Better messaging

Author: KillianLucas

interpreter/core/llm/llm.py

@@ -234,7 +234,7 @@ def fixed_litellm_completions(**params):
 
     if "local" in params.get("model"):
         # Kinda hacky, but this helps
-        params["stop"] = ["<|assistant|>", "<|end|>"]
+        params["stop"] = ["<|assistant|>", "<|end|>", "<|eot_id|>"]
 
     # Run completion
     first_error = None

interpreter/terminal_interface/profiles/defaults/local.py

@@ -10,6 +10,8 @@
 
 from interpreter import interpreter
 
+model = None
+
 
 def download_model(models_dir, models, interpreter):
     # Get RAM and disk information
@@ -20,15 +22,11 @@ def download_model(models_dir, models, interpreter):
         1024 * 1024 * 1024
     )  # Convert bytes to GB
 
-    time.sleep(1)
-
     # Display the users hardware specs
     interpreter.display_message(
         f"Your machine has `{total_ram:.2f}GB` of RAM, and `{free_disk_space:.2f}GB` of free storage space."
     )
 
-    time.sleep(2)
-
     if total_ram < 10:
         interpreter.display_message(
             f"\nYour computer realistically can only run smaller models less than 4GB, Phi-2 might be the best model for your computer.\n"
@@ -42,14 +40,10 @@ def download_model(models_dir, models, interpreter):
             f"\nYour computer should have enough RAM to run any model below.\n"
         )
 
-    time.sleep(1)
-
     interpreter.display_message(
         f"In general, the larger the model, the better the performance, but choose a model that best fits your computer's hardware. \nOnly models you have the storage space to download are shown:\n"
     )
 
-    time.sleep(1)
-
     try:
         model_list = [
             {
@@ -182,7 +176,7 @@ def download_model(models_dir, models, interpreter):
 
 # START OF LOCAL MODEL PROVIDER LOGIC
 interpreter.display_message(
-    "> Open Interpreter is compatible with several local model providers.\n"
+    "\n**Open Interpreter** supports multiple local model providers.\n"
 )
 
 # Define the choices for local models
@@ -197,7 +191,7 @@ def download_model(models_dir, models, interpreter):
 questions = [
     inquirer.List(
         "model",
-        message="What one would you like to use?",
+        message="Select a provider",
         choices=choices,
     ),
 ]
@@ -231,41 +225,23 @@ def download_model(models_dir, models, interpreter):
 elif selected_model == "Ollama":
     try:
         # List out all downloaded ollama models. Will fail if ollama isn't installed
-        def list_ollama_models():
-            result = subprocess.run(
-                ["ollama", "list"], capture_output=True, text=True, check=True
-            )
-            lines = result.stdout.split("\n")
-            names = [
-                line.split()[0].replace(":latest", "")
-                for line in lines[1:]
-                if line.strip()
-            ]  # Extract names, trim out ":latest", skip header
-            return names
-
-        llama3_installed = True
-        names = list_ollama_models()
-        if "llama3" not in names:
-            # If a user has other models installed but not llama3, let's display the correct message
-            if not names:
-                llama3_installed = False
-            names.insert(0, "llama3")
-
-        # If there are models, prompt them to select one
-        time.sleep(1)
+        result = subprocess.run(
+            ["ollama", "list"], capture_output=True, text=True, check=True
+        )
+        lines = result.stdout.split("\n")
+        names = [
+            line.split()[0].replace(":latest", "") for line in lines[1:] if line.strip()
+        ]  # Extract names, trim out ":latest", skip header
 
-        if llama3_installed:
-            interpreter.display_message(
-                f"**{len(names)} Ollama model{'s' if len(names) != 1 else ''} found.** To download a new model, run `ollama run <model-name>`, then start a new interpreter session. \n\n For a full list of downloadable models, check out [https://ollama.com/library](https://ollama.com/library) \n"
-            )
+        for model in ["llama3", "phi3", "wizardlm2"]:
+            if model not in names:
+                names.append("→ Download " + model)
 
         # Create a new inquirer selection from the names
         name_question = [
             inquirer.List(
                 "name",
-                message="Select a downloaded Ollama model:"
-                if llama3_installed
-                else "No models found. Select a model to install:",
+                message="Select a model",
                 choices=names,
             ),
         ]
@@ -276,17 +252,16 @@ def list_ollama_models():
 
         selected_name = name_answer["name"]
 
-        if selected_name == "llama3":
-            # If the user selects llama3, we need to check if it's installed, and if not, install it
-            all_models = list_ollama_models()
-            if "llama3" not in all_models:
-                interpreter.display_message(f"\nDownloading Llama3...\n")
-                subprocess.run(["ollama", "pull", "llama3"], check=True)
+        if "download" in selected_name.lower():
+            model = selected_name.split(" ")[-1]
+            interpreter.display_message(f"\nDownloading {model}...\n")
+            subprocess.run(["ollama", "pull", model], check=True)
+        else:
+            model = selected_name.strip()
 
         # Set the model to the selected model
-        interpreter.llm.model = f"ollama/{selected_name}"
-        interpreter.display_message(f"\nUsing Ollama model: `{selected_name}` \n")
-        time.sleep(1)
+        interpreter.llm.model = f"ollama/{model}"
+        interpreter.display_message(f"> Model set to `{model}`")
 
     # If Ollama is not installed or not recognized as a command, prompt the user to download Ollama and try again
     except (subprocess.CalledProcessError, FileNotFoundError) as e:
@@ -359,15 +334,17 @@ def list_ollama_models():
     models = [f for f in os.listdir(models_dir) if f.endswith(".llamafile")]
 
     if not models:
-        print("\nThere are no models currently downloaded. Let's download a new one.\n")
+        print(
+            "\nNo models currently downloaded. Please select a new model to download.\n"
+        )
         model_path = download_model(models_dir, models, interpreter)
     else:
         # Prompt the user to select a downloaded model or download a new one
-        model_choices = models + [" ↓ Download new model"]
+        model_choices = models + ["↓ Download new model"]
         questions = [
             inquirer.List(
                 "model",
-                message="Select a Llamafile model to run or download a new one:",
+                message="Select a model",
                 choices=model_choices,
             )
         ]
@@ -376,7 +353,7 @@ def list_ollama_models():
         if answers == None:
             exit()
 
-        if answers["model"] == " ↓ Download new model":
+        if answers["model"] == "↓ Download new model":
             model_path = download_model(models_dir, models, interpreter)
         else:
             model_path = os.path.join(models_dir, answers["model"])
@@ -392,11 +369,8 @@ def list_ollama_models():
                     text=True,
                 )
 
-                print("Waiting for the model to load...")
                 for line in process.stdout:
                     if "llama server listening at http://127.0.0.1:8080" in line:
-                        print("\nModel loaded \n")
-                        time.sleep(1)
                         break  # Exit the loop once the server is ready
             except Exception as e:
                 process.kill()  # Force kill if not terminated after timeout
@@ -409,6 +383,9 @@ def list_ollama_models():
     interpreter.llm.api_base = "http://localhost:8080/v1"
     interpreter.llm.supports_functions = False
 
+    model_name = model_path.split("/")[-1]
+    interpreter.display_message(f"> Model set to `{model_name}`")
+
 user_ram = total_ram = psutil.virtual_memory().total / (
     1024 * 1024 * 1024
 )  # Convert bytes to GB
@@ -419,6 +396,15 @@ def list_ollama_models():
 else:
     interpreter.llm.max_tokens = 1000
     interpreter.llm.context_window = 3000
+
+# Display intro message
+if interpreter.auto_run == False:
+    interpreter.display_message(
+        "**Open Interpreter** will require approval before running code."
+        + "\n\nUse `interpreter -y` to bypass this."
+        + "\n\nPress `CTRL-C` to exit.\n"
+    )
+
 # Set the system message to a minimal version for all local models.
 interpreter.system_message = """
 You are Open Interpreter, a world-class programmer that can execute code on the user's machine.
@@ -450,14 +436,14 @@ def list_ollama_models():
 Now, your turn:
 """
 
-interpreter.user_message_template = "{content} Please send me some code that would be able to answer my question, in the form of ```python\n... the code ...\n``` or ```shell\n... the code ...\n```"
+# interpreter.user_message_template = "{content} Please send me some code that would be able to answer my question, in the form of ```python\n... the code ...\n``` or ```shell\n... the code ...\n```"
 interpreter.code_output_template = "I executed that code. This was the ouput: {content}\n\nWhat does this output mean / what's next (if anything, or are we done)?"
 interpreter.empty_code_output_template = "The code above was executed on my machine. It produced no text output. what's next (if anything, or are we done?)"
 interpreter.code_output_sender = "user"
 interpreter.max_output = 500
 interpreter.llm.context_window = 8000
 interpreter.force_task_completion = False
-# interpreter.user_message_template = "{content}. If my question must be solved by running code on my computer, send me code to run enclosed in ```python or ```shell. Otherwise, don't send code and answer like a chatbot. Be concise, don't include anything unnecessary. Don't use placeholders, I can't edit code."
+interpreter.user_message_template = "{content}. If my question must be solved by running code on my computer, send me code to run enclosed in ```python or ```shell. Otherwise, don't send code. Be concise, don't include anything unnecessary. Don't use placeholders, I can't edit code."
 
 # Set offline for all local models
 interpreter.offline = True