fixes to get optional params

litellm/llms/palm.py

@@ -56,48 +56,44 @@ def completion(
     ## COMPLETION CALL
     response = palm.chat(messages=prompt)
 
-
+    ## LOGGING
-    if "stream" in optional_params and optional_params["stream"] == True:
+    logging_obj.post_call(
-        return response.iter_lines()
+            input=prompt,
-    else:
+            api_key="",
-        ## LOGGING
+            original_response=response,
-        logging_obj.post_call(
+            additional_args={"complete_input_dict": {}},
-                input=prompt,
-                api_key="",
-                original_response=response,
-                additional_args={"complete_input_dict": {}},
-            )
-        print_verbose(f"raw model_response: {response}")
-        ## RESPONSE OBJECT
-        completion_response = response.last
-
-        if "error" in completion_response:
-            raise PalmError(
-                message=completion_response["error"],
-                status_code=response.status_code,
-            )
-        else:
-            try:
-                model_response["choices"][0]["message"]["content"] = completion_response
-            except:
-                raise PalmError(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(
-            encoding.encode(prompt)
-        ) 
-        completion_tokens = len(
-            encoding.encode(model_response["choices"][0]["message"]["content"])
         )
+    print_verbose(f"raw model_response: {response}")
+    ## RESPONSE OBJECT
+    completion_response = response.last
 
-        model_response["created"] = time.time()
+    if "error" in completion_response:
-        model_response["model"] = "palm/" + model
+        raise PalmError(
-        model_response["usage"] = {
+            message=completion_response["error"],
-            "prompt_tokens": prompt_tokens,
+            status_code=response.status_code,
-            "completion_tokens": completion_tokens,
+        )
-            "total_tokens": prompt_tokens + completion_tokens,
+    else:
-        }
+        try:
-        return model_response
+            model_response["choices"][0]["message"]["content"] = completion_response
+        except:
+            raise PalmError(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(
+        encoding.encode(prompt)
+    ) 
+    completion_tokens = len(
+        encoding.encode(model_response["choices"][0]["message"]["content"])
+    )
+
+    model_response["created"] = time.time()
+    model_response["model"] = "palm/" + model
+    model_response["usage"] = {
+        "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

litellm/llms/sagemaker.py

@@ -6,6 +6,7 @@ import time
 from typing import Callable
 from litellm.utils import ModelResponse, get_secret
 import sys
+from copy import deepcopy
 
 class SagemakerError(Exception):
     def __init__(self, status_code, message):
@@ -60,9 +61,12 @@ def completion(
                 )
         else:
             prompt += f"{message['content']}"
+    # pop streaming if it's in the optional params as 'stream' raises an error with sagemaker
+    inference_params = deepcopy(optional_params)
+    inference_params.pop("stream", None)
     data = {
         "inputs": prompt,
-        "parameters": optional_params
+        "parameters": inference_params
     }
 
     ## LOGGING
@@ -79,46 +83,43 @@ def completion(
         CustomAttributes="accept_eula=true",
     )
     response = response["Body"].read().decode("utf8")
-    if "stream" in optional_params and optional_params["stream"] == True:
+    ## LOGGING
-        return response.iter_lines()
+    logging_obj.post_call(
-    else:
+            input=prompt,
-        ## LOGGING
+            api_key="",
-        logging_obj.post_call(
+            original_response=response,
-                input=prompt,
+            additional_args={"complete_input_dict": data},
-                api_key="",
-                original_response=response,
-                additional_args={"complete_input_dict": data},
-            )
-        print_verbose(f"raw model_response: {response}")
-        ## RESPONSE OBJECT
-        completion_response = json.loads(response)
-        if "error" in completion_response:
-            raise SagemakerError(
-                message=completion_response["error"],
-                status_code=response.status_code,
-            )
-        else:
-            try:
-                model_response["choices"][0]["message"]["content"] = completion_response[0]["generation"]
-            except:
-                raise SagemakerError(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(
-            encoding.encode(prompt)
-        ) 
-        completion_tokens = len(
-            encoding.encode(model_response["choices"][0]["message"]["content"])
         )
+    print_verbose(f"raw model_response: {response}")
+    ## RESPONSE OBJECT
+    completion_response = json.loads(response)
+    if "error" in completion_response:
+        raise SagemakerError(
+            message=completion_response["error"],
+            status_code=response.status_code,
+        )
+    else:
+        try:
+            model_response["choices"][0]["message"]["content"] = completion_response[0]["generation"]
+        except:
+            raise SagemakerError(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
+
+    model_response["created"] = time.time()
+    model_response["model"] = model
+    model_response["usage"] = {
+        "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

litellm/main.py

@@ -977,7 +977,6 @@ def completion(
                 encoding=encoding,
                 logging_obj=logging
             )
-
             if "stream" in optional_params and optional_params["stream"]==True: ## [BETA]
                 # sagemaker does not support streaming as of now so we're faking streaming:
                 # https://discuss.huggingface.co/t/streaming-output-text-when-deploying-on-sagemaker/39611

litellm/tests/test_completion.py

@@ -926,7 +926,7 @@ def test_completion_with_fallbacks():
     except Exception as e:
         pytest.fail(f"Error occurred: {e}")
 
-test_completion_with_fallbacks()
+# test_completion_with_fallbacks()
 # def test_completion_with_fallbacks_multiple_keys():
 #     print(f"backup key 1: {os.getenv('BACKUP_OPENAI_API_KEY_1')}")
 #     print(f"backup key 2: {os.getenv('BACKUP_OPENAI_API_KEY_2')}")

litellm/tests/test_streaming.py

@@ -709,7 +709,7 @@ def test_completion_sagemaker_stream():
     except Exception as e:
         pytest.fail(f"Error occurred: {e}")
 
-# test_completion_sagemaker_stream()
+test_completion_sagemaker_stream()
 
 # test on openai completion call
 def test_openai_text_completion_call():

litellm/utils.py

@@ -977,9 +977,9 @@ def get_optional_params(  # use the openai defaults
             raise ValueError("LiteLLM.Exception: Function calling is not supported by this provider")
 
     def _check_valid_arg(supported_params): 
-        print(f"checking params for {model}")
+        print_verbose(f"checking params for {model}")
-        print(f"params passed in {passed_params}")
+        print_verbose(f"params passed in {passed_params}")
-        print(f"non-default params passed in {non_default_params}")
+        print_verbose(f"non-default params passed in {non_default_params}")
         unsupported_params = [k for k in non_default_params.keys() if k not in supported_params]
         if unsupported_params:
             raise ValueError("LiteLLM.Exception: Unsupported parameters passed: {}".format(', '.join(unsupported_params)))
@@ -1225,7 +1225,6 @@ def get_optional_params(  # use the openai defaults
     for k in passed_params.keys(): 
         if k not in default_params.keys(): 
             optional_params[k] = passed_params[k]
-    print(f"final params going to model: {optional_params}")
     return optional_params
 
 def get_llm_provider(model: str, custom_llm_provider: Optional[str] = None):
@@ -3441,14 +3440,15 @@ def completion_with_split_tests(models={}, messages=[], use_client=False, overri
 
 def completion_with_fallbacks(**kwargs):
     print(f"kwargs inside completion_with_fallbacks: {kwargs}")
-    nested_kwargs = kwargs.pop("kwargs")
+    nested_kwargs = kwargs.pop("kwargs", {})
     response = None
     rate_limited_models = set()
     model_expiration_times = {}
     start_time = time.time()
     original_model = kwargs["model"]
-    fallbacks = [kwargs["model"]] + nested_kwargs["fallbacks"]
+    fallbacks = [kwargs["model"]] + nested_kwargs.get("fallbacks", [])
-    del nested_kwargs["fallbacks"]  # remove fallbacks so it's not recursive
+    if "fallbacks" in nested_kwargs:
+        del nested_kwargs["fallbacks"]  # remove fallbacks so it's not recursive
 
     while response == None and time.time() - start_time < 45:
         for model in fallbacks:

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "litellm"
-version = "0.1.811"
+version = "0.1.812"
 description = "Library to easily interface with LLM API providers"
 authors = ["BerriAI"]
 license = "MIT License"