Merge pull request #1416 from OpenInterpreter/development

Merge Development Branch

Author: KillianLucas

docs/getting-started/setup.mdx

@@ -5,7 +5,7 @@ title: Setup
 <iframe
   width="560"
   height="315"
-  src="https://www.youtube.com/watch?v=5sk3t8ilDR8"
+  src="https://www.youtube.com/embed/5sk3t8ilDR8"
   frameBorder="0"
   allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
   allowFullScreen

docs/guides/profiles.mdx

@@ -2,6 +2,15 @@
 title: Profiles
 ---
 
+<iframe
+  width="560"
+  height="315"
+  src="https://www.youtube.com/embed/NxfdrGQrkHQ"
+  frameBorder="0"
+  allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
+  allowFullScreen
+></iframe>
+
 Profiles are a powerful way to customize your instance of Open Interpreter.
 
 Profiles are Python files that configure Open Interpreter. A wide range of fields from the [model](/settings/all-settings#model-selection) to the [context window](/settings/all-settings#context-window) to the [message templates](/settings/all-settings#user-message-template) can be configured in a Profile. This allows you to save multiple variations of Open Interpreter to optimize for your specific use-cases.

docs/language-models/custom-models.mdx

@@ -9,21 +9,20 @@ As long as your system can accept an input and stream an output (and can be inte
 Simply replace the OpenAI-compatible `completions` function in your language model with one of your own:
 
 ```python
-def custom_language_model(openai_message):
+def custom_language_model(messages, model, stream, max_tokens):
     """
     OpenAI-compatible completions function (this one just echoes what the user said back).
+    To make it OpenAI-compatible and parsable, `choices` has to be the root property.
+    The property `delta` is used to signify streaming.
     """
-    users_content = openai_message[-1].get("content") # Get last message's content
-
-    # To make it OpenAI-compatible, we yield this first:
-    yield {"delta": {"role": "assistant"}}
+    users_content = messages[-1].get("content") # Get last message's content
 
     for character in users_content:
-        yield {"delta": {"content": character}}
+        yield {"choices": [{"delta": {"content": character}}]}
 
 # Tell Open Interpreter to power the language model with this function
 
-interpreter.llm.completion = custom_language_model
+interpreter.llm.completions = custom_language_model
 ```
 
 Then, set the following settings:
@@ -39,4 +38,4 @@ And start using it:
 
 ```
 interpreter.chat("Hi!") # Returns/displays "Hi!" character by character
-```
+```

interpreter/core/async_core.py

@@ -701,47 +701,45 @@ async def openai_compatible_generator():
     async def chat_completion(request: ChatCompletionRequest):
         # Convert to LMC
 
-        user_messages = []
-        for message in reversed(request.messages):
-            if message.role == "user":
-                user_messages.append(message)
-            else:
-                break
-        user_messages.reverse()
-
-        for message in user_messages:
-            if type(message.content) == str:
-                async_interpreter.messages.append(
-                    {"role": "user", "type": "message", "content": message.content}
-                )
-            if type(message.content) == list:
-                for content in message.content:
-                    if content["type"] == "text":
-                        async_interpreter.messages.append(
-                            {"role": "user", "type": "message", "content": content}
+        last_message = request.messages[-1]
+
+        if last_message.role != "user":
+            raise ValueError("Last message must be from the user.")
+
+        if last_message.content == "{STOP}":
+            # Handle special STOP token
+            return
+
+        if type(last_message.content) == str:
+            async_interpreter.messages.append(last_message)
+        if type(last_message.content) == list:
+            for content in last_message.content:
+                if content["type"] == "text":
+                    async_interpreter.messages.append(
+                        {"role": "user", "type": "message", "content": content}
+                    )
+                elif content["type"] == "image_url":
+                    if "url" not in content["image_url"]:
+                        raise Exception("`url` must be in `image_url`.")
+                    url = content["image_url"]["url"]
+                    print(url[:100])
+                    if "base64," not in url:
+                        raise Exception(
+                            '''Image must be in the format: "data:image/jpeg;base64,{base64_image}"'''
                         )
-                    elif content["type"] == "image_url":
-                        if "url" not in content["image_url"]:
-                            raise Exception("`url` must be in `image_url`.")
-                        url = content["image_url"]["url"]
-                        print(url[:100])
-                        if "base64," not in url:
-                            raise Exception(
-                                '''Image must be in the format: "data:image/jpeg;base64,{base64_image}"'''
-                            )
 
-                        # data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAA6oA...
+                    # data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAA6oA...
 
-                        data = url.split("base64,")[1]
-                        format = "base64." + url.split(";")[0].split("/")[1]
-                        async_interpreter.messages.append(
-                            {
-                                "role": "user",
-                                "type": "image",
-                                "format": format,
-                                "content": data,
-                            }
-                        )
+                    data = url.split("base64,")[1]
+                    format = "base64." + url.split(";")[0].split("/")[1]
+                    async_interpreter.messages.append(
+                        {
+                            "role": "user",
+                            "type": "image",
+                            "format": format,
+                            "content": data,
+                        }
+                    )
 
         if request.stream:
             return StreamingResponse(

interpreter/core/llm/llm.py

@@ -25,13 +25,27 @@
 from .run_tool_calling_llm import run_tool_calling_llm
 from .utils.convert_to_openai_messages import convert_to_openai_messages
 
+import logging
+
+# Create or get the logger
+logger = logging.getLogger('LiteLLM')
+
+class SuppressDebugFilter(logging.Filter):
+    def filter(self, record):
+        # Suppress only the specific message containing the keywords
+        if "cost map" in record.getMessage():
+            return False  # Suppress this log message
+        return True  # Allow all other messages
 
 class Llm:
     """
     A stateless LMC-style LLM with some helpful properties.
     """
 
     def __init__(self, interpreter):
+        # Add the filter to the logger
+        logger.addFilter(SuppressDebugFilter())
+
         # Store a reference to parent interpreter
         self.interpreter = interpreter
 
@@ -311,6 +325,9 @@ def load(self):
         if self._is_loaded:
             return
 
+        if self.model.startswith("ollama/") and not ":" in self.model:
+            self.model = self.model + ":latest"
+
         self._is_loaded = True
 
         if self.model.startswith("ollama/"):
@@ -323,7 +340,7 @@ def load(self):
                 if response.ok:
                     data = response.json()
                     names = [
-                        model["name"].replace(":latest", "")
+                        model["name"]
                         for model in data["models"]
                         if "name" in model and model["name"]
                     ]
@@ -358,6 +375,7 @@ def load(self):
                     self.max_tokens = int(self.context_window * 0.2)
 
             # Send a ping, which will actually load the model
+            model_name = model_name.replace(":latest", "")
             print(f"Loading {model_name}...\n")
 
             old_max_tokens = self.max_tokens
@@ -398,6 +416,8 @@ def fixed_litellm_completions(**params):
     else:
         litellm.drop_params = True
 
+    params["model"] = params["model"].replace(":latest", "")
+
     # Run completion
     attempts = 4
     first_error = None

interpreter/terminal_interface/local_setup.py

@@ -257,7 +257,7 @@ def download_model(models_dir, models, interpreter):
             names=[name for name in names if not any(word.lower() in name.lower() for word in priority_models)]
             names=priority_models_found+names
 
-            for model in ["llama3", "phi3", "wizardlm2", "codestral"]:
+            for model in ["llama3.1", "phi3", "mistral-nemo", "gemma2", "codestral"]:
                 if model not in names:
                     names.append("↓ Download " + model)