Audio upload/transcribe not working on llama-swap

[JamesMowery]

Getting a 500 error when attempting to upload/transcribe audio files (only tested a small mp3). llama-server saying that audio isn't supported (but it works fine with OWUI).

llms output llama-swap error log:

Oct 31 11:34:57 phoenix llama-swap[63587]: main: server is listening on http://127.0.0.1:5818 - starting the main loop
Oct 31 11:34:57 phoenix llama-swap[63587]: srv  update_slots: all slots are idle
Oct 31 11:34:58 phoenix llama-swap[63587]: srv  log_server_r: request: GET /health 127.0.0.1 200
Oct 31 11:34:58 phoenix llama-swap[63587]: [INFO] <Qwen3-VL-8B-Instruct-BF16> Health check passed on http://localhost:5818/health
Oct 31 11:34:58 phoenix llama-swap[63587]: [DEBUG] <Qwen3-VL-8B-Instruct-BF16> swapState() State transitioned from starting to ready
Oct 31 11:34:58 phoenix llama-swap[63587]: got exception: {"code":500,"message":"audio input is not supported - hint: if this is unexpected, you may need to provide the mmproj","type":"server_error"}
Oct 31 11:34:58 phoenix llama-swap[63587]: srv  log_server_r: request: POST /v1/chat/completions 127.0.0.1 500
Oct 31 11:34:58 phoenix llama-swap[63587]: [DEBUG] <Qwen3-VL-8B-Instruct-BF16> request /v1/chat/completions - start: 10.265148086s, total: 10.277691021s
Oct 31 11:34:58 phoenix llama-swap[63587]: [WARN] metrics skipped, HTTP status=500, path=/v1/chat/completions
Oct 31 11:34:58 phoenix llama-swap[63587]: [INFO] Request ::1 "POST /v1/chat/completions HTTP/1.1" 500 151 "llms.py/1.0" 10.278364192s

Open-WebUI output:

Oct 31 11:36:40 phoenix llama-swap[63587]:        eval time =     637.21 ms /    19 tokens (   33.54 ms per token,    29.82 tokens per second)
Oct 31 11:36:40 phoenix llama-swap[63587]:       total time =     704.88 ms /   242 tokens
Oct 31 11:36:40 phoenix llama-swap[63587]: slot      release: id  0 | task 145 | stop processing: n_tokens = 248, truncated = 0
Oct 31 11:36:40 phoenix llama-swap[63587]: srv  update_slots: all slots are idle
Oct 31 11:36:40 phoenix llama-swap[63587]: srv  log_server_r: request: POST /v1/chat/completions 127.0.0.1 200
Oct 31 11:36:40 phoenix llama-swap[63587]: [DEBUG] <Qwen3-VL-8B-Instruct-BF16> request /v1/chat/completions - start: 0s, total: 707.956034ms
Oct 31 11:36:40 phoenix llama-swap[63587]: [INFO] Request ::1 "POST /v1/chat/completions HTTP/1.1" 200 661 "Python/3.12 aiohttp/3.12.15" 708.027884ms

[JamesMowery]

Source audio file I tested it with, just in case it's useful:

the-pill-cut.mp3

[mythz]

This audio file works with the other Audio capable models:

So it's a case of llama.cpp/swap not supporting embedded audio messages.

If that's the case we'd need a custom OpenAiProvider for llama.cpp/swap which I'll look at setting up in the near future. For now I'll need to port all these latest UI changes to the C# version.