Fix Monitor.update_model_call_limit always rejecting every call

[Chessing234]

Bug

Monitor.update_model_call_limit exits early and returns False whenever the model is not already present in self.model_call_limit_global, but the dict is only ever initialized to {} in Monitor.__init__ and no other code populates it. The first update for any model always fails, the dict stays empty, and every subsequent update also fails - so the /update_model_call_limit/{model}/{limit} endpoint is unreachable in practice.

https://github.com/lm-sys/FastChat/blob/587d5cf/fastchat/serve/call_monitor.py#L73-L77

def update_model_call_limit(self, model: str, limit: int) -> bool:
    if model not in self.model_call_limit_global:
        return False
    self.model_call_limit_global[model] = limit
    return True

Root cause

The early-return guards against the only code path that would populate the dict, so the dict never gets populated in the first place.

Fix

Drop the early-return and upsert the limit unconditionally. The neighbouring accessors already treat a missing key as "no limit configured":

• get_model_call_limit returns -1 when the key is absent.
• is_model_limit_reached returns False when the key is absent.

so allowing the first call to create the entry is consistent with the rest of the class.