adding support for vllm

litellm/__init__.py

@@ -195,6 +195,7 @@ provider_list = [
     "azure",
     "sagemaker",
     "bedrock",
+    "vllm",
     "custom", # custom apis
 ]
 

litellm/llms/huggingface_restapi.py

@@ -54,8 +54,8 @@ def completion(
         model_prompt_details = custom_prompt_dict[model]
         prompt = custom_prompt(
             role_dict=model_prompt_details["roles"], 
-            initial_prompt_value=model_prompt_details["pre_message_sep"],  
+            initial_prompt_value=model_prompt_details["initial_prompt_value"],  
-            final_prompt_value=model_prompt_details["post_message_sep"], 
+            final_prompt_value=model_prompt_details["final_prompt_value"], 
             messages=messages
         )
     else:

litellm/llms/together_ai.py

@@ -41,14 +41,13 @@ def completion(
     logger_fn=None,
 ):
     headers = validate_environment(api_key)
-    model = model
     if model in custom_prompt_dict:
         # check if the model has a registered custom prompt
         model_prompt_details = custom_prompt_dict[model]
         prompt = custom_prompt(
             role_dict=model_prompt_details["roles"], 
-            initial_prompt_value=model_prompt_details["pre_message_sep"],  
+            initial_prompt_value=model_prompt_details["initial_prompt_value"],  
-            final_prompt_value=model_prompt_details["post_message_sep"], 
+            final_prompt_value=model_prompt_details["final_prompt_value"], 
             messages=messages
         )
     else:

litellm/llms/vllm.py

@@ -0,0 +1,97 @@
+import os
+import json
+from enum import Enum
+import requests
+import time
+from typing import Callable
+from litellm.utils import ModelResponse
+from .prompt_templates.factory import prompt_factory, custom_prompt
+
+class VLLMError(Exception):
+    def __init__(self, status_code, message):
+        self.status_code = status_code
+        self.message = message
+        super().__init__(
+            self.message
+        )  # Call the base class constructor with the parameters it needs
+
+# check if vllm is installed
+def validate_environment():
+    try: 
+        from vllm import LLM, SamplingParams
+        return LLM, SamplingParams
+    except:
+        raise VLLMError(status_code=0, message="The vllm package is not installed in your environment. Run - `pip install vllm` before proceeding.")
+
+def completion(
+    model: str,
+    messages: list,
+    model_response: ModelResponse,
+    print_verbose: Callable,
+    encoding,
+    logging_obj,
+    custom_prompt_dict={},
+    optional_params=None,
+    litellm_params=None,
+    logger_fn=None,
+):
+    LLM, SamplingParams = validate_environment()
+    try:
+        llm = LLM(model=model)
+    except Exception as e:
+        raise VLLMError(status_code=0, message=str(e))
+    sampling_params = SamplingParams(**optional_params)
+    if model in custom_prompt_dict:
+        # check if the model has a registered custom prompt
+        model_prompt_details = custom_prompt_dict[model]
+        prompt = custom_prompt(
+            role_dict=model_prompt_details["roles"], 
+            initial_prompt_value=model_prompt_details["initial_prompt_value"],  
+            final_prompt_value=model_prompt_details["final_prompt_value"], 
+            messages=messages
+        )
+    else:
+        prompt = prompt_factory(model=model, messages=messages)
+
+
+    ## LOGGING
+    logging_obj.pre_call(
+        input=prompt,
+        api_key="",
+        additional_args={"complete_input_dict": sampling_params},
+    )
+
+    outputs = llm.generate(prompt, sampling_params)
+
+    
+    ## COMPLETION CALL
+    if "stream" in optional_params and optional_params["stream"] == True:
+        return iter(outputs)
+    else:
+        ## LOGGING
+        logging_obj.post_call(
+            input=prompt,
+            api_key="",
+            original_response=outputs,
+            additional_args={"complete_input_dict": sampling_params},
+        )
+        print_verbose(f"raw model_response: {outputs}")
+        ## RESPONSE OBJECT
+        model_response["choices"][0]["message"]["content"] = outputs[0].outputs[0].text
+
+        ## CALCULATING USAGE
+        prompt_tokens = len(outputs[0].prompt_token_ids)  
+        completion_tokens = len(outputs[0].outputs[0].token_ids)  
+
+        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
+    pass

litellm/main.py

@@ -27,6 +27,7 @@ from .llms import huggingface_restapi
 from .llms import replicate
 from .llms import aleph_alpha
 from .llms import baseten
+from .llms import vllm
 import tiktoken
 from concurrent.futures import ThreadPoolExecutor
 from typing import Callable, List, Optional, Dict
@@ -670,20 +671,18 @@ def completion(
                 encoding=encoding,
                 logging_obj=logging
             )
-            
+
-            # TODO: Add streaming for sagemaker
+            if "stream" in optional_params and optional_params["stream"] == True: ## [BETA]
-            # if "stream" in optional_params and optional_params["stream"] == True:
+                # don't try to access stream object,
-            #     # don't try to access stream object,
+                response = CustomStreamWrapper(
-            #     response = CustomStreamWrapper(
+                    iter(model_response), model, custom_llm_provider="sagemaker", logging_obj=logging
-            #         model_response, model, custom_llm_provider="ai21", logging_obj=logging
+                )
-            #     )
+                return response
-            #     return response
+
-            
             ## RESPONSE OBJECT
             response = model_response
-        elif custom_llm_provider == "bedrock":
+        elif custom_llm_provider == "vllm":
-            # boto3 reads keys from .env
+            model_response = vllm.completion(
-            model_response = bedrock.completion(
                 model=model,
                 messages=messages,
                 model_response=model_response,
@@ -695,17 +694,15 @@ def completion(
                 logging_obj=logging
             )
             
-            # TODO: Add streaming for bedrock
+            if "stream" in optional_params and optional_params["stream"] == True: ## [BETA]
-            # if "stream" in optional_params and optional_params["stream"] == True:
+                # don't try to access stream object,
-            #     # don't try to access stream object,
+                response = CustomStreamWrapper(
-            #     response = CustomStreamWrapper(
+                    model_response, model, custom_llm_provider="vllm", logging_obj=logging
-            #         model_response, model, custom_llm_provider="ai21", logging_obj=logging
+                )
-            #     )
+                return response
-            #     return response
+
-            
             ## RESPONSE OBJECT
             response = model_response
-
         elif custom_llm_provider == "ollama":
             endpoint = (
                 litellm.api_base if litellm.api_base is not None else api_base

litellm/tests/test_completion.py

@@ -435,7 +435,6 @@ def test_completion_together_ai():
     except Exception as e:
         pytest.fail(f"Error occurred: {e}")
 
-test_completion_together_ai()
 # def test_customprompt_together_ai():
 #     try:
 #         litellm.register_prompt_template(
@@ -462,6 +461,20 @@ def test_completion_sagemaker():
     except Exception as e:
         pytest.fail(f"Error occurred: {e}")
 
+######## Test VLLM ########
+# def test_completion_vllm():
+#     try:
+#         response = completion(
+#             model="vllm/facebook/opt-125m", 
+#             messages=messages,
+#             temperature=0.2,
+#             max_tokens=80,
+#         )
+#         print(response)
+#     except Exception as e:
+#         pytest.fail(f"Error occurred: {e}")
+
+# test_completion_vllm()
 
 # def test_completion_custom_api_base():
 #     try:

litellm/utils.py

@@ -1385,23 +1385,33 @@ def modify_integration(integration_name, integration_params):
 
 
 # custom prompt helper function
-def register_prompt_template(model: str, roles: dict, pre_message_sep: str, post_message_sep: str):
+def register_prompt_template(model: str, roles: dict, initial_prompt_value: str = "", final_prompt_value: str = ""):
     """
     Example usage:
     ```
     import litellm 
     litellm.register_prompt_template(
-	    model="bloomz",
+	    model="llama-2",
-	    roles={"system":"<|im_start|>system", "assistant":"<|im_start|>assistant", "user":"<|im_start|>user"}
+	    roles={
-	    pre_message_sep: "\n",
+            "system": {
-	    post_message_sep: "<|im_end|>\n"
+                "pre_message": "[INST] <<SYS>>\n",
+                "post_message": "\n<</SYS>>\n [/INST]\n"
+            },
+            "user": { # follow this format https://github.com/facebookresearch/llama/blob/77062717054710e352a99add63d160274ce670c6/llama/generation.py#L348
+                "pre_message": "[INST] ",
+                "post_message": " [/INST]\n"
+            }, 
+            "assistant": {
+                "post_message": "\n" # follows this - https://replicate.com/blog/how-to-prompt-llama
+            }
+        },
     )
     ```
     """
     litellm.custom_prompt_dict[model] = {
         "roles": roles,
-        "pre_message_sep": pre_message_sep,
+        "initial_prompt_value": initial_prompt_value,
-        "post_message_sep": post_message_sep
+        "final_prompt_value": final_prompt_value
     }
     return litellm.custom_prompt_dict
 
@@ -1844,6 +1854,14 @@ def exception_type(model, original_exception, custom_llm_provider):
                         llm_provider="together_ai",
                         model=model
                     )
+            elif custom_llm_provider == "vllm":
+                if hasattr(original_exception, "status_code"):
+                    if original_exception.status_code == 0:
+                        raise APIConnectionError(
+                            message=f"VLLMException - {original_exception.message}",
+                            llm_provider="vllm",
+                            model=model
+                        )
         else:
             raise original_exception
     except Exception as e:
@@ -2080,6 +2098,9 @@ class CustomStreamWrapper:
             elif self.custom_llm_provider and self.custom_llm_provider == "ai21": #ai21 doesn't provide streaming
                 chunk = next(self.completion_stream)
                 completion_obj["content"] = self.handle_ai21_chunk(chunk)
+            elif self.custom_llm_provider and self.custom_llm_provider == "vllm":
+                chunk = next(self.completion_stream)
+                completion_obj["content"] = chunk[0].outputs[0].text
             elif self.model in litellm.aleph_alpha_models: #ai21 doesn't provide streaming
                 chunk = next(self.completion_stream)
                 completion_obj["content"] = self.handle_aleph_alpha_chunk(chunk)

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "litellm"
-version = "0.1.545"
+version = "0.1.548"
 description = "Library to easily interface with LLM API providers"
 authors = ["BerriAI"]
 license = "MIT License"