templated system messages

8 files changed, 62 insertions, 16 deletions

diff --git a/continuedev/src/continuedev/core/config.py b/continuedev/src/continuedev/core/config.py
index 957609c5..91a47c8e 100644
--- a/continuedev/src/continuedev/core/config.py
+++ b/continuedev/src/continuedev/core/config.py
@@ -85,6 +85,7 @@ class ContinueConfig(BaseModel):
     slash_commands: Optional[List[SlashCommand]] = DEFAULT_SLASH_COMMANDS
     on_traceback: Optional[List[OnTracebackSteps]] = [
         OnTracebackSteps(step_name="DefaultOnTracebackStep")]
+    system_message: Optional[str] = None
 
     # Want to force these to be the slash commands for now
     @validator('slash_commands', pre=True)
diff --git a/continuedev/src/continuedev/core/sdk.py b/continuedev/src/continuedev/core/sdk.py
index eb60109c..ac57c122 100644
--- a/continuedev/src/continuedev/core/sdk.py
+++ b/continuedev/src/continuedev/core/sdk.py
@@ -34,10 +34,12 @@ MODEL_PROVIDER_TO_ENV_VAR = {
 class Models:
     provider_keys: Dict[ModelProvider, str] = {}
     model_providers: List[ModelProvider]
+    system_message: str
 
     def __init__(self, sdk: "ContinueSDK", model_providers: List[ModelProvider]):
         self.sdk = sdk
         self.model_providers = model_providers
+        self.system_message = sdk.config.system_message
 
     @classmethod
     async def create(cls, sdk: "ContinueSDK", with_providers: List[ModelProvider] = ["openai"]) -> "Models":
@@ -53,12 +55,12 @@ class Models:
     def __load_openai_model(self, model: str) -> OpenAI:
         api_key = self.provider_keys["openai"]
         if api_key == "":
-            return ProxyServer(self.sdk.ide.unique_id, model)
-        return OpenAI(api_key=api_key, default_model=model)
+            return ProxyServer(self.sdk.ide.unique_id, model, system_message=self.system_message)
+        return OpenAI(api_key=api_key, default_model=model, system_message=self.system_message)
 
     def __load_hf_inference_api_model(self, model: str) -> HuggingFaceInferenceAPI:
         api_key = self.provider_keys["hf_inference_api"]
-        return HuggingFaceInferenceAPI(api_key=api_key, model=model)
+        return HuggingFaceInferenceAPI(api_key=api_key, model=model, system_message=self.system_message)
 
     @cached_property
     def starcoder(self):
@@ -82,7 +84,7 @@ class Models:
 
     @cached_property
     def ggml(self):
-        return GGML()
+        return GGML(system_message=self.system_message)
 
     def __model_from_name(self, model_name: str):
         if model_name == "starcoder":
diff --git a/continuedev/src/continuedev/libs/llm/ggml.py b/continuedev/src/continuedev/libs/llm/ggml.py
index d3589b70..6007fdb4 100644
--- a/continuedev/src/continuedev/libs/llm/ggml.py
+++ b/continuedev/src/continuedev/libs/llm/ggml.py
@@ -33,7 +33,7 @@ class GGML(LLM):
 
         args = {**self.default_args, **kwargs}
         messages = compile_chat_messages(
-            self.name, with_history, args["max_tokens"], prompt, functions=args.get("functions", None))
+            self.name, with_history, args["max_tokens"], prompt, functions=args.get("functions", None), system_message=self.system_message)
 
         async with aiohttp.ClientSession() as session:
             async with session.post(f"{SERVER_URL}/v1/completions", json={
@@ -50,7 +50,7 @@ class GGML(LLM):
     async def stream_chat(self, messages: List[ChatMessage] = [], **kwargs) -> Generator[Union[Any, List, Dict], None, None]:
         args = {**self.default_args, **kwargs}
         messages = compile_chat_messages(
-            self.name, messages, args["max_tokens"], None, functions=args.get("functions", None))
+            self.name, messages, args["max_tokens"], None, functions=args.get("functions", None), system_message=self.system_message)
         args["stream"] = True
 
         async with aiohttp.ClientSession() as session:
@@ -77,7 +77,7 @@ class GGML(LLM):
 
         async with aiohttp.ClientSession() as session:
             async with session.post(f"{SERVER_URL}/v1/completions", json={
-                "messages": compile_chat_messages(args["model"], with_history, args["max_tokens"], prompt, functions=None),
+                "messages": compile_chat_messages(args["model"], with_history, args["max_tokens"], prompt, functions=None, system_message=self.system_message),
                 **args
             }) as resp:
                 try:
diff --git a/continuedev/src/continuedev/libs/llm/hf_inference_api.py b/continuedev/src/continuedev/libs/llm/hf_inference_api.py
index 803ba122..7e11fbbe 100644
--- a/continuedev/src/continuedev/libs/llm/hf_inference_api.py
+++ b/continuedev/src/continuedev/libs/llm/hf_inference_api.py
@@ -11,9 +11,10 @@ class HuggingFaceInferenceAPI(LLM):
     api_key: str
     model: str
 
-    def __init__(self, api_key: str, model: str):
+    def __init__(self, api_key: str, model: str, system_message: str = None):
         self.api_key = api_key
         self.model = model
+        self.system_message = system_message  # TODO: Nothing being done with this
 
     def complete(self, prompt: str, with_history: List[ChatMessage] = [], **kwargs):
         """Return the completion of the text with the given temperature."""
diff --git a/continuedev/src/continuedev/libs/llm/openai.py b/continuedev/src/continuedev/libs/llm/openai.py
index f0877d90..d973f19e 100644
--- a/continuedev/src/continuedev/libs/llm/openai.py
+++ b/continuedev/src/continuedev/libs/llm/openai.py
@@ -37,7 +37,7 @@ class OpenAI(LLM):
         if args["model"] in CHAT_MODELS:
             async for chunk in await openai.ChatCompletion.acreate(
                 messages=compile_chat_messages(
-                    args["model"], with_history, args["max_tokens"], prompt, functions=None),
+                    args["model"], with_history, args["max_tokens"], prompt, functions=None, system_message=self.system_message),
                 **args,
             ):
                 if "content" in chunk.choices[0].delta:
@@ -58,7 +58,7 @@ class OpenAI(LLM):
 
         async for chunk in await openai.ChatCompletion.acreate(
             messages=compile_chat_messages(
-                args["model"], messages, args["max_tokens"], functions=args.get("functions", None)),
+                args["model"], messages, args["max_tokens"], functions=args.get("functions", None), system_message=self.system_message),
             **args,
         ):
             yield chunk.choices[0].delta
@@ -69,7 +69,7 @@ class OpenAI(LLM):
         if args["model"] in CHAT_MODELS:
             resp = (await openai.ChatCompletion.acreate(
                 messages=compile_chat_messages(
-                    args["model"], with_history, args["max_tokens"], prompt, functions=None),
+                    args["model"], with_history, args["max_tokens"], prompt, functions=None, system_message=self.system_message),
                 **args,
             )).choices[0].message.content
         else:
diff --git a/continuedev/src/continuedev/libs/llm/proxy_server.py b/continuedev/src/continuedev/libs/llm/proxy_server.py
index eab6e441..3ec492f3 100644
--- a/continuedev/src/continuedev/libs/llm/proxy_server.py
+++ b/continuedev/src/continuedev/libs/llm/proxy_server.py
@@ -38,7 +38,7 @@ class ProxyServer(LLM):
 
         async with aiohttp.ClientSession(connector=aiohttp.TCPConnector(ssl_context=ssl_context)) as session:
             async with session.post(f"{SERVER_URL}/complete", json={
-                "messages": compile_chat_messages(args["model"], with_history, args["max_tokens"], prompt, functions=None),
+                "messages": compile_chat_messages(args["model"], with_history, args["max_tokens"], prompt, functions=None, system_message=self.system_message),
                 "unique_id": self.unique_id,
                 **args
             }) as resp:
@@ -50,7 +50,7 @@ class ProxyServer(LLM):
     async def stream_chat(self, messages: List[ChatMessage] = [], **kwargs) -> Coroutine[Any, Any, Generator[Union[Any, List, Dict], None, None]]:
         args = {**self.default_args, **kwargs}
         messages = compile_chat_messages(
-            self.default_model, messages, args["max_tokens"], None, functions=args.get("functions", None))
+            self.default_model, messages, args["max_tokens"], None, functions=args.get("functions", None), system_message=self.system_message)
 
         async with aiohttp.ClientSession(connector=aiohttp.TCPConnector(ssl_context=ssl_context)) as session:
             async with session.post(f"{SERVER_URL}/stream_chat", json={
@@ -74,7 +74,7 @@ class ProxyServer(LLM):
     async def stream_complete(self, prompt, with_history: List[ChatMessage] = [], **kwargs) -> Generator[Union[Any, List, Dict], None, None]:
         args = {**self.default_args, **kwargs}
         messages = compile_chat_messages(
-            self.default_model, with_history, args["max_tokens"], prompt, functions=args.get("functions", None))
+            self.default_model, with_history, args["max_tokens"], prompt, functions=args.get("functions", None), system_message=self.system_message)
 
         async with aiohttp.ClientSession(connector=aiohttp.TCPConnector(ssl_context=ssl_context)) as session:
             async with session.post(f"{SERVER_URL}/stream_complete", json={
diff --git a/continuedev/src/continuedev/libs/util/count_tokens.py b/continuedev/src/continuedev/libs/util/count_tokens.py
index e1baeca1..1ca98fe6 100644
--- a/continuedev/src/continuedev/libs/util/count_tokens.py
+++ b/continuedev/src/continuedev/libs/util/count_tokens.py
@@ -1,6 +1,7 @@
 import json
 from typing import Dict, List, Union
 from ...core.main import ChatMessage
+from .templating import render_system_message
 import tiktoken
 
 aliases = {
@@ -85,13 +86,15 @@ def compile_chat_messages(model: str, msgs: List[ChatMessage], max_tokens: int,
         for function in functions:
             prompt_tokens += count_tokens(model, json.dumps(function))
 
+    rendered_system_message = render_system_message(system_message)
+
     msgs = prune_chat_history(model,
-                              msgs, MAX_TOKENS_FOR_MODEL[model], prompt_tokens + max_tokens + count_tokens(model, system_message))
+                              msgs, MAX_TOKENS_FOR_MODEL[model], prompt_tokens + max_tokens + count_tokens(model, rendered_system_message))
     history = []
     if system_message:
         history.append({
             "role": "system",
-            "content": system_message
+            "content": rendered_system_message
         })
     history += [msg.to_dict(with_functions=functions is not None)
                 for msg in msgs]
diff --git a/continuedev/src/continuedev/libs/util/templating.py b/continuedev/src/continuedev/libs/util/templating.py
new file mode 100644
index 00000000..ebfc2e31
--- /dev/null
+++ b/continuedev/src/continuedev/libs/util/templating.py
@@ -0,0 +1,39 @@
+import os
+import chevron
+
+
+def get_vars_in_template(template):
+    """
+    Get the variables in a template
+    """
+    return [token[1] for token in chevron.tokenizer.tokenize(template) if token[0] == 'variable']
+
+
+def escape_var(var: str) -> str:
+    """
+    Escape a variable so it can be used in a template
+    """
+    return var.replace(os.path.sep, '').replace('.', '')
+
+
+def render_system_message(system_message: str) -> str:
+    """
+    Render system message with mustache syntax.
+    Right now it only supports rendering absolute file paths as their contents.
+    """
+    vars = get_vars_in_template(system_message)
+
+    args = {}
+    for var in vars:
+        if var.startswith(os.path.sep):
+            # Escape vars which are filenames, because mustache doesn't allow / in variable names
+            escaped_var = escape_var(var)
+            system_message = system_message.replace(
+                var, escaped_var)
+
+            if os.path.exists(var):
+                args[escaped_var] = open(var, 'r').read()
+            else:
+                args[escaped_var] = ''
+
+    return chevron.render(system_message, args)