mirror of
https://github.com/crewAIInc/crewAI.git
synced 2026-01-23 07:08:14 +00:00
Compare commits
5 Commits
devin/1768
...
devin/1740
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
c956588586 | ||
|
|
e8d61d32db | ||
|
|
1e7292d0fa | ||
|
|
b7c988b3ac | ||
|
|
6d4c591eda |
@@ -92,9 +92,43 @@ def suppress_warnings():
|
|||||||
|
|
||||||
|
|
||||||
class LLM:
|
class LLM:
|
||||||
|
"""
|
||||||
|
A wrapper class for language model interactions using litellm.
|
||||||
|
|
||||||
|
This class provides a unified interface for interacting with various language models
|
||||||
|
through litellm. It handles model configuration, context window sizing, and callback
|
||||||
|
management.
|
||||||
|
|
||||||
|
Args:
|
||||||
|
model (str): The identifier for the language model to use. Must be a valid model ID
|
||||||
|
with a provider prefix (e.g., 'openai/gpt-4'). Cannot be a numeric value without
|
||||||
|
a provider prefix.
|
||||||
|
timeout (Optional[Union[float, int]]): The timeout for API calls in seconds.
|
||||||
|
temperature (Optional[float]): Controls randomness in the model's output.
|
||||||
|
top_p (Optional[float]): Controls diversity via nucleus sampling.
|
||||||
|
n (Optional[int]): Number of completions to generate.
|
||||||
|
stop (Optional[Union[str, List[str]]]): Sequences where the model should stop generating.
|
||||||
|
max_completion_tokens (Optional[int]): Maximum number of tokens to generate.
|
||||||
|
max_tokens (Optional[int]): Alias for max_completion_tokens.
|
||||||
|
presence_penalty (Optional[float]): Penalizes repeated tokens.
|
||||||
|
frequency_penalty (Optional[float]): Penalizes frequent tokens.
|
||||||
|
logit_bias (Optional[Dict[int, float]]): Modifies likelihood of specific tokens.
|
||||||
|
response_format (Optional[Dict[str, Any]]): Specifies the format for the model's response.
|
||||||
|
seed (Optional[int]): Seed for deterministic outputs.
|
||||||
|
logprobs (Optional[bool]): Whether to return log probabilities.
|
||||||
|
top_logprobs (Optional[int]): Number of most likely tokens to return probabilities for.
|
||||||
|
base_url (Optional[str]): Base URL for API calls.
|
||||||
|
api_version (Optional[str]): API version to use.
|
||||||
|
api_key (Optional[str]): API key for authentication.
|
||||||
|
callbacks (List[Any]): List of callback functions.
|
||||||
|
**kwargs: Additional keyword arguments to pass to the model.
|
||||||
|
|
||||||
|
Raises:
|
||||||
|
ValueError: If the model ID is empty, whitespace, or a numeric value without a provider prefix.
|
||||||
|
"""
|
||||||
def __init__(
|
def __init__(
|
||||||
self,
|
self,
|
||||||
model: str,
|
model: Union[str, Any],
|
||||||
timeout: Optional[Union[float, int]] = None,
|
timeout: Optional[Union[float, int]] = None,
|
||||||
temperature: Optional[float] = None,
|
temperature: Optional[float] = None,
|
||||||
top_p: Optional[float] = None,
|
top_p: Optional[float] = None,
|
||||||
@@ -115,6 +149,16 @@ class LLM:
|
|||||||
callbacks: List[Any] = [],
|
callbacks: List[Any] = [],
|
||||||
**kwargs,
|
**kwargs,
|
||||||
):
|
):
|
||||||
|
# Only validate model ID if it's not None and is a numeric value without a provider prefix
|
||||||
|
if model is not None and (
|
||||||
|
isinstance(model, (int, float)) or
|
||||||
|
(isinstance(model, str) and model.strip() and model.strip().isdigit())
|
||||||
|
):
|
||||||
|
raise ValueError(
|
||||||
|
f"Invalid model ID: {model}. Model ID cannot be a numeric value without a provider prefix. "
|
||||||
|
"Please specify a valid model ID with a provider prefix, e.g., 'openai/gpt-4'."
|
||||||
|
)
|
||||||
|
|
||||||
self.model = model
|
self.model = model
|
||||||
self.timeout = timeout
|
self.timeout = timeout
|
||||||
self.temperature = temperature
|
self.temperature = temperature
|
||||||
@@ -186,7 +230,10 @@ class LLM:
|
|||||||
|
|
||||||
def supports_function_calling(self) -> bool:
|
def supports_function_calling(self) -> bool:
|
||||||
try:
|
try:
|
||||||
params = get_supported_openai_params(model=self.model)
|
# Handle None model case
|
||||||
|
if self.model is None:
|
||||||
|
return False
|
||||||
|
params = get_supported_openai_params(model=str(self.model))
|
||||||
return "response_format" in params
|
return "response_format" in params
|
||||||
except Exception as e:
|
except Exception as e:
|
||||||
logging.error(f"Failed to get supported params: {str(e)}")
|
logging.error(f"Failed to get supported params: {str(e)}")
|
||||||
@@ -194,7 +241,10 @@ class LLM:
|
|||||||
|
|
||||||
def supports_stop_words(self) -> bool:
|
def supports_stop_words(self) -> bool:
|
||||||
try:
|
try:
|
||||||
params = get_supported_openai_params(model=self.model)
|
# Handle None model case
|
||||||
|
if self.model is None:
|
||||||
|
return False
|
||||||
|
params = get_supported_openai_params(model=str(self.model))
|
||||||
return "stop" in params
|
return "stop" in params
|
||||||
except Exception as e:
|
except Exception as e:
|
||||||
logging.error(f"Failed to get supported params: {str(e)}")
|
logging.error(f"Failed to get supported params: {str(e)}")
|
||||||
@@ -208,8 +258,10 @@ class LLM:
|
|||||||
self.context_window_size = int(
|
self.context_window_size = int(
|
||||||
DEFAULT_CONTEXT_WINDOW_SIZE * CONTEXT_WINDOW_USAGE_RATIO
|
DEFAULT_CONTEXT_WINDOW_SIZE * CONTEXT_WINDOW_USAGE_RATIO
|
||||||
)
|
)
|
||||||
|
# Ensure model is a string before calling startswith
|
||||||
|
model_str = str(self.model) if not isinstance(self.model, str) else self.model
|
||||||
for key, value in LLM_CONTEXT_WINDOW_SIZES.items():
|
for key, value in LLM_CONTEXT_WINDOW_SIZES.items():
|
||||||
if self.model.startswith(key):
|
if model_str.startswith(key):
|
||||||
self.context_window_size = int(value * CONTEXT_WINDOW_USAGE_RATIO)
|
self.context_window_size = int(value * CONTEXT_WINDOW_USAGE_RATIO)
|
||||||
return self.context_window_size
|
return self.context_window_size
|
||||||
|
|
||||||
|
|||||||
43
tests/unit/test_llm.py
Normal file
43
tests/unit/test_llm.py
Normal file
@@ -0,0 +1,43 @@
|
|||||||
|
import pytest
|
||||||
|
|
||||||
|
from crewai.llm import LLM
|
||||||
|
|
||||||
|
|
||||||
|
@pytest.mark.parametrize(
|
||||||
|
"invalid_model,error_message",
|
||||||
|
[
|
||||||
|
(3420, "Invalid model ID: 3420. Model ID cannot be a numeric value without a provider prefix."),
|
||||||
|
("3420", "Invalid model ID: 3420. Model ID cannot be a numeric value without a provider prefix."),
|
||||||
|
(3.14, "Invalid model ID: 3.14. Model ID cannot be a numeric value without a provider prefix."),
|
||||||
|
],
|
||||||
|
)
|
||||||
|
def test_invalid_numeric_model_ids(invalid_model, error_message):
|
||||||
|
"""Test that numeric model IDs are rejected."""
|
||||||
|
with pytest.raises(ValueError, match=error_message):
|
||||||
|
LLM(model=invalid_model)
|
||||||
|
|
||||||
|
|
||||||
|
@pytest.mark.parametrize(
|
||||||
|
"valid_model",
|
||||||
|
[
|
||||||
|
"openai/gpt-4",
|
||||||
|
"gpt-3.5-turbo",
|
||||||
|
"anthropic/claude-2",
|
||||||
|
],
|
||||||
|
)
|
||||||
|
def test_valid_model_ids(valid_model):
|
||||||
|
"""Test that valid model IDs are accepted."""
|
||||||
|
llm = LLM(model=valid_model)
|
||||||
|
assert llm.model == valid_model
|
||||||
|
|
||||||
|
|
||||||
|
def test_empty_model_id():
|
||||||
|
"""Test that empty model IDs are rejected."""
|
||||||
|
with pytest.raises(ValueError, match="Invalid model ID: ''. Model ID cannot be empty or whitespace."):
|
||||||
|
LLM(model="")
|
||||||
|
|
||||||
|
|
||||||
|
def test_whitespace_model_id():
|
||||||
|
"""Test that whitespace model IDs are rejected."""
|
||||||
|
with pytest.raises(ValueError, match="Invalid model ID: ' '. Model ID cannot be empty or whitespace."):
|
||||||
|
LLM(model=" ")
|
||||||
Reference in New Issue
Block a user