File size: 1,636 Bytes
51e64f4 d90d9dc a5b969d d90d9dc a5b969d 51e64f4 a5b969d 5c7f0a5 51e64f4 a5b969d 51e64f4 5c7f0a5 51e64f4 a5b969d 51e64f4 d90d9dc 5c7f0a5 d90d9dc 5c7f0a5 a5b969d 5c7f0a5 51e64f4 5c7f0a5 a5b969d 51e64f4 a5b969d 51e64f4 a5b969d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 |
from command_classifier import classify_command
from executor import execute_command
def get_assistant_response(pipe, command: str, execute: bool, max_new_tokens: int = 1024) -> str:
"""
Process the user command using the language model pipeline and optionally execute a system command.
Args:
pipe: The Hugging Face model pipeline
command (str): The user's text input
execute (bool): Whether or not to execute a real command on the machine
max_new_tokens (int): Maximum tokens to generate (longer = more detailed)
Returns:
str: Response from the assistant (and optionally OS output)
"""
command_type = classify_command(command)
# If execution is allowed, try to run a shell command
if execute:
exec_output = execute_command(command_type, command)
if exec_output:
return f"🛠️ Executed:\n{exec_output}"
# Prepare prompt for chat-style models like Qwen3
prompt = f"<|user|>\n{command}\n<|assistant|>\n"
try:
output = pipe(
prompt,
return_full_text=True,
max_new_tokens=max_new_tokens,
do_sample=True,
temperature=0.7,
top_p=0.9
)
generated_text = output[0]["generated_text"]
# Extract only assistant response
if "<|assistant|>" in generated_text:
response = generated_text.split("<|assistant|>")[-1].strip()
else:
response = generated_text.strip()
return f"🤖 ({command_type}) {response}"
except Exception as e:
return f"⚠️ Model error: {e}"
|