phoenix/litellm
1
0
Fork 0
forked from phoenix/litellm-mirror
Code Pull requests Activity

Merge branch 'main' into litellm_proxy_team_cache_update

Browse source
This commit is contained in:
Krish Dholakia 2024-07-19 21:07:26 -07:00 • committed by GitHub
commit f797597202
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
15 changed files with 222 additions and 61 deletions
Download patch file Download diff file Expand all files Collapse all files

110
litellm/llms/bedrock_httpx.py
View file

@ -75,6 +75,7 @@ BEDROCK_CONVERSE_MODELS = [
"anthropic.claude-v2:1",
"anthropic.claude-v1",
"anthropic.claude-instant-v1",
"ai21.jamba-instruct-v1:0",
]
@ -195,13 +196,39 @@ async def make_call(
if client is None:
client = _get_async_httpx_client() # Create a new client if none provided
response = await client.post(api_base, headers=headers, data=data, stream=True)
response = await client.post(
api_base,
headers=headers,
data=data,
stream=True if "ai21" not in api_base else False,
)
if response.status_code != 200:
raise BedrockError(status_code=response.status_code, message=response.text)
if "ai21" in api_base:
aws_bedrock_process_response = BedrockConverseLLM()
model_response: (
ModelResponse
) = aws_bedrock_process_response.process_response(
model=model,
response=response,
model_response=litellm.ModelResponse(),
stream=True,
logging_obj=logging_obj,
optional_params={},
api_key="",
data=data,
messages=messages,
print_verbose=litellm.print_verbose,
encoding=litellm.encoding,
) # type: ignore
completion_stream: Any = MockResponseIterator(model_response=model_response)
else:
decoder = AWSEventStreamDecoder(model=model)
completion_stream = decoder.aiter_bytes(response.aiter_bytes(chunk_size=1024))
completion_stream = decoder.aiter_bytes(
response.aiter_bytes(chunk_size=1024)
)
# LOGGING
logging_obj.post_call(
@ -233,11 +260,33 @@ def make_sync_call(
if client is None:
client = _get_httpx_client() # Create a new client if none provided
response = client.post(api_base, headers=headers, data=data, stream=True)
response = client.post(
api_base,
headers=headers,
data=data,
stream=True if "ai21" not in api_base else False,
)
if response.status_code != 200:
raise BedrockError(status_code=response.status_code, message=response.read())
if "ai21" in api_base:
aws_bedrock_process_response = BedrockConverseLLM()
model_response: ModelResponse = aws_bedrock_process_response.process_response(
model=model,
response=response,
model_response=litellm.ModelResponse(),
stream=True,
logging_obj=logging_obj,
optional_params={},
api_key="",
data=data,
messages=messages,
print_verbose=litellm.print_verbose,
encoding=litellm.encoding,
) # type: ignore
completion_stream: Any = MockResponseIterator(model_response=model_response)
else:
decoder = AWSEventStreamDecoder(model=model)
completion_stream = decoder.iter_bytes(response.iter_bytes(chunk_size=1024))
@ -1348,7 +1397,7 @@ class BedrockConverseLLM(BaseLLM):
response: Union[requests.Response, httpx.Response],
model_response: ModelResponse,
stream: bool,
logging_obj: Logging,
logging_obj: Optional[Logging],
optional_params: dict,
api_key: str,
data: Union[dict, str],
@ -1358,6 +1407,7 @@ class BedrockConverseLLM(BaseLLM):
) -> Union[ModelResponse, CustomStreamWrapper]:
## LOGGING
if logging_obj is not None:
logging_obj.post_call(
input=messages,
api_key=api_key,
@ -1900,7 +1950,7 @@ class BedrockConverseLLM(BaseLLM):
if acompletion:
if isinstance(client, HTTPHandler):
client = None
if stream is True and provider != "ai21":
if stream is True:
return self.async_streaming(
model=model,
messages=messages,
@ -1937,7 +1987,7 @@ class BedrockConverseLLM(BaseLLM):
client=client,
) # type: ignore
if (stream is not None and stream is True) and provider != "ai21":
if stream is not None and stream is True:
streaming_response = CustomStreamWrapper(
completion_stream=None,
@ -1981,7 +2031,7 @@ class BedrockConverseLLM(BaseLLM):
model=model,
response=response,
model_response=model_response,
stream=stream,
stream=stream if isinstance(stream, bool) else False,
logging_obj=logging_obj,
optional_params=optional_params,
api_key="",
@ -2168,3 +2218,49 @@ class AWSEventStreamDecoder:
return None
return chunk.decode() # type: ignore[no-any-return]
class MockResponseIterator: # for returning ai21 streaming responses
def __init__(self, model_response):
self.model_response = model_response
self.is_done = False
# Sync iterator
def __iter__(self):
return self
def _chunk_parser(self, chunk_data: ModelResponse) -> GenericStreamingChunk:
try:
chunk_usage: litellm.Usage = getattr(chunk_data, "usage")
processed_chunk = GenericStreamingChunk(
text=chunk_data.choices[0].message.content or "", # type: ignore
tool_use=None,
is_finished=True,
finish_reason=chunk_data.choices[0].finish_reason, # type: ignore
usage=ConverseTokenUsageBlock(
inputTokens=chunk_usage.prompt_tokens,
outputTokens=chunk_usage.completion_tokens,
totalTokens=chunk_usage.total_tokens,
),
index=0,
)
return processed_chunk
except Exception:
raise ValueError(f"Failed to decode chunk: {chunk_data}")
def __next__(self):
if self.is_done:
raise StopIteration
self.is_done = True
return self._chunk_parser(self.model_response)
# Async iterator
def __aiter__(self):
return self
async def __anext__(self):
if self.is_done:
raise StopAsyncIteration
self.is_done = True
return self._chunk_parser(self.model_response)

7
litellm/main.py
View file

@ -4798,9 +4798,10 @@ async def ahealth_check(
if isinstance(stack_trace, str):
stack_trace = stack_trace[:1000]
if model not in litellm.model_cost and mode is None:
raise Exception(
"Missing `mode`. Set the `mode` for the model - https://docs.litellm.ai/docs/proxy/health#embedding-models"
)
return {
"error": "Missing `mode`. Set the `mode` for the model - https://docs.litellm.ai/docs/proxy/health#embedding-models"
}
error_to_return = str(e) + " stack trace: " + stack_trace
return {"error": error_to_return}

10
litellm/model_prices_and_context_window_backup.json
View file

@ -2803,6 +2803,16 @@
"litellm_provider": "bedrock",
"mode": "chat"
},
"ai21.jamba-instruct-v1:0": {
"max_tokens": 4096,
"max_input_tokens": 70000,
"max_output_tokens": 4096,
"input_cost_per_token": 0.0000005,
"output_cost_per_token": 0.0000007,
"litellm_provider": "bedrock",
"mode": "chat",
"supports_system_messages": true
},
"amazon.titan-text-lite-v1": {
"max_tokens": 4000,
"max_input_tokens": 42000,

2
litellm/proxy/proxy_config.yaml
View file

@ -4,5 +4,7 @@ model_list:
model: fireworks_ai/accounts/fireworks/models/llama-v3-70b-instruct
api_key: "os.environ/FIREWORKS_AI_API_KEY"
router_settings:
enable_tag_filtering: True # 👈 Key Change
general_settings:
master_key: sk-1234

2
litellm/tests/test_bedrock_completion.py
View file

@ -592,6 +592,8 @@ def test_bedrock_claude_3(image_url):
assert len(response.choices) > 0
assert len(response.choices[0].message.content) > 0
except litellm.InternalServerError:
pass
except RateLimitError:
pass
except Exception as e:

8
litellm/tests/test_completion.py
View file

@ -1348,7 +1348,10 @@ def test_completion_fireworks_ai():
pytest.fail(f"Error occurred: {e}")
def test_completion_fireworks_ai_bad_api_base():
@pytest.mark.parametrize(
"api_key, api_base", [(None, "my-bad-api-base"), ("my-bad-api-key", None)]
)
def test_completion_fireworks_ai_dynamic_params(api_key, api_base):
try:
litellm.set_verbose = True
messages = [
@ -1361,7 +1364,8 @@ def test_completion_fireworks_ai_bad_api_base():
response = completion(
model="fireworks_ai/accounts/fireworks/models/mixtral-8x7b-instruct",
messages=messages,
api_base="my-bad-api-base",
api_base=api_base,
api_key=api_key,
)
pytest.fail(f"This call should have failed!")
except Exception as e:

4
litellm/tests/test_completion_cost.py
View file

@ -706,9 +706,9 @@ def test_vertex_ai_completion_cost():
print("calculated_input_cost: {}".format(calculated_input_cost))
# @pytest.mark.skip(reason="new test - WIP, working on fixing this")
@pytest.mark.skip(reason="new test - WIP, working on fixing this")
def test_vertex_ai_medlm_completion_cost():
"""Test for medlm completion cost."""
"""Test for medlm completion cost ."""
with pytest.raises(Exception) as e:
model = "vertex_ai/medlm-medium"

1
litellm/tests/test_exceptions.py
View file

@ -90,6 +90,7 @@ def test_context_window(model):
models = ["command-nightly"]
@pytest.mark.skip(reason="duplicate test.")
@pytest.mark.parametrize("model", models)
def test_context_window_with_fallbacks(model):
ctx_window_fallback_dict = {

16
litellm/tests/test_router_timeout.py
View file

@ -1,8 +1,12 @@
#### What this tests ####
# This tests if the router timeout error handling during fallbacks
import sys, os, time
import traceback, asyncio
import asyncio
import os
import sys
import time
import traceback
import pytest
sys.path.insert(
@ -12,9 +16,10 @@ sys.path.insert(
import os
from dotenv import load_dotenv
import litellm
from litellm import Router
from dotenv import load_dotenv
load_dotenv()
@ -37,6 +42,7 @@ def test_router_timeouts():
"litellm_params": {
"model": "claude-instant-1.2",
"api_key": "os.environ/ANTHROPIC_API_KEY",
"mock_response": "hello world",
},
"tpm": 20000,
},
@ -90,7 +96,9 @@ def test_router_timeouts():
@pytest.mark.asyncio
async def test_router_timeouts_bedrock():
import openai, uuid
import uuid
import openai
# Model list for OpenAI and Anthropic models
_model_list = [

24
litellm/tests/test_streaming.py
View file

@ -1312,22 +1312,22 @@ async def test_completion_replicate_llama3_streaming(sync_mode):
# pytest.fail(f"Error occurred: {e}")
@pytest.mark.parametrize("sync_mode", [True]) # False
@pytest.mark.parametrize("sync_mode", [True, False]) #
@pytest.mark.parametrize(
"model",
"model, region",
[
"bedrock/cohere.command-r-plus-v1:0",
"anthropic.claude-3-sonnet-20240229-v1:0",
"anthropic.claude-instant-v1",
"bedrock/ai21.j2-mid",
"mistral.mistral-7b-instruct-v0:2",
"bedrock/amazon.titan-tg1-large",
"meta.llama3-8b-instruct-v1:0",
"cohere.command-text-v14",
["bedrock/ai21.jamba-instruct-v1:0", "us-east-1"],
["bedrock/cohere.command-r-plus-v1:0", None],
["anthropic.claude-3-sonnet-20240229-v1:0", None],
["anthropic.claude-instant-v1", None],
["mistral.mistral-7b-instruct-v0:2", None],
["bedrock/amazon.titan-tg1-large", None],
["meta.llama3-8b-instruct-v1:0", None],
["cohere.command-text-v14", None],
],
)
@pytest.mark.asyncio
async def test_bedrock_httpx_streaming(sync_mode, model):
async def test_bedrock_httpx_streaming(sync_mode, model, region):
try:
litellm.set_verbose = True
if sync_mode:
@ -1337,6 +1337,7 @@ async def test_bedrock_httpx_streaming(sync_mode, model):
messages=messages,
max_tokens=10, # type: ignore
stream=True,
aws_region_name=region,
)
complete_response = ""
# Add any assertions here to check the response
@ -1358,6 +1359,7 @@ async def test_bedrock_httpx_streaming(sync_mode, model):
messages=messages,
max_tokens=100, # type: ignore
stream=True,
aws_region_name=region,
)
complete_response = ""
# Add any assertions here to check the response

26
litellm/tests/test_token_counter.py
View file

@ -20,7 +20,12 @@ from litellm import (
token_counter,
)
from litellm.tests.large_text import text
from litellm.tests.messages_with_counts import MESSAGES_TEXT, MESSAGES_WITH_IMAGES, MESSAGES_WITH_TOOLS
from litellm.tests.messages_with_counts import (
MESSAGES_TEXT,
MESSAGES_WITH_IMAGES,
MESSAGES_WITH_TOOLS,
)
def test_token_counter_normal_plus_function_calling():
try:
@ -55,27 +60,28 @@ def test_token_counter_normal_plus_function_calling():
except Exception as e:
pytest.fail(f"An exception occurred - {str(e)}")
# test_token_counter_normal_plus_function_calling()
@pytest.mark.parametrize(
"message_count_pair",
MESSAGES_TEXT,
)
def test_token_counter_textonly(message_count_pair):
counted_tokens = token_counter(
model="gpt-35-turbo",
messages=[message_count_pair["message"]]
model="gpt-35-turbo", messages=[message_count_pair["message"]]
)
assert counted_tokens == message_count_pair["count"]
@pytest.mark.parametrize(
"message_count_pair",
MESSAGES_WITH_IMAGES,
)
def test_token_counter_with_images(message_count_pair):
counted_tokens = token_counter(
model="gpt-4o",
messages=[message_count_pair["message"]]
model="gpt-4o", messages=[message_count_pair["message"]]
)
assert counted_tokens == message_count_pair["count"]
@ -327,3 +333,13 @@ def test_get_modified_max_tokens(
), "Got={}, Expected={}, Params={}".format(
calculated_value, expected_value, args
)
def test_empty_tools():
messages = [{"role": "user", "content": "hey, how's it going?", "tool_calls": None}]
result = token_counter(
messages=messages,
)
print(result)

41
litellm/utils.py
View file

@ -1911,7 +1911,7 @@ def token_counter(
# use tiktoken, anthropic, cohere, llama2, or llama3's tokenizer depending on the model
is_tool_call = False
num_tokens = 0
if text == None:
if text is None:
if messages is not None:
print_verbose(f"token_counter messages received: {messages}")
text = ""
@ -1937,7 +1937,7 @@ def token_counter(
num_tokens += calculage_img_tokens(
data=image_url_str, mode="auto"
)
if "tool_calls" in message:
if message.get("tool_calls"):
is_tool_call = True
for tool_call in message["tool_calls"]:
if "function" in tool_call:
@ -4398,44 +4398,44 @@ def get_llm_provider(
if custom_llm_provider == "perplexity":
# perplexity is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.perplexity.ai
api_base = api_base or "https://api.perplexity.ai"
dynamic_api_key = get_secret("PERPLEXITYAI_API_KEY")
dynamic_api_key = api_key or get_secret("PERPLEXITYAI_API_KEY")
elif custom_llm_provider == "anyscale":
# anyscale is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.endpoints.anyscale.com/v1
api_base = api_base or "https://api.endpoints.anyscale.com/v1"
dynamic_api_key = get_secret("ANYSCALE_API_KEY")
dynamic_api_key = api_key or get_secret("ANYSCALE_API_KEY")
elif custom_llm_provider == "deepinfra":
# deepinfra is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.endpoints.anyscale.com/v1
api_base = api_base or "https://api.deepinfra.com/v1/openai"
dynamic_api_key = get_secret("DEEPINFRA_API_KEY")
dynamic_api_key = api_key or get_secret("DEEPINFRA_API_KEY")
elif custom_llm_provider == "empower":
api_base = api_base or "https://app.empower.dev/api/v1"
dynamic_api_key = get_secret("EMPOWER_API_KEY")
dynamic_api_key = api_key or get_secret("EMPOWER_API_KEY")
elif custom_llm_provider == "groq":
# groq is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.groq.com/openai/v1
api_base = api_base or "https://api.groq.com/openai/v1"
dynamic_api_key = get_secret("GROQ_API_KEY")
dynamic_api_key = api_key or get_secret("GROQ_API_KEY")
elif custom_llm_provider == "nvidia_nim":
# nvidia_nim is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.endpoints.anyscale.com/v1
api_base = api_base or "https://integrate.api.nvidia.com/v1"
dynamic_api_key = get_secret("NVIDIA_NIM_API_KEY")
dynamic_api_key = api_key or get_secret("NVIDIA_NIM_API_KEY")
elif custom_llm_provider == "volcengine":
# volcengine is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.endpoints.anyscale.com/v1
api_base = api_base or "https://ark.cn-beijing.volces.com/api/v3"
dynamic_api_key = get_secret("VOLCENGINE_API_KEY")
dynamic_api_key = api_key or get_secret("VOLCENGINE_API_KEY")
elif custom_llm_provider == "codestral":
# codestral is openai compatible, we just need to set this to custom_openai and have the api_base be https://codestral.mistral.ai/v1
api_base = api_base or "https://codestral.mistral.ai/v1"
dynamic_api_key = get_secret("CODESTRAL_API_KEY")
dynamic_api_key = api_key or get_secret("CODESTRAL_API_KEY")
elif custom_llm_provider == "deepseek":
# deepseek is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.deepseek.com/v1
api_base = api_base or "https://api.deepseek.com/v1"
dynamic_api_key = get_secret("DEEPSEEK_API_KEY")
dynamic_api_key = api_key or get_secret("DEEPSEEK_API_KEY")
elif custom_llm_provider == "fireworks_ai":
# fireworks is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.fireworks.ai/inference/v1
if not model.startswith("accounts/fireworks/models"):
model = f"accounts/fireworks/models/{model}"
api_base = api_base or "https://api.fireworks.ai/inference/v1"
dynamic_api_key = (
dynamic_api_key = api_key or (
get_secret("FIREWORKS_API_KEY")
or get_secret("FIREWORKS_AI_API_KEY")
or get_secret("FIREWORKSAI_API_KEY")
@ -4465,10 +4465,10 @@ def get_llm_provider(
elif custom_llm_provider == "voyage":
# voyage is openai compatible, we just need to set this to custom_openai and have the api_base be https://api.voyageai.com/v1
api_base = "https://api.voyageai.com/v1"
dynamic_api_key = get_secret("VOYAGE_API_KEY")
dynamic_api_key = api_key or get_secret("VOYAGE_API_KEY")
elif custom_llm_provider == "together_ai":
api_base = "https://api.together.xyz/v1"
dynamic_api_key = (
dynamic_api_key = api_key or (
get_secret("TOGETHER_API_KEY")
or get_secret("TOGETHER_AI_API_KEY")
or get_secret("TOGETHERAI_API_KEY")
@ -4476,8 +4476,10 @@ def get_llm_provider(
)
elif custom_llm_provider == "friendliai":
api_base = "https://inference.friendli.ai/v1"
dynamic_api_key = get_secret("FRIENDLIAI_API_KEY") or get_secret(
"FRIENDLI_TOKEN"
dynamic_api_key = (
api_key
or get_secret("FRIENDLIAI_API_KEY")
or get_secret("FRIENDLI_TOKEN")
)
if api_base is not None and not isinstance(api_base, str):
raise Exception(
@ -6813,6 +6815,13 @@ def exception_type(
model=model,
llm_provider="bedrock",
)
elif "Could not process image" in error_str:
exception_mapping_worked = True
raise litellm.InternalServerError(
message=f"BedrockException - {error_str}",
model=model,
llm_provider="bedrock",
)
elif hasattr(original_exception, "status_code"):
if original_exception.status_code == 500:
exception_mapping_worked = True

10
model_prices_and_context_window.json
View file

@ -2803,6 +2803,16 @@
"litellm_provider": "bedrock",
"mode": "chat"
},
"ai21.jamba-instruct-v1:0": {
"max_tokens": 4096,
"max_input_tokens": 70000,
"max_output_tokens": 4096,
"input_cost_per_token": 0.0000005,
"output_cost_per_token": 0.0000007,
"litellm_provider": "bedrock",
"mode": "chat",
"supports_system_messages": true
},
"amazon.titan-text-lite-v1": {
"max_tokens": 4000,
"max_input_tokens": 42000,

2
ui/litellm-dashboard/src/components/leftnav.tsx
View file

@ -48,7 +48,7 @@ const Sidebar: React.FC<SidebarProps> = ({
style={{ height: "100%", borderRight: 0 }}
>
<Menu.Item key="1" onClick={() => setPage("api-keys")}>
<Text>API Keys</Text>
<Text>Virtual Keys</Text>
</Menu.Item>
<Menu.Item key="3" onClick={() => setPage("llm-playground")}>
<Text>Test Key</Text>