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

clean out AI21 Init Client calls

Browse source
This commit is contained in:
ishaan-jaff 2023-09-04 10:08:53 -07:00
parent f2b0fa90ab
commit 38564ddc82
2 changed files with 94 additions and 107 deletions
Download patch file Download diff file Expand all files Collapse all files

188
litellm/llms/ai21.py
View file

@ -1,4 +1,5 @@
import os, json import os
import json
from enum import Enum from enum import Enum
import requests import requests
import time import time
@ -13,115 +14,102 @@ class AI21Error(Exception):
self.message self.message
) # Call the base class constructor with the parameters it needs ) # Call the base class constructor with the parameters it needs
def validate_environment(api_key):
if api_key is None:
raise ValueError(
"Missing AI21 API Key - A call is being made to ai21 but no key is set either in the environment variables or via params"
)
headers = {
"accept": "application/json",
"content-type": "application/json",
"Authorization": "Bearer " + api_key,
}
return headers
class AI21LLM: def completion(
def __init__( model: str,
self, encoding, logging_obj, api_key=None messages: list,
): model_response: ModelResponse,
self.encoding = encoding print_verbose: Callable,
self.completion_url_fragment_1 = "https://api.ai21.com/studio/v1/" encoding,
self.completion_url_fragment_2 = "/complete" api_key,
self.api_key = api_key logging_obj,
self.logging_obj = logging_obj optional_params=None,
self.validate_environment(api_key=api_key) litellm_params=None,
logger_fn=None,
def validate_environment( ):
self, api_key headers = validate_environment(api_key)
): # set up the environment required to run the model model = model
# set the api key prompt = ""
if self.api_key == None: for message in messages:
raise ValueError( if "role" in message:
"Missing AI21 API Key - A call is being made to ai21 but no key is set either in the environment variables or via params" if message["role"] == "user":
) prompt += (
self.api_key = api_key f"{message['content']}"
self.headers = { )
"accept": "application/json",
"content-type": "application/json",
"Authorization": "Bearer " + self.api_key,
}
def completion(
self,
model: str,
messages: list,
model_response: ModelResponse,
print_verbose: Callable,
optional_params=None,
litellm_params=None,
logger_fn=None,
): # logic for parsing in - calling - parsing out model completion calls
model = model
prompt = ""
for message in messages:
if "role" in message:
if message["role"] == "user":
prompt += (
f"{message['content']}"
)
else:
prompt += (
f"{message['content']}"
)
else: else:
prompt += f"{message['content']}" prompt += (
data = { f"{message['content']}"
"prompt": prompt, )
# "instruction": prompt, # some baseten models require the prompt to be passed in via the 'instruction' kwarg else:
**optional_params, prompt += f"{message['content']}"
} data = {
"prompt": prompt,
# "instruction": prompt, # some baseten models require the prompt to be passed in via the 'instruction' kwarg
**optional_params,
}
## LOGGING ## LOGGING
self.logging_obj.pre_call( logging_obj.pre_call(
input=prompt, input=prompt,
api_key=self.api_key, api_key=api_key,
additional_args={"complete_input_dict": data}, additional_args={"complete_input_dict": data},
) )
## COMPLETION CALL ## COMPLETION CALL
response = requests.post( response = requests.post(
self.completion_url_fragment_1 + model + self.completion_url_fragment_2, headers=self.headers, data=json.dumps(data) "https://api.ai21.com/studio/v1/" + model + "/complete", headers=headers, data=json.dumps(data)
) )
if "stream" in optional_params and optional_params["stream"] == True: if "stream" in optional_params and optional_params["stream"] == True:
return response.iter_lines() return response.iter_lines()
else: else:
## LOGGING ## LOGGING
self.logging_obj.post_call( logging_obj.post_call(
input=prompt, input=prompt,
api_key=self.api_key, api_key=api_key,
original_response=response.text, original_response=response.text,
additional_args={"complete_input_dict": data}, additional_args={"complete_input_dict": data},
) )
print_verbose(f"raw model_response: {response.text}") print_verbose(f"raw model_response: {response.text}")
## RESPONSE OBJECT ## RESPONSE OBJECT
completion_response = response.json() completion_response = response.json()
if "error" in completion_response: if "error" in completion_response:
raise AI21Error( raise AI21Error(
message=completion_response["error"], message=completion_response["error"],
status_code=response.status_code, status_code=response.status_code,
)
else:
try:
model_response["choices"][0]["message"]["content"] = completion_response["completions"][0]["data"]["text"]
except:
raise AI21Error(message=json.dumps(completion_response), status_code=response.status_code)
## CALCULATING USAGE - baseten charges on time, not tokens - have some mapping of cost here.
prompt_tokens = len(
self.encoding.encode(prompt)
)
completion_tokens = len(
self.encoding.encode(model_response["choices"][0]["message"]["content"])
) )
else:
try:
model_response["choices"][0]["message"]["content"] = completion_response["completions"][0]["data"]["text"]
except:
raise AI21Error(message=json.dumps(completion_response), status_code=response.status_code)
model_response["created"] = time.time() ## CALCULATING USAGE - baseten charges on time, not tokens - have some mapping of cost here.
model_response["model"] = model prompt_tokens = len(
model_response["usage"] = { encoding.encode(prompt)
"prompt_tokens": prompt_tokens, )
"completion_tokens": completion_tokens, completion_tokens = len(
"total_tokens": prompt_tokens + completion_tokens, encoding.encode(model_response["choices"][0]["message"]["content"])
} )
return model_response
def embedding( model_response["created"] = time.time()
self, model_response["model"] = model
): # logic for parsing in - calling - parsing out model embedding calls model_response["usage"] = {
pass "prompt_tokens": prompt_tokens,
"completion_tokens": completion_tokens,
"total_tokens": prompt_tokens + completion_tokens,
}
return model_response
def embedding():
# logic for parsing in - calling - parsing out model embedding calls
pass

13
litellm/main.py
View file

@ -21,9 +21,9 @@ from litellm.utils import (
) )
from .llms import anthropic from .llms import anthropic
from .llms import together_ai from .llms import together_ai
from .llms import ai21
from .llms.huggingface_restapi import HuggingfaceRestAPILLM from .llms.huggingface_restapi import HuggingfaceRestAPILLM
from .llms.baseten import BasetenLLM from .llms.baseten import BasetenLLM
from .llms.ai21 import AI21LLM
from .llms.aleph_alpha import AlephAlphaLLM from .llms.aleph_alpha import AlephAlphaLLM
import tiktoken import tiktoken
from concurrent.futures import ThreadPoolExecutor from concurrent.futures import ThreadPoolExecutor
@ -657,12 +657,8 @@ def completion(
api_key api_key
or litellm.ai21_key or litellm.ai21_key
or os.environ.get("AI21_API_KEY") or os.environ.get("AI21_API_KEY")
) )
ai21_client = AI21LLM( model_response = ai21.completion(
encoding=encoding, api_key=ai21_key, logging_obj=logging
)
model_response = ai21_client.completion(
model=model, model=model,
messages=messages, messages=messages,
model_response=model_response, model_response=model_response,
@ -670,6 +666,9 @@ def completion(
optional_params=optional_params, optional_params=optional_params,
litellm_params=litellm_params, litellm_params=litellm_params,
logger_fn=logger_fn, logger_fn=logger_fn,
encoding=encoding,
api_key=ai21_key,
logging_obj=logging
) )
if "stream" in optional_params and optional_params["stream"] == True: if "stream" in optional_params and optional_params["stream"] == True: