https://docs.all-hands.dev/ logo
Join Slack
Powered by
<@U090ER7JPHA> In openhands/openhands repo, invest...
# slackbot-chatter
m
@nutritious-balloon-65098 In openhands/openhands repo, investigate why grok-code-fast-1 model encounters 
Missing required parameters for function 'execute_bash': {'security_risk'}
errors
n
I'm on it! can track my progress here.
Investigation Summary: grok-code-fast-1 Missing security_risk Parameter Error I investigated why the grok-code-fast-1 model encounters 
Missing required parameters for function 'execute_bash': {'security_risk'}
errors in the OpenHands repository. Root Cause Identified The issue occurs because: 1. grok-code-fast-1 is not recognized as supporting native function calling - It's missing from the 
FUNCTION_CALLING_PATTERNS
list in 
openhands/llm/model_features.py
2. OpenHands falls back to text-based function call conversion - Since the model isn't in the function calling patterns, OpenHands uses the function call converter (
openhands/llm/fn_call_converter.py
) 3. Strict parameter validation fails - The converter requires all mandatory parameters including 
security_risk
, but grok-code-fast-1 omits this parameter when generating text-based function calls 4. Error occurs during validation - The validation error is thrown at line 693 in 
fn_call_converter.py
Evidence Found • Model capabilities: Research confirms grok-code-fast-1 supports "tool use", "function calling", and "structured output" • Current configuration: The model is only listed in 
SUPPORTS_STOP_WORDS_FALSE_PATTERNS
but missing from 
FUNCTION_CALLING_PATTERNS
• Recent commit: Found commit d5d5e265f that added grok-code-fast-1 to the stop words exclusion list but didn't add function calling support Recommended Solution Add grok-code-fast-1 to the 
FUNCTION_CALLING_PATTERNS
list in `openhands/llm/model_features.py`:
Copy code
FUNCTION_CALLING_PATTERNS: list[str] = [
    # ... existing patterns ...
    # xAI Grok models  
    'grok-code-fast-1',
    # ... rest of patterns ...
]
This solution: • ✅ Addresses the root cause by enabling native function calling • ✅ Aligns with the model's actual capabilities • ✅ Follows existing patterns for other function-calling models • ✅ Provides better performance and reliability than text conversion • ✅ Has minimal risk (only affects grok-code-fast-1) No code changes were made as this was an investigation task. The user should implement the recommended solution by adding 'grok-code-fast-1' to the FUNCTION_CALLING_PATTERNS list to resolve the missing security_risk parameter errors. <sub>View full conversation</sub>
m
What’s the config that turns on function calling?
n
Investigation Summary: grok-code-fast-1 Missing security_risk Parameter Error I investigated why the grok-code-fast-1 model encounters 
Missing required parameters for function 'execute_bash': {'security_risk'}
errors in the OpenHands repository. Root Cause Identified The issue occurs because: 1. grok-code-fast-1 is not recognized as supporting native function calling - It's missing from the 
FUNCTION_CALLING_PATTERNS
list in 
openhands/llm/model_features.py
2. OpenHands falls back to text-based function call conversion - Since the model isn't in the function calling patterns, OpenHands uses the function call converter (
openhands/llm/fn_call_converter.py
) 3. The converter strictly validates required parameters - When parsing text-based function calls, it requires all mandatory parameters including 
security_risk
4. The model omits the security_risk parameter - When generating text-based function calls, grok-code-fast-1 doesn't consistently include the 
security_risk
parameter, causing validation to fail at line 693 in 
fn_call_converter.py
Configuration That Forces Native Function Calling Answer to your question: The configuration that forces turning on native function calling is the 
FUNCTION_CALLING_PATTERNS
list in 
openhands/llm/model_features.py
. Models listed in this array are recognized as supporting native function calling and bypass the text-based function call converter entirely. Recommended Solution Add 
'grok-code-fast-1'
to the 
FUNCTION_CALLING_PATTERNS
list in 
openhands/llm/model_features.py
. Research confirms that grok-code-fast-1 officially supports "tool use", "function calling", and "structured output" according to xAI documentation. This would enable native function calling for grok-code-fast-1, eliminating the need for error-prone text-based function call conversion and resolving the missing 
security_risk
parameter issue. Key Files Involved • 
openhands/llm/model_features.py
- Contains function calling patterns • 
openhands/llm/fn_call_converter.py
- Where the validation error occurs • 
openhands/agenthub/codeact_agent/tools/bash.py
- Defines the execute_bash tool with required security_risk parameter No code changes were made as this was an investigation task. The solution would require adding grok-code-fast-1 to the function calling patterns to enable its native function calling capabilities. <sub>View full conversation</sub>
Investigation Summary: grok-code-fast-1 Missing security_risk Parameter Error Root Cause The 
grok-code-fast-1
model encounters 
Missing required parameters for function 'execute_bash': {'security_risk'}
errors because: 1. grok-code-fast-1 is not recognized as supporting native function calling - It's missing from 
FUNCTION_CALLING_PATTERNS
in 
openhands/llm/model_features.py
2. OpenHands falls back to text-based function call conversion - When 
is_function_calling_active()
returns 
False
, the system uses 
convert_fncall_messages_to_non_fncall_messages()
and 
convert_non_fncall_messages_to_fncall_messages()
3. The converter strictly validates required parameters - The validation occurs in 
_extract_and_validate_params()
in 
openhands/llm/fn_call_converter.py
at line 693 4. The model omits the security_risk parameter - When generating text-based function calls, grok-code-fast-1 doesn't include the required 
security_risk
parameter Configuration That Forces Native Function Calling Answer to your question: Yes, the 
native_tool_calling
LLM config would work! Here's how:
Copy code
# In LLMConfig (openhands/core/config/llm_config.py):
native_tool_calling: bool | None = Field(default=None)
How it works:
Copy code
# In openhands/llm/llm.py lines 539-542:
features = get_features(self.config.model)
if self.config.native_tool_calling is None:
    self._function_calling_active = features.supports_function_calling  # Uses model patterns
else:
    self._function_calling_active = self.config.native_tool_calling     # Uses explicit config
To fix the grok-code-fast-1 issue, users can set:
Copy code
[llm]
model = "grok-code-fast-1"
native_tool_calling = true
This would bypass the problematic function call converter and use native function calling directly. Recommended Long-term Solution Add 
'grok-code-fast-1'
to 
FUNCTION_CALLING_PATTERNS
in 
openhands/llm/model_features.py
since research confirms the model supports native function calling, tool use, and structured outputs. Key Insight The 
native_tool_calling
config provides a manual override that can force any model to use (or not use) native function calling, regardless of what the model feature patterns indicate. This is the immediate workaround for the grok-code-fast-1 issue. <sub>View full conversation</sub>
m
@nutritious-balloon-65098 I’ve tested that enable 
native_tool_calling
config works, which proves this 
grok-code-fast-1
supports native tool calling. Please create a PR to add this model to 
FUNCTION_CALLING_PATTERNS
list
n
Uh oh! There was an unexpected error starting the job :(
Open in Slack
PreviousNext
Powered by