Feature/update sciphi local logic rebased (#120)

* tweak local logic for sciphi

* fix

sciphi/interface/llm/sciphi_interface.py

@@ -71,8 +71,8 @@ class SciPhiLLMInterface(LLMInterface):
 
     def __init__(
         self,
-        rag_interface: RAGInterface,
         config: SciPhiConfig = SciPhiConfig(),
+        rag_interface: Optional[RAGInterface] = None,
         *args,
         **kwargs,
     ) -> None:
@@ -98,11 +98,22 @@ def get_chat_completion(
         if not added_system_prompt:
             prompt = f"### System:\n{SciPhiLLMInterface.ALPACA_CHAT_SYSTEM_PROMPT}.\n\n{prompt}"
 
-        context = self.rag_interface.get_contexts([last_user_message])[0]
-        prompt += f"### Response:\n{SciPhiFormatter.RETRIEVAL_TOKEN} {SciPhiFormatter.INIT_PARAGRAPH_TOKEN}{context}{SciPhiFormatter.END_PARAGRAPH_TOKEN}"
+        # TODO - Cleanup RAG logic checks across this script.
+
+        if not generation_config.model_name:
+            raise ValueError("No model name provided")
+        if "RAG" in generation_config.model_name:
+            if not self.rag_interface:
+                raise ValueError(
+                    "RAG generation requested but no RAG interface provided"
+                )
+            context = self.rag_interface.get_contexts([last_user_message])[0]
+            prompt += f"### Response:\n{SciPhiFormatter.RETRIEVAL_TOKEN} {SciPhiFormatter.INIT_PARAGRAPH_TOKEN}{context}{SciPhiFormatter.END_PARAGRAPH_TOKEN}"
+        else:
+            prompt += f"### Response:\n"
         latest_completion = self.model.get_instruct_completion(
             prompt, generation_config
-        ).strip()
+        )
 
         return SciPhiFormatter.remove_cruft(latest_completion)
 
@@ -114,6 +125,20 @@ def get_completion(
         logger.debug(
             f"Requesting completion from local vLLM with model={generation_config.model_name} and prompt={prompt}"
         )
+
+        # TODO - Cleanup and consolidate RAG logic checks across this script.
+
+        if not generation_config.model_name:
+            raise ValueError("No model name provided")
+
+        if "RAG" not in generation_config.model_name:
+            return self.model.get_instruct_completion(
+                prompt, generation_config
+            ).strip()
+        if not self.rag_interface:
+            raise ValueError(
+                "RAG model requested, but no RAG interface provided"
+            )
         completion = ""
         while True:
             prompt_with_context = (
@@ -133,7 +158,7 @@ def get_completion(
             )
             context = self.rag_interface.get_contexts([context_query])[0]
             completion += f"{SciPhiFormatter.INIT_PARAGRAPH_TOKEN}{context}{SciPhiFormatter.END_PARAGRAPH_TOKEN}"
-        return SciPhiFormatter.remove_cruft(completion)
+        return SciPhiFormatter.remove_cruft(completion).strip()
 
     def get_batch_completion(
         self, prompts: List[str], generation_config: GenerationConfig

sciphi/llm/models/sciphi_llm.py

@@ -28,7 +28,10 @@ class SciPhiConfig(LLMConfig):
     sub_provider_name: LLMProviderName = LLMProviderName.VLLM
 
     # SciPhi Extras...
-    mode = SciPhiProviderMode.REMOTE
+    mode: SciPhiProviderMode = SciPhiProviderMode.REMOTE
+    ## Local
+    model_name: Optional[str] = None
+    ## Remote
     api_base: Optional[str] = "https://api.sciphi.ai/v1"
     api_key: Optional[str] = None
 
@@ -47,14 +50,15 @@ def __init__(
             SciPhiProviderMode.REMOTE,
             SciPhiProviderMode.LOCAL_VLLM,
         ]:
-            # Remote and local vLLM are both powered by vLLM
-            assert self.config.sub_provider_name == LLMProviderName.VLLM
             from sciphi.llm.models.vllm_llm import (
                 vLLM,
                 vLLMConfig,
                 vLLMProviderMode,
             )
 
+            # Remote and local vLLM are both powered by vLLM
+            assert self.config.sub_provider_name == LLMProviderName.VLLM
+
             if self.config.mode == SciPhiProviderMode.REMOTE:
                 api_key = config.api_key or os.getenv("SCIPHI_API_KEY")
                 if not api_key:
@@ -64,11 +68,21 @@ def __init__(
                 self.model = vLLM(
                     vLLMConfig(
                         provider_name=config.provider_name,
+                        model_name=config.model_name,
                         api_base=config.api_base,
                         api_key=api_key,
                         mode=vLLMProviderMode.REMOTE,
                     ),
                 )
+            elif self.config.mode == SciPhiProviderMode.LOCAL_VLLM:
+                self.model = vLLM(
+                    vLLMConfig(
+                        provider_name=config.provider_name,
+                        model_name=config.model_name,
+                        mode=vLLMProviderMode.LOCAL,
+                    ),
+                )
+
         elif self.config.mode == SciPhiProviderMode.LOCAL_HF:
             from sciphi.llm.models import hugging_face_llm  # noqa F401
 

sciphi/scripts/sciphi_chat.py

@@ -7,7 +7,7 @@
     SciPhiLLMInterface,
     SciPhiWikiRAGInterface,
 )
-from sciphi.llm import GenerationConfig
+from sciphi.llm import GenerationConfig, SciPhiConfig
 
 
 def filter_relevant_args(dataclass_type, args_dict):
@@ -20,10 +20,22 @@ def filter_relevant_args(dataclass_type, args_dict):
 
 def main(
     query: str = "Who is the president of the United States?",
+    mode="remote",
     llm_model_name="SciPhi/SciPhi-Self-RAG-Mistral-7B-32k",
 ):
     rag_interface = SciPhiWikiRAGInterface()
-    llm_interface = SciPhiLLMInterface(rag_interface)
+
+    config = None
+    if mode == "local":
+        from sciphi.llm.models.sciphi_llm import SciPhiProviderMode
+
+        config = SciPhiConfig(
+            mode=SciPhiProviderMode.LOCAL_VLLM, model_name=llm_model_name
+        )
+    else:
+        config = SciPhiConfig()
+
+    llm_interface = SciPhiLLMInterface(config, rag_interface)
 
     generation_config = GenerationConfig(
         model_name=llm_model_name,
@@ -36,14 +48,14 @@ def main(
             "role": "system",
             "content": "You are a helpful and informative professor. You give long, accurate, and detailed explanations to student questions. You answer EVERY question that is given to you. You retrieve data multiple times if necessary.",
         },
-        {
-            "role": "user",
-            "content": "Who is the president of the United States?",
-        },
-        {
-            "role": "assistant",
-            "content": "Joe Biden is the current president of the United States.",
-        },
+        # {
+        #     "role": "user",
+        #     "content": "Who is the president of the United States?",
+        # },
+        # {
+        #     "role": "assistant",
+        #     "content": "Joe Biden is the current president of the United States.",
+        # },
         {
             "role": "user",
             "content": query,