starfleet/computer
starfleet/computer
Watch
1
0
Fork
You've already forked computer
 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

feat: Q4_K_M CUDA wheel rebuild (q6_K) for 12GB headless #10

Merged
troed merged 3 commits from devel into main 2026-08-27 14:52:14 +02:00
Conversation 0 Commits 3 Files changed 2 +160 -2
troed commented 2026-08-27 14:52:00 +02:00
Owner
Copy link

Enable Q4_K_M on 12GB headless (RTX A2000).

  • PyPI qwentts-cpp-python 0.3.1 pins qwentts.cpp 7df559a (ABI 2) without CUDA getrows for GGML_TYPE_Q6_K → Q4_K_M GGUFs (Serveurperso/Qwen3-TTS-GGUF) abort unsupported src0 type: q6_K.
  • Newer qwentts.cpp a8a7716 bumps QT_ABI 2→4 (qt_tts_params removes codec_chunk_sec) → Python binding at ABI 2 mismatches → --ref-text requires --ref-wav abort.
  • Fix: stay on 7df559a (ABI 2) + cherry-pick ggml c044c6f Q6_K getrows, single-arch 86-real build. Verified headless: Talker 1765→1012 MB, nvidia-smi 11439→10597 (1309 free), no abort, TTS warmup ok.

Adds scripts/build_qwentts_wheel.sh (clone qwentts.cpp/qwentts-cpp-python, cmake -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=86-real, build_native.py, wheel 1cu128, pip --no-deps) and README Q4_K_M VRAM table + rebuild docs.

Deployed to headless, tested Q4_K_M synthesis, service stable.

Enable Q4_K_M on 12GB headless (RTX A2000). - PyPI qwentts-cpp-python 0.3.1 pins qwentts.cpp 7df559a (ABI 2) without CUDA getrows for GGML_TYPE_Q6_K → Q4_K_M GGUFs (Serveurperso/Qwen3-TTS-GGUF) abort `unsupported src0 type: q6_K`. - Newer qwentts.cpp a8a7716 bumps QT_ABI 2→4 (qt_tts_params removes codec_chunk_sec) → Python binding at ABI 2 mismatches → `--ref-text requires --ref-wav` abort. - Fix: stay on 7df559a (ABI 2) + cherry-pick ggml c044c6f Q6_K getrows, single-arch 86-real build. Verified headless: Talker 1765→1012 MB, nvidia-smi 11439→10597 (1309 free), no abort, TTS warmup ok. Adds scripts/build_qwentts_wheel.sh (clone qwentts.cpp/qwentts-cpp-python, cmake -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=86-real, build_native.py, wheel 1cu128, pip --no-deps) and README Q4_K_M VRAM table + rebuild docs. Deployed to headless, tested Q4_K_M synthesis, service stable.
docs: update demo video to vwTvn6qsVtQGsSW8TJMBDj
All checks were successful
CI / Sanity check (ubuntu-latest) (pull_request) Successful in 4m18s
Details
77082e83cb 
The PyPI qwentts-cpp-python 0.3.1 wheel (qwentts.cpp 7df559a) lacks the
ggml-cuda getrows kernel for GGML_TYPE_Q6_K, so Q4_K_M GGUFs abort with
unsupported src0 type: q6_K. Rebuilding against main/a8a7716+ includes the
case. Add scripts/build_qwentts_wheel.sh (clone+build_native+wheel) and
document VRAM table and rebuild steps in README.
fix: build_qwentts_wheel default ref master and tolerant checkout
All checks were successful
CI / Sanity check (ubuntu-latest) (pull_request) Successful in 3m52s
Details
91a59532ba 
qwentts.cpp default branch is master, not main. Also shallow clone
already on correct ref, so skip unnecessary checkout that aborted the
build with set -e.
troed merged commit 3946978a4c into main 2026-08-27 14:52:14 +02:00
troed deleted branch devel 2026-08-27 14:52:15 +02:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
starfleet/computer!10
No description provided.