Test feedback submission for Hybrid/Recurrent models

[JamePeng]

Test Feedback: Hybrid / Recurrent Models

Hi all,

I am collecting real-world test results for hybrid+recurrent model/multimodel and HybridCheckpointCache manager.

If you're testing any of these, please reply with:

• model & repo
• llama-cpp-python commit/branch
• important flags (pls set the verbose=True) and runtime logs
• hardware and system environment
• tokens/s & memory/vram usage
• any crashes, quality issues, potent bugs, or surprising results

Even one log or good point helps a lot. Thank you!

[yamikumo-DSD]

I read the latest update.
In this part of MTMDChatHander class

           if self._mtmd_cpp.mtmd_support_vision(self.mtmd_ctx) and self.verbose:
                print(f"{self.log_prefix}(_init_mtmd_context): Vision support detected.", file=sys.stderr)
           else:
                raise ValueError("Vision is not supported by this model")

, I guess that ValueError raised when verbose=False is not your intention.
Isn't this what you wanna do?

           if self._mtmd_cpp.mtmd_support_vision(self.mtmd_ctx):
                if self.verbose:
                    print(f"{self.log_prefix}(_init_mtmd_context): Vision support detected.", file=sys.stderr)
                else:
                    pass # No logging
           else:
                raise ValueError("Vision is not supported by this model")

[JamePeng]

Yeah, thank you!

[JamePeng]

Test result of Qwen3.5 :)