Small fix for ollama truncation check. Small fix for pythonic tool repr rendering.

Author: monoxgas

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "rigging"
-version = "3.3.4"
+version = "3.3.5"
 description = "LLM Interaction Framework"
 authors = ["Nick Landers <monoxgas@gmail.com>"]
 license = "MIT"

rigging/generator/litellm_.py

@@ -254,13 +254,17 @@ def _warn_on_input_truncation(
         # - https://github.com/ollama/ollama/issues/7987
         # - https://github.com/ollama/ollama/issues/4967
 
+        # Not perfect
+        if "ollama" not in self.model.lower():
+            return
+
         # We can't check with usage info
         if not response.usage:
             return
 
         # Get a general view of how long we might expect the input prompt to
-        # We'll use a gracious 4 char per token estimate
-        input_tokens_estimate = int(sum(len(message.content) for message in messages) / 4)
+        # We'll use a gracious 10 char per token estimate
+        input_tokens_estimate = int(sum(len(message.content) for message in messages) / 10)
 
         # Check if the response reports that accepted input tokens are less than this
         if response.usage.input_tokens < input_tokens_estimate:

rigging/transform/pythonic_tools.py

@@ -104,7 +104,7 @@ def _attempt_parse_tool_calls_from_string(candidate_str: str) -> list[ToolCall]
 def _render_tool_call_to_pythonic_string(tool_call: ToolCall) -> str:
     """Renders a single ToolCall object to its pythonic string representation."""
     args_dict = json.loads(tool_call.function.arguments)
-    args_str = ", ".join(f"{key}={json.dumps(value)}" for key, value in args_dict.items())
+    args_str = ", ".join(f"{key}={value!r}" for key, value in args_dict.items())
     return f"{tool_call.function.name}({args_str})"