8000 Eval bug: `llama-tts` fails (abort) with longer lines · Issue #13712 · ggml-org/llama.cpp · GitHub
[go: up one dir, main page]
Skip to content

Eval bug: llama-tts fails (abort) with longer lines #13712

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
Disonantemus opened this issue May 22, 2025 · 1 comment · Fixed by #13713
Closed

Eval bug: llama-tts fails (abort) with longer lines #13712

Disonantemus opened this issue May 22, 2025 · 1 comment · Fixed by #13713
Labels
bug Something isn't working

Comments

@Disonantemus
Copy link
Disonantemus commented May 22, 2025 •
edited
Loading

Name and Version

Last (binaries) version from Releases:
llama-b5456-bin-ubuntu-vulkan-x64.zip

$ llama-tts --version
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce GTX 1660 SUPER (NVIDIA) | uma: 0 | fp16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: none
version: 5456 (cc74d5be)
built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu

Operating systems

Linux

GGML backends

Vulkan

Hardware

System (updated today):

OS: Arch Linux x86_64
Kernel: Linux 6.12.24-1-lts
Shell: bash 5.2.37
WM: dwm (X11)
Terminal: tmux 3.5a
CPU: Intel(R) Core(TM) i7-4790 (8) @ 3.60 GHz
GPU: NVIDIA GeForce GTX 1660 SUPER [Discrete]
Memory: 2.83 GiB / 15.56 GiB (18%)
Locale: en_US.UTF-8

Models

--tts-oute-default (OuteTTS-0.2-500M)

Problem description & steps to reproduce

  1. This works:

    $ llama-tts --tts-oute-default -p "The main goal of llama.cpp is to enable LLM inference with minimal setup and state-of-the-art performance" && aplay output.wav

  2. This doesn't (and abort):

    $ llama-tts --tts-oute-default -p "The main goal of llama.cpp is to enable LLM inference with minimal setup and state-of-the-art performance on a wide range of hardware" && aplay output.wav

  3. Difference is only the prompt:

    "on a wide range of hardware"

First Bad Commit

No response

Relevant log output

|1750|><|875|><|933|><|1595|><|1406|><|861|><|437|><|747|><|1542|><|639|><|607|><|1308|><|1427|><|1141|><|1450|><|1304|><|1492|><|1656|>'
main: codes audio size: 544
/home/runner/work/llama.cpp/llama.cpp/src/llama-context.cpp:897: GGML_ASSERT((cparams.causal_attn || cparams.n_ubatch >= n_tokens_all) && "non-causal attention requires n_ubatch >= n_tokens") failed
[New LWP 260885]
[New LWP 260884]
[New LWP 260883]
[New LWP 260882]
[New LWP 260879]
[New LWP 260877]
[New LWP 260876]
[New LWP 260875]
[New LWP 260867]

This GDB supports auto-downloading debuginfo from the following URLs:
  <https://debuginfod.archlinux.org>
Enable debuginfod for this session? (y or [n]) [answered N; input not from terminal]
Debuginfod has been disabled.
To make this setting permanent, add 'set debuginfod enabled off' to .gdbinit.
Function(s) ^std::(move|forward|as_const|(__)?addressof) will be skipped when stepping.
Function(s) ^std::(shared|unique)_ptr<.*>::(get|operator) will be skipped when stepping.
Function(s) ^std::(basic_string|vector|array|deque|(forward_)?list|(unordered_|flat_)?(multi)?(map|set)|span)<.*>::(c?r?(begin|end)|front|back|data|size|empty) will be skipped when stepping.
Function(s) ^std::(basic_string|vector|array|deque|span)<.*>::operator.] will be skipped when stepping.
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/usr/lib/libthread_db.so.1".
0x0000789957280e22 in ?? () from /usr/lib/libc.so.6
#0  0x0000789957280e22 in ?? () from /usr/lib/libc.so.6
#1  0x0000789957274fda in ?? () from /usr/lib/libc.so.6
#2  0x0000789957275024 in ?? () from /usr/lib/libc.so.6
#3  0x00007899572e592f in wait4 () from /usr/lib/libc.so.6
#4  0x00007899578b7f6d in ggml_abort () from /home/kuro/Exec5/libggml-base.so
#5  0x0000789957a03c17 in llama_context::decode(llama_batch&) () from /home/kuro/Exec5/libllama.so
#6  0x0000789957a03d78 in llama_decode () from /home/kuro/Exec5/libllama.so
#7  0x0000568382fde354 in main ()
[Inferior 1 (process 260866) detached]
Aborted (core dumped)
@ggerganov ggerganov mentioned this issue May 22, 2025
Merged
@ggerganov
Copy link
Member

Will be fixed with #13713.

@ggerganov ggerganov added bug Something isn't working and removed bug-unconfirmed labels May 22, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
bug Something isn't working
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

tts : fix n_ubatch + make WavTokenizer cache-less ggml-org/llama.cpp
2 participants
@ggerganov @Disonantemus
0