Troubleshooting
Messages on the stats screen
| Message |
Fix |
| NO LINK (red light in SETUP) |
the server isn't running or the address is wrong: start it, or check the address in SETUP. If it stopped at startup, its window says why |
| LINK LOST |
the server stopped answering mid-request; nothing was written. Retry. |
| NO REPLY |
the server didn't answer in time: check it's still running, or try a smaller request |
| SRV BUSY |
the server's queue is full; the device retries by itself |
| NO TARGET |
nothing to generate: paint some bars or turn on AR |
| TIME SIG |
the scene's clips use different time signatures |
| TOO FEW BARS |
the scene is shorter than CONTEXT: lower it in ENGINE, or lengthen the clips |
| REJECTED |
the server refused a setting, and the screen says which (for example, density on a melodic track) |
| MODEL GONE |
the chosen model isn't on the server any more: pick one in ENGINE |
| CHECKS FAIL |
every attempt was rejected by NOVELTY or SILENCE: raise ATTEMPTS or ESC, or turn a check off |
| CTX OVERFLOW |
the scene is too long for the model: ignore some tracks or lower CONTEXT |
| CTX LIMIT |
the takes may be cut short: lower CONTEXT or generate fewer bars |
| TAKES LOST |
some takes failed the checks; the others are ready |
| TAG FAILED |
Live didn't load the tag device: drag MIDI-GPT Instrument from the MIDI-GPT folder onto the track |
| SRV FAULT · BAD REQUEST |
something broke on the server or in the device: retry, and if it repeats, report it with the server window's lines and its failed_requests.jsonl |
Before generating
GENERATE checks the settings first, and instead of sending a request the server would refuse, the OUTPUT
screen says what to change:
| Message |
Change |
| BARS/STEP > CONTEXT |
lower BARS/STEP |
| REMIX + RHYTHM |
turn one of them off on that track |
| REMIX AT 0% |
raise AMOUNT: at 0% remix would change nothing |
| REMIX / EXACT RHYTHM NEEDS TRK/STEP 1 |
set TRK/STEP to 1 (or AUTO) |
| MASK K ≥ TOP-K · MASK P ≥ TOP-P |
lower MASK K / MASK P: they would skip every choice |
Other problems
| Symptom |
Fix |
| The server window says the model is gated, or no access |
accept the license at huggingface.co/Metacreation/MIDI-GPT with your Hugging Face account and save a token (see Install) |
Windows: the server stops with a DLL error (WinError 126, c10.dll) |
PyTorch needs Microsoft's Visual C++ runtime: install vc_redist.x64.exe |
| Windows: "Windows protected your PC" when starting the server |
click More info › Run anyway, or right-click the .bat › Properties › Unblock (the installer unblocks the files itself) |
| Port 3456 already in use |
another MIDI-GPT server is running (MIDI-GPT for REAPER uses the same port): use it, or start this one with --port 3457 and change the address |
| The device won't load after a download |
clear the macOS quarantine flag (see Install) |
| The first generation is slow |
the server loads the model on the first request; later ones are faster |
| Generation is slow overall |
on a Mac the GPU (MPS) can be slower than the CPU: start the server with start-server.sh --device cpu. Else try yellow_small, or run the server on a faster machine |
Model limits
- DENSITY only works on drum tracks; POLY and LENGTH only on melodic tracks.
- REGISTER only applies with SCALE on.
- REMIX replaces the rhythm settings on the same track. POLY LIM and DENS LIM win over it: a wider chord
or a fuller bar comes back with fewer notes.
Files
|
|
| Settings |
~/Library/Application Support/MIDI-GPT/live.json |
| Server |
~/Library/Application Support/MIDI-GPT/server (its window shows the log) |
| Models and the Hugging Face token |
~/.cache/huggingface |
| Requests that failed on the server |
failed_requests.jsonl in the server folder |
Still stuck? Report it with what you did and what the
stats screen said.