A run manager with callbacks for the LLM.
Returns:
result: ChatResult with model generation
"""
if self.streaming:
stream_iter = self._stream(
messages, stop=stop, run_manager=run_manager, **kwargs
)
if stream_iter:
return generate_from_stream(stream_iter)
messages_dicts = _create_message_dicts(messages)
response = self._handle_request(messages_dicts, stop, streaming=False, **kwargs)
message = self._process_response(response)
generation = ChatGeneration(
message=message,
generation_info={
"finish_reason": message.response_metadata["finish_reason"]
},
)
return ChatResult(generations=[generation])
def _stream(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> Iterator[ChatGenerationChunk]:
"""
Stream the output of the SambaNovaCloud chat model.
Args:
messages: the prompt composed of a list of messages.
stop: a list of strings on which the model should stop generating.
If generation stops due to a stop token, the stop token itself
SHOULD BE INCLUDED as part of the output. This is not enforced
across models right now, but it's a good practice to follow since
it makes it much easier to parse the output of the model
downstream and understand why generation stopped.
run_manager: A run manager with callbacks for the LLM.
Yields:
chunk: ChatGenerationChunk with model partial generation
"""
messages_dicts = _create_message_dicts(messages)
response = self._handle_request(messages_dicts, stop, streaming=True, **kwargs)
for ai_message_chunk in self._process_stream_response(response):
chunk = ChatGenerationChunk(message=ai_message_chunk)
if run_manager:
run_manager.on_llm_new_token(chunk.text, chunk=chunk)
yield chunk
[docs]
class ChatSambaStudio(BaseChatModel):
"""
SambaStudio chat model.
Setup:
To use, you should have the environment variables:
`SAMBASTUDIO_URL` set with your SambaStudio deployed endpoint URL.
`SAMBASTUDIO_API_KEY` set with your SambaStudio deployed endpoint Key.
https://docs.sambanova.ai/sambastudio/latest/index.html
Example:
.. code-block:: python
ChatSambaStudio(
sambastudio_url = set with your SambaStudio deployed endpoint URL,
sambastudio_api_key = set with your SambaStudio deployed endpoint Key.
model = model or expert name (set for Bundle endpoints),
max_tokens = max number of tokens to generate,
temperature = model temperature,
top_p = model top p,
top_k = model top k,
do_sample = wether to do sample
process_prompt = wether to process prompt
(set for Bundle generic v1 and v2 endpoints)
stream_options = include usage to get generation metrics
special_tokens = start, start_role, end_role, end special tokens
(set for Bundle generic v1 and v2 endpoints when process prompt
set to false or for StandAlone v1 and v2 endpoints)
model_kwargs: Optional = Extra Key word arguments to pass to the model.
)
Key init args — completion params:
model: str
The name of the model to use, e.g., Meta-Llama-3-70B-Instruct-4096
(set for Bundle endpoints).
streaming: bool
Whether to use streaming
max_tokens: inthandler when using non streaming methods
max tokens to generate
temperature: float
model temperature
top_p: float
model top p
top_k: int
model top k
do_sample: bool
wether to do sample
process_prompt:
wether to process prompt (set for Bundle generic v1 and v2 endpoints)
stream_options: dict
stream options, include usage to get generation metrics
special_tokens: dict
start, start_role, end_role and end special tokens
(set for Bundle generic v1 and v2 endpoints when process prompt set to false
or for StandAlone v1 and v2 endpoints) default to llama3 special tokens
model_kwargs: dict
Extra Key word arguments to pass to the model.
Key init args — client params:
sambastudio_url: str
SambaStudio endpoint Url
sambastudio_api_key: str
SambaStudio endpoint api key
Instantiate:
.. code-block:: python
from langchain_community.chat_models import ChatSambaStudio
chat = ChatSambaStudio=(
sambastudio_url = set with your SambaStudio deployed endpoint URL,
sambastudio_api_key = set with your SambaStudio deployed endpoint Key.
model = model or expert name (set for Bundle endpoints),
max_tokens = max number of tokens to generate,
temperature = model temperature,
top_p = model top p,
top_k = model top k,
do_sample = wether to do sample
process_prompt = wether to process prompt
(set for Bundle generic v1 and v2 endpoints)
stream_options = include usage to get generation metrics
special_tokens = start, start_role, end_role, and special tokens
(set for Bundle generic v1 and v2 endpoints when process prompt
set to false or for StandAlone v1 and v2 endpoints)
model_kwargs: Optional = Extra Key word arguments to pass to the model.
)
Invoke:
.. code-block:: python
messages = [
SystemMessage(content="your are an AI assistant."),
HumanMessage(content="tell me a joke."),
]
response = chat.invoke(messages)
Stream:
.. code-block:: python
for chunk in chat.stream(messages):
print(chunk.content, end="", flush=True)
Async:
.. code-block:: python
response = chat.ainvoke(messages)
await response
Tool calling:
.. code-block:: python
from pydantic import BaseModel, Field
class GetWeather(BaseModel):
'''Get the current weather in a given location'''
location: str = Field(
...,
description="The city and state, e.g. Los Angeles, CA"
)
llm_with_tools = llm.bind_tools([GetWeather, GetPopulation])
ai_msg = llm_with_tools.invoke("Should I bring my umbrella today in LA?")
ai_msg.tool_calls
.. code-block:: python
[
{
'name': 'GetWeather',
'args': {'location': 'Los Angeles, CA'},
'id': 'call_adf61180ea2b4d228a'
}
]
Structured output:
.. code-block:: python
from typing import Optional
from pydantic import BaseModel, Field
class Joke(BaseModel):
'''Joke to tell user.'''
setup: str = Field(description="The setup of the joke")
punchline: str = Field(description="The punchline to the joke")
structured_model = llm.with_structured_output(Joke)
structured_model.invoke("Tell me a joke about cats")
.. code-block:: python
Joke(setup="Why did the cat join a band?",
punchline="Because it wanted to be the purr-cussionist!")
See `ChatSambaStudio.with_structured_output()` for more.
Token usage:
.. code-block:: python
response = chat.invoke(messages)
print(response.response_metadata["usage"]["prompt_tokens"]
print(response.response_metadata["usage"]["total_tokens"]
Response metadata
.. code-block:: python
response = chat.invoke(messages)
print(response.response_metadata)
"""
sambastudio_url: str = Field(default="")
"""SambaStudio Url"""
sambastudio_api_key: SecretStr = Field(default=SecretStr(""))
"""SambaStudio api key"""
base_url: str = Field(default="", exclude=True)
"""SambaStudio non streaming Url"""
streaming_url: str = Field(default="", exclude=True)
"""SambaStudio streaming Url"""
model: Optional[str] = Field(default=None)
"""The name of the model or expert to use (for Bundle endpoints)"""
streaming: bool = Field(default=False)
"""Whether to use streaming handler when using non streaming methods"""
max_tokens: int = Field(default=1024)
"""max tokens to generate"""
temperature: Optional[float] = Field(default=0.7)
"""model temperature"""
top_p: Optional[float] = Field(default=None)
"""model top p"""
top_k: Optional[int] = Field(default=None)
"""model top k"""
do_sample: Optional[bool] = Field(default=None)
"""whether to do sampling"""
process_prompt: Optional[bool] = Field(default=True)
"""whether process prompt (for Bundle generic v1 and v2 endpoints)"""
stream_options: Dict[str, Any] = Field(default={"include_usage": True})
"""stream options, include usage to get generation metrics"""
special_tokens: Dict[str, Any] = Field(
default={
"start": "<|begin_of_text|>",
"start_role": "<|begin_of_text|><|start_header_id|>{role}<|end_header_id|>",
"end_role": "<|eot_id|>",
"end": "<|start_header_id|>assistant<|end_header_id|>\n",
}
)
"""start, start_role, end_role and end special tokens
(set for Bundle generic v1 and v2 endpoints when process prompt set to false
or for StandAlone v1 and v2 endpoints)
default to llama3 special tokens"""
model_kwargs: Optional[Dict[str, Any]] = None
"""Key word arguments to pass to the model."""
additional_headers: Dict[str, Any] = Field(default={})
"""Additional headers to send in request"""
[docs]
class Config:
populate_by_name = True
@classmethod
def is_lc_serializable(cls) -> bool:
"""Return whether this model can be serialized by Langchain."""
return False
@property
def lc_secrets(self) -> Dict[str, str]:
return {
"sambastudio_url": "sambastudio_url",
"sambastudio_api_key": "sambastudio_api_key",
}
@property
def _identifying_params(self) -> Dict[str, Any]:
"""Return a dictionary of identifying parameters.
This information is used by the LangChain callback system, which
is used for tracing purposes make it possible to monitor LLMs.
"""
return {
"model": self.model,
"streaming": self.streaming,
"max_tokens": self.max_tokens,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
"do_sample": self.do_sample,
"process_prompt": self.process_prompt,
"stream_options": self.stream_options,
"special_tokens": self.special_tokens,
"model_kwargs": self.model_kwargs,
}
@property
def _llm_type(self) -> str:
"""Get the type of language model used by this chat model."""
return "sambastudio-chatmodel"
def __init__(self, **kwargs: Any) -> None:
"""init and validate environment variables"""
kwargs["sambastudio_url"] = get_from_dict_or_env(
kwargs, "sambastudio_url", "SAMBASTUDIO_URL"
)
kwargs["sambastudio_api_key"] = convert_to_secret_str(
get_from_dict_or_env(kwargs, "sambastudio_api_key", "SAMBASTUDIO_API_KEY")
)
kwargs["base_url"], kwargs["streaming_url"] = self._get_sambastudio_urls(
kwargs["sambastudio_url"]
)
super().__init__(**kwargs)
[docs]
def with_structured_output(
self,
schema: Optional[Union[Dict[str, Any], Type[BaseModel]]] = None,
*,
method: Literal[
"function_calling", "json_mode", "json_schema"
] = "function_calling",
include_raw: bool = False,
**kwargs: Any,
) -> Runnable[LanguageModelInput, Union[Dict[str, Any], BaseModel]]:
"""Model wrapper that returns outputs formatted to match the given schema.
Args:
schema:
The output schema. Can be passed in as:
- an OpenAI function/tool schema,
- a JSON Schema,
- a TypedDict class,
- or a Pydantic class.
If `schema` is a Pydantic class then the model output will be a
Pydantic instance of that class, and the model-generated fields will be
validated by the Pydantic class. Otherwise the model output will be a
dict and will not be validated. See :meth:`langchain_core.utils.function_calling.convert_to_openai_tool`
for more on how to properly specify types and descriptions of
schema fields when specifying a Pydantic or TypedDict class.
method:
The method for steering model generation, either "function_calling"
"json_mode" or "json_schema".
If "function_calling" then the schema will be converted
to an OpenAI function and the returned model will make use of the
function-calling API. If "json_mode" or "json_schema" then OpenAI's
JSON mode will be used.
Note that if using "json_mode" or "json_schema" then you must include instructions
for formatting the output into the desired schema into the model call.
include_raw:
If False then only the parsed structured output is returned. If
an error occurs during model output parsing it will be raised. If True
then both the raw model response (a BaseMessage) and the parsed model
response will be returned. If an error occurs during output parsing it
will be caught and returned as well. The final output is always a dict
with keys "raw", "parsed", and "parsing_error".
Returns:
A Runnable that takes same inputs as a :class:`langchain_core.language_models.chat.BaseChatModel`.
If `include_raw` is False and `schema` is a Pydantic class, Runnable outputs
an instance of `schema` (i.e., a Pydantic object).
Otherwise, if `include_raw` is False then Runnable outputs a dict.
If `include_raw` is True, then Runnable outputs a dict with keys:
- `"raw"`: BaseMessage
- `"parsed"`: None if there was a parsing error, otherwise the type depends on the `schema` as described above.
- `"parsing_error"`: Optional[BaseException]
Example: schema=Pydantic class, method="function_calling", include_raw=False:
.. code-block:: python
from typing import Optional
from langchain_community.chat_models import ChatSambaStudio
from pydantic import BaseModel, Field
class AnswerWithJustification(BaseModel):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: str = Field(
description="A justification for the answer."
)
llm = ChatSambaStudio(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(AnswerWithJustification)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> AnswerWithJustification(
# answer='They weigh the same',
# justification='A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same.'
# )
Example: schema=Pydantic class, method="function_calling", include_raw=True:
.. code-block:: python
from langchain_community.chat_models import ChatSambaStudio
from pydantic import BaseModel
class AnswerWithJustification(BaseModel):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: str
llm = ChatSambaStudio(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(
AnswerWithJustification, include_raw=True
)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'raw': AIMessage(content='', additional_kwargs={'tool_calls': [{'function': {'arguments': '{"answer": "They weigh the same.", "justification": "A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount."}', 'name': 'AnswerWithJustification'}, 'id': 'call_17a431fc6a4240e1bd', 'type': 'function'}]}, response_metadata={'finish_reason': 'tool_calls', 'usage': {'acceptance_rate': 5, 'completion_tokens': 53, 'completion_tokens_after_first_per_sec': 343.7964936837758, 'completion_tokens_after_first_per_sec_first_ten': 439.1205661878638, 'completion_tokens_per_sec': 162.8511306784833, 'end_time': 1731527851.0698032, 'is_last_response': True, 'prompt_tokens': 213, 'start_time': 1731527850.7137961, 'time_to_first_token': 0.20475482940673828, 'total_latency': 0.32545061111450196, 'total_tokens': 266, 'total_tokens_per_sec': 817.3283162354066}, 'model_name': 'Meta-Llama-3.1-70B-Instruct', 'system_fingerprint': 'fastcoe', 'created': 1731527850}, id='95667eaf-447f-4b53-bb6e-b6e1094ded88', tool_calls=[{'name': 'AnswerWithJustification', 'args': {'answer': 'They weigh the same.', 'justification': 'A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount.'}, 'id': 'call_17a431fc6a4240e1bd', 'type': 'tool_call'}]),
# 'parsed': AnswerWithJustification(answer='They weigh the same.', justification='A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount.'),
# 'parsing_error': None
# }
Example: schema=TypedDict class, method="function_calling", include_raw=False:
.. code-block:: python
# IMPORTANT: If you are using Python <=3.8, you need to import Annotated
# from typing_extensions, not from typing.
from typing_extensions import Annotated, TypedDict
from langchain_community.chat_models import ChatSambaStudio
class AnswerWithJustification(TypedDict):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: Annotated[
Optional[str], None, "A justification for the answer."
]
llm = ChatSambaStudio(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(AnswerWithJustification)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'answer': 'They weigh the same',
# 'justification': 'A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount.'
# }
Example: schema=OpenAI function schema, method="function_calling", include_raw=False:
.. code-block:: python
from langchain_community.chat_models import ChatSambaStudio
oai_schema = {
'name': 'AnswerWithJustification',
'description': 'An answer to the user question along with justification for the answer.',
'parameters': {
'type': 'object',
'properties': {
'answer': {'type': 'string'},
'justification': {'description': 'A justification for the answer.', 'type': 'string'}
},
'required': ['answer']
}
}
llm = ChatSambaStudio(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(oai_schema)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'answer': 'They weigh the same',
# 'justification': 'A pound is a unit of weight or mass, so one pound of bricks and one pound of feathers both weigh the same amount.'
# }
Example: schema=Pydantic class, method="json_mode", include_raw=True:
.. code-block::
from langchain_community.chat_models import ChatSambaStudio
from pydantic import BaseModel
class AnswerWithJustification(BaseModel):
answer: str
justification: str
llm = ChatSambaStudio(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(
AnswerWithJustification,
method="json_mode",
include_raw=True
)
structured_llm.invoke(
"Answer the following question. "
"Make sure to return a JSON blob with keys 'answer' and 'justification'.\n\n"
"What's heavier a pound of bricks or a pound of feathers?"
)
# -> {
# 'raw': AIMessage(content='{\n "answer": "They are the same weight",\n "justification": "A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities."\n}', additional_kwargs={}, response_metadata={'finish_reason': 'stop', 'usage': {'acceptance_rate': 5.3125, 'completion_tokens': 79, 'completion_tokens_after_first_per_sec': 292.65701089829776, 'completion_tokens_after_first_per_sec_first_ten': 346.43324678555325, 'completion_tokens_per_sec': 200.012158915008, 'end_time': 1731528071.1708555, 'is_last_response': True, 'prompt_tokens': 70, 'start_time': 1731528070.737394, 'time_to_first_token': 0.16693782806396484, 'total_latency': 0.3949759876026827, 'total_tokens': 149, 'total_tokens_per_sec': 377.2381225105847}, 'model_name': 'Meta-Llama-3.1-70B-Instruct', 'system_fingerprint': 'fastcoe', 'created': 1731528070}, id='83208297-3eb9-4021-a856-ca78a15758df'),
# 'parsed': AnswerWithJustification(answer='They are the same weight', justification='A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities.'),
# 'parsing_error': None
# }
Example: schema=None, method="json_mode", include_raw=True:
.. code-block::
from langchain_community.chat_models import ChatSambaStudio
llm = ChatSambaStudio(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(method="json_mode", include_raw=True)
structured_llm.invoke(
"Answer the following question. "
"Make sure to return a JSON blob with keys 'answer' and 'justification'.\n\n"
"What's heavier a pound of bricks or a pound of feathers?"
)
# -> {
# 'raw': AIMessage(content='{\n "answer": "They are the same weight",\n "justification": "A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities."\n}', additional_kwargs={}, response_metadata={'finish_reason': 'stop', 'usage': {'acceptance_rate': 4.722222222222222, 'completion_tokens': 79, 'completion_tokens_after_first_per_sec': 357.1315485254867, 'completion_tokens_after_first_per_sec_first_ten': 416.83279609305305, 'completion_tokens_per_sec': 240.92819585198137, 'end_time': 1731528164.8474727, 'is_last_response': True, 'prompt_tokens': 70, 'start_time': 1731528164.4906917, 'time_to_first_token': 0.13837409019470215, 'total_latency': 0.3278985247892492, 'total_tokens': 149, 'total_tokens_per_sec': 454.4088757208256}, 'model_name': 'Meta-Llama-3.1-70B-Instruct', 'system_fingerprint': 'fastcoe', 'created': 1731528164}, id='15261eaf-8a25-42ef-8ed5-f63d8bf5b1b0'),
# 'parsed': {
# 'answer': 'They are the same weight',
# 'justification': 'A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities.'},
# },
# 'parsing_error': None
# }
Example: schema=None, method="json_schema", include_raw=True:
.. code-block::
from langchain_community.chat_models import ChatSambaStudio
class AnswerWithJustification(BaseModel):
answer: str
justification: str
llm = ChatSambaStudio(model="Meta-Llama-3.1-70B-Instruct", temperature=0)
structured_llm = llm.with_structured_output(AnswerWithJustification, method="json_schema", include_raw=True)
structured_llm.invoke(
"Answer the following question. "
"Make sure to return a JSON blob with keys 'answer' and 'justification'.\n\n"
"What's heavier a pound of bricks or a pound of feathers?"
)
# -> {
# 'raw': AIMessage(content='{\n "answer": "They are the same weight",\n "justification": "A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities."\n}', additional_kwargs={}, response_metadata={'finish_reason': 'stop', 'usage': {'acceptance_rate': 5.3125, 'completion_tokens': 79, 'completion_tokens_after_first_per_sec': 292.65701089829776, 'completion_tokens_after_first_per_sec_first_ten': 346.43324678555325, 'completion_tokens_per_sec': 200.012158915008, 'end_time': 1731528071.1708555, 'is_last_response': True, 'prompt_tokens': 70, 'start_time': 1731528070.737394, 'time_to_first_token': 0.16693782806396484, 'total_latency': 0.3949759876026827, 'total_tokens': 149, 'total_tokens_per_sec': 377.2381225105847}, 'model_name': 'Meta-Llama-3.1-70B-Instruct', 'system_fingerprint': 'fastcoe', 'created': 1731528070}, id='83208297-3eb9-4021-a856-ca78a15758df'),
# 'parsed': AnswerWithJustification(answer='They are the same weight', justification='A pound is a unit of weight or mass, so a pound of bricks and a pound of feathers both weigh the same amount, one pound. The difference is in their density and volume. A pound of feathers would take up more space than a pound of bricks due to the difference in their densities.'),
# 'parsing_error': None
# }
""" # noqa: E501
if kwargs:
raise ValueError(f"Received unsupported arguments {kwargs}")
is_pydantic_schema = _is_pydantic_class(schema)
if method == "function_calling":
if schema is None:
raise ValueError(
"schema must be specified when method is 'function_calling'. "
"Received None."
)
tool_name = convert_to_openai_tool(schema)["function"]["name"]
llm = self.bind_tools([schema], tool_choice=tool_name)
if is_pydantic_schema:
output_parser: OutputParserLike[Any] = PydanticToolsParser(
tools=[schema], # type: ignore[list-item]
first_tool_only=True,
)
else:
output_parser = JsonOutputKeyToolsParser(
key_name=tool_name, first_tool_only=True
)
elif method == "json_mode":
llm = self
# TODO bind response format when json mode available by API
# llm = self.bind(response_format={"type": "json_object"})
if is_pydantic_schema:
schema = cast(Type[BaseModel], schema)
output_parser = PydanticOutputParser(pydantic_object=schema)
else:
output_parser = JsonOutputParser()
elif method == "json_schema":
if schema is None:
raise ValueError(
"schema must be specified when method is not 'json_mode'. "
"Received None."
)
llm = self
# TODO bind response format when json schema available by API,
# update example
# llm = self.bind(
# response_format={"type": "json_object", "json_schema": schema}
# )
if is_pydantic_schema:
schema = cast(Type[BaseModel], schema)
output_parser = PydanticOutputParser(pydantic_object=schema)
else:
output_parser = JsonOutputParser()
else:
raise ValueError(
f"Unrecognized method argument. Expected one of 'function_calling' or "
f"'json_mode'. Received: '{method}'"
)
if include_raw:
parser_assign = RunnablePassthrough.assign(
parsed=itemgetter("raw") | output_parser, parsing_error=lambda _: None
)
parser_none = RunnablePassthrough.assign(parsed=lambda _: None)
parser_with_fallback = parser_assign.with_fallbacks(
[parser_none], exception_key="parsing_error"
)
return RunnableMap(raw=llm) | parser_with_fallback
else:
return llm | output_parser
def _get_role(self, message: BaseMessage) -> str:
"""
Get the role of LangChain BaseMessage
Args:
message: LangChain BaseMessage
Returns:
str: Role of the LangChain BaseMessage
"""
if isinstance(message, SystemMessage):
role = "system"
elif isinstance(message, HumanMessage):
role = "user"
elif isinstance(message, AIMessage):
role = "assistant"
elif isinstance(message, ToolMessage):
role = "tool"
elif isinstance(message, ChatMessage):
role = message.role
else:
raise TypeError(f"Got unknown type {message}")
return role
def _messages_to_string(self, messages: List[BaseMessage], **kwargs: Any) -> str:
"""
Convert a list of BaseMessages to a:
- dumped json string with Role / content dict structure
when process_prompt is true,
- string with special tokens if process_prompt is false
for generic V1 and V2 endpoints
Args:
messages: list of BaseMessages
Returns:
str: string to send as model input depending on process_prompt param
"""
if self.process_prompt:
messages_dict: Dict[str, Any] = {
"conversation_id": "sambaverse-conversation-id",
"messages": [],
**kwargs,
}
for message in messages:
if isinstance(message, AIMessage):
message_dict = {
"message_id": message.id,
"role": self._get_role(message),
"content": message.content,
}
if "tool_calls" in message.additional_kwargs:
message_dict["tool_calls"] = message.additional_kwargs[
"tool_calls"
]
if message_dict["content"] == "":
message_dict["content"] = None
elif isinstance(message, ToolMessage):
message_dict = {
"message_id": message.id,
"role": self._get_role(message),
"content": message.content,
"tool_call_id": message.tool_call_id,
}
else:
message_dict = {
"message_id": message.id,
"role": self._get_role(message),
"content": message.content,
}
messages_dict["messages"].append(message_dict)
messages_string = json.dumps(messages_dict)
else:
if "tools" in kwargs.keys():
raise NotImplementedError(
"tool calling not supported in API Generic V2 "
"without process_prompt, switch to OpenAI compatible API "
"or Generic V2 API with process_prompt=True"
)
messages_string = self.special_tokens["start"]
for message in messages:
messages_string += self.special_tokens["start_role"].format(
role=self._get_role(message)
)
messages_string += f" {message.content} "
messages_string += self.special_tokens["end_role"]
messages_string += self.special_tokens["end"]
return messages_string
def _get_sambastudio_urls(self, url: str) -> Tuple[str, str]:
"""
Get streaming and non streaming URLs from the given URL
Args:
url: string with sambastudio base or streaming endpoint url
Returns:
base_url: string with url to do non streaming calls
streaming_url: string with url to do streaming calls
"""
if "chat/completions" in url:
base_url = url
stream_url = url
else:
if "stream" in url:
base_url = url.replace("stream/", "")
stream_url = url
else:
base_url = url
if "generic" in url:
stream_url = "generic/stream".join(url.split("generic"))
else:
raise ValueError("Unsupported URL")
return base_url, stream_url
def _handle_request(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
streaming: Optional[bool] = False,
**kwargs: Any,
) -> Response:
"""
Performs a post request to the LLM API.
Args:
messages_dicts: List of role / content dicts to use as input.
stop: list of stop tokens
streaming: wether to do a streaming call
Returns:
A request Response object
"""
# create request payload for openai compatible API
if "chat/completions" in self.sambastudio_url:
messages_dicts = _create_message_dicts(messages)
data = {
"messages": messages_dicts,
"max_tokens": self.max_tokens,
"stop": stop,
"model": self.model,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
"stream": streaming,
"stream_options": self.stream_options,
**kwargs,
}
data = {key: value for key, value in data.items() if value is not None}
headers = {
"Authorization": f"Bearer "
f"{self.sambastudio_api_key.get_secret_value()}",
"Content-Type": "application/json",
**self.additional_headers,
}
# create request payload for generic v2 API
elif "api/v2/predict/generic" in self.sambastudio_url:
items = [
{"id": "item0", "value": self._messages_to_string(messages, **kwargs)}
]
params: Dict[str, Any] = {
"select_expert": self.model,
"process_prompt": self.process_prompt,
"max_tokens_to_generate": self.max_tokens,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
"do_sample": self.do_sample,
}
if self.model_kwargs is not None:
params = {**params, **self.model_kwargs}
params = {key: value for key, value in params.items() if value is not None}
data = {"items": items, "params": params}
headers = {
"key": self.sambastudio_api_key.get_secret_value(),
**self.additional_headers,
}
# create request payload for generic v1 API
elif "api/predict/generic" in self.sambastudio_url:
if "tools" in kwargs.keys():
raise NotImplementedError(
"tool calling not supported in API Generic V1, "
"switch to OpenAI compatible API or Generic V2 API"
)
params = {
"select_expert": self.model,
"process_prompt": self.process_prompt,
"max_tokens_to_generate": self.max_tokens,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
"do_sample": self.do_sample,
**kwargs,
}
if self.model_kwargs is not None:
params = {**params, **self.model_kwargs}
params = {
key: {"type": type(value).__name__, "value": str(value)}
for key, value in params.items()
if value is not None
}
if streaming:
data = {
"instance": self._messages_to_string(messages),
"params": params,
}
else:
data = {
"instances": [self._messages_to_string(messages)],
"params": params,
}
headers = {
"key": self.sambastudio_api_key.get_secret_value(),
**self.additional_headers,
}
else:
raise ValueError(
f"Unsupported URL{self.sambastudio_url}"
"only openai, generic v1 and generic v2 APIs are supported"
)
http_session = requests.Session()
if streaming:
response = http_session.post(
self.streaming_url, headers=headers, json=data, stream=True
)
else:
response = http_session.post(
self.base_url, headers=headers, json=data, stream=False
)
if response.status_code != 200:
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}."
f"{response.text}."
)
return response
def _process_response(self, response: Response) -> AIMessage:
"""
Process a non streaming response from the api
Args:
response: A request Response object
Returns
generation: an AIMessage with model generation
"""
# Extract json payload form response
try:
response_dict = response.json()
except Exception as e:
raise RuntimeError(
f"Sambanova /complete call failed couldn't get JSON response {e}"
f"response: {response.text}"
)
additional_kwargs: Dict[str, Any] = {}
tool_calls = []
invalid_tool_calls = []
# process response payload for openai compatible API
if "chat/completions" in self.sambastudio_url:
content = response_dict["choices"][0]["message"].get("content", "")
if content is None:
content = ""
id = response_dict["id"]
response_metadata = {
"finish_reason": response_dict["choices"][0]["finish_reason"],
"usage": response_dict.get("usage"),
"model_name": response_dict["model"],
"system_fingerprint": response_dict["system_fingerprint"],
"created": response_dict["created"],
}
raw_tool_calls = response_dict["choices"][0]["message"].get("tool_calls")
if raw_tool_calls:
additional_kwargs["tool_calls"] = raw_tool_calls
for raw_tool_call in raw_tool_calls:
if isinstance(raw_tool_call["function"]["arguments"], dict):
raw_tool_call["function"]["arguments"] = json.dumps(
raw_tool_call["function"].get("arguments", {})
)
try:
tool_calls.append(
parse_tool_call(raw_tool_call, return_id=True)
)
except Exception as e:
invalid_tool_calls.append(
make_invalid_tool_call(raw_tool_call, str(e))
)
# process response payload for generic v2 API
elif "api/v2/predict/generic" in self.sambastudio_url:
content = response_dict["items"][0]["value"]["completion"]
id = response_dict["items"][0]["id"]
response_metadata = response_dict["items"][0]
raw_tool_calls = response_dict["items"][0]["value"].get("tool_calls")
if raw_tool_calls:
additional_kwargs["tool_calls"] = raw_tool_calls
for raw_tool_call in raw_tool_calls:
if isinstance(raw_tool_call["function"]["arguments"], dict):
raw_tool_call["function"]["arguments"] = json.dumps(
raw_tool_call["function"].get("arguments", {})
)
try:
tool_calls.append(
parse_tool_call(raw_tool_call, return_id=True)
)
except Exception as e:
invalid_tool_calls.append(
make_invalid_tool_call(raw_tool_call, str(e))
)
# process response payload for generic v1 API
elif "api/predict/generic" in self.sambastudio_url:
content = response_dict["predictions"][0]["completion"]
id = None
response_metadata = response_dict
else:
raise ValueError(
f"Unsupported URL{self.sambastudio_url}"
"only openai, generic v1 and generic v2 APIs are supported"
)
return AIMessage(
content=content,
additional_kwargs=additional_kwargs,
tool_calls=tool_calls,
invalid_tool_calls=invalid_tool_calls,
response_metadata=response_metadata,
id=id,
)
def _process_stream_response(
self, response: Response
) -> Iterator[BaseMessageChunk]:
"""
Process a streaming response from the api
Args:
response: An iterable request Response object
Yields:
generation: an AIMessageChunk with model partial generation
"""
try:
import sseclient
except ImportError:
raise ImportError(
"could not import sseclient library"
"Please install it with `pip install sseclient-py`."
)
# process response payload for openai compatible API
if "chat/completions" in self.sambastudio_url:
finish_reason = ""
client = sseclient.SSEClient(response)
for event in client.events():
if event.event == "error_event":
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}."
f"{event.data}."
)
try:
# check if the response is not a final event ("[DONE]")
if event.data != "[DONE]":
if isinstance(event.data, str):
data = json.loads(event.data)
else:
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}."
f"{event.data}."
)
if data.get("error"):
raise RuntimeError(
f"Sambanova /complete call failed with status code "
f"{response.status_code}."
f"{event.data}."
)
if len(data["choices"]) > 0:
finish_reason = data["choices"][0].get("finish_reason")
content = data["choices"][0]["delta"]["content"]
id = data["id"]
metadata = {}
else:
content = ""
id = data["id"]
metadata = {
"finish_reason": finish_reason,
"usage": data.get("usage"),
"model_name": data["model"],
"system_fingerprint": data["system_fingerprint"],
"created": data["created"],
}
if data.get("usage") is not None:
content = ""
id = data["id"]
metadata = {
"finish_reason": finish_reason,
"usage": data.get("usage"),
"model_name": data["model"],
"system_fingerprint": data["system_fingerprint"],
"created": data["created"],
}
yield AIMessageChunk(
content=content,
id=id,
response_metadata=metadata,
additional_kwargs={},
)
except Exception as e:
raise RuntimeError(
f"Error getting content chunk raw streamed response: {e}"
f"data: {event.data}"
)
# process response payload for generic v2 API
elif "api/v2/predict/generic" in self.sambastudio_url:
for line in response.iter_lines():
try:
data = json.loads(line)
content = data["result"]["items"][0]["value"]["stream_token"]
id = data["result"]["items"][0]["id"]
if data["result"]["items"][0]["value"]["is_last_response"]:
metadata = {
"finish_reason": data["result"]["items"][0]["value"].get(
"stop_reason"
),
"prompt": data["result"]["items"][0]["value"].get("prompt"),
"usage": {
"prompt_tokens_count": data["result"]["items"][0][
"value"
].get("prompt_tokens_count"),
"completion_tokens_count": data["result"]["items"][0][
"value"
].get("completion_tokens_count"),
"total_tokens_count": data["result"]["items"][0][
"value"
].get("total_tokens_count"),
"start_time": data["result"]["items"][0]["value"].get(
"start_time"
),
"end_time": data["result"]["items"][0]["value"].get(
"end_time"
),
"model_execution_time": data["result"]["items"][0][
"value"
].get("model_execution_time"),
"time_to_first_token": data["result"]["items"][0][
"value"
].get("time_to_first_token"),
"throughput_after_first_token": data["result"]["items"][
0
]["value"].get("throughput_after_first_token"),
"batch_size_used": data["result"]["items"][0][
"value"
].get("batch_size_used"),
},
}
else:
metadata = {}
yield AIMessageChunk(
content=content,
id=id,
response_metadata=metadata,
additional_kwargs={},
)
except Exception as e:
raise RuntimeError(
f"Error getting content chunk raw streamed response: {e}"
f"line: {line}"
)
# process response payload for generic v1 API
elif "api/predict/generic" in self.sambastudio_url:
for line in response.iter_lines():
try:
data = json.loads(line)
content = data["result"]["responses"][0]["stream_token"]
id = None
if data["result"]["responses"][0]["is_last_response"]:
metadata = {
"finish_reason": data["result"]["responses"][0].get(
"stop_reason"
),
"prompt": data["result"]["responses"][0].get("prompt"),
"usage": {
"prompt_tokens_count": data["result"]["responses"][
0
].get("prompt_tokens_count"),
"completion_tokens_count": data["result"]["responses"][
0
].get("completion_tokens_count"),
"total_tokens_count": data["result"]["responses"][
0
].get("total_tokens_count"),
"start_time": data["result"]["responses"][0].get(
"start_time"
),
"end_time": data["result"]["responses"][0].get(
"end_time"
),
"model_execution_time": data["result"]["responses"][
0
].get("model_execution_time"),
"time_to_first_token": data["result"]["responses"][
0
].get("time_to_first_token"),
"throughput_after_first_token": data["result"][
"responses"
][0].get("throughput_after_first_token"),
"batch_size_used": data["result"]["responses"][0].get(
"batch_size_used"
),
},
}
else:
metadata = {}
yield AIMessageChunk(
content=content,
id=id,
response_metadata=metadata,
additional_kwargs={},
)
except Exception as e:
raise RuntimeError(
f"Error getting content chunk raw streamed response: {e}"
f"line: {line}"
)
else:
raise ValueError(
f"Unsupported URL{self.sambastudio_url}"
"only openai, generic v1 and generic v2 APIs are supported"
)
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
"""
Call SambaStudio models.
Args:
messages: the prompt composed of a list of messages.
stop: a list of strings on which the model should stop generating.
If generation stops due to a stop token, the stop token itself
SHOULD BE INCLUDED as part of the output. This is not enforced
across models right now, but it's a good practice to follow since
it makes it much easier to parse the output of the model
downstream and understand why generation stopped.
run_manager: A run manager with callbacks for the LLM.
Returns:
result: ChatResult with model generation
"""
if self.streaming:
stream_iter = self._stream(
messages, stop=stop, run_manager=run_manager, **kwargs
)
if stream_iter:
return generate_from_stream(stream_iter)
response = self._handle_request(messages, stop, streaming=False, **kwargs)
message = self._process_response(response)
generation = ChatGeneration(message=message)
return ChatResult(generations=[generation])
def _stream(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> Iterator[ChatGenerationChunk]:
"""
Stream the output of the SambaStudio model.
Args:
messages: the prompt composed of a list of messages.
stop: a list of strings on which the model should stop generating.
If generation stops due to a stop token, the stop token itself
SHOULD BE INCLUDED as part of the output. This is not enforced
across models right now, but it's a good practice to follow since
it makes it much easier to parse the output of the model
downstream and understand why generation stopped.
run_manager: A run manager with callbacks for the LLM.
Yields:
chunk: ChatGenerationChunk with model partial generation
"""
response = self._handle_request(messages, stop, streaming=True, **kwargs)
for ai_message_chunk in self._process_stream_response(response):
chunk = ChatGenerationChunk(message=ai_message_chunk)
if run_manager:
run_manager.on_llm_new_token(chunk.text, chunk=chunk)
yield chunk
