Compare commits

..
5 Commits
Author SHA1 Message Date
bhetherman e845016b32 Pin tts-server Sopro model to a known-good HF revision
Loading from the unpinned main branch let an upstream repo restructure
(dropped vocoder_streaming.safetensors) crash-loop the tts container on
every boot for days; pin to the last snapshot with the full file set.
2026-09-27 02:02:43 -04:00
bhethermanandClaude Sonnet 5 2b05169b66 Bump spotify-voice-assistant submodule pointer
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-09-05 14:47:36 -04:00
bhethermanandClaude Sonnet 5 5e832c5496 Bump spotify-voice-assistant submodule pointer
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-09-05 14:33:54 -04:00
bhethermanandClaude Sonnet 5 96c8c547dd Bump spotify-voice-assistant submodule pointer
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-09-05 14:01:53 -04:00
bhethermanandClaude Sonnet 5 0ba2ea45be Add MTP_N_PARALLEL to entrypoint.sh, bump submodule pointers
entrypoint.sh now honors MTP_N_PARALLEL (default 1, unchanged) for
-np instead of hardcoding 1, so open-webui's docker-compose.mtp.yaml
can request multiple llama-server slots.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-09-05 13:40:59 -04:00
4 changed files with 14 additions and 5 deletions

No files matched your search

+2 -1
View File
@@ -9,6 +9,7 @@ set -euo pipefail
: "${MTP_CTX_SIZE:=8192}"
: "${MTP_BATCH_SIZE:=512}"
: "${MTP_TEMP:=0.6}"
: "${MTP_N_PARALLEL:=1}"
args=(
--model "$MTP_TARGET_GGUF"
@@ -18,7 +19,7 @@ args=(
--spec-draft-relaxed-top-n "$MTP_RELAXED_TOP_N" --spec-draft-backend-sampling
--spec-draft-model "$MTP_DRAFT_GGUF" -ngl 999 --spec-draft-ngl 999
--flash-attn auto -c "$MTP_CTX_SIZE" -b "$MTP_BATCH_SIZE" -ub "$MTP_BATCH_SIZE"
--context-shift --keep 4 -np 1 --temp "$MTP_TEMP"
--context-shift --keep 4 -np "$MTP_N_PARALLEL" --temp "$MTP_TEMP"
)
if [[ -n "${MTP_AUTH_TOKEN:-}" ]]; then
+10 -2
View File
@@ -50,8 +50,16 @@ log.info("Loading Kokoro (lang_code=%s) on %s", LANG_CODE, DEVICE)
kokoro_pipeline = KPipeline(lang_code=LANG_CODE, device=DEVICE)
log.info("Kokoro ready")
log.info("Loading Sopro on %s", DEVICE)
sopro_model = SoproTTS.from_pretrained("samuel-vitorino/sopro-v2-turbo", device=DEVICE)
# Pinned instead of tracking the mutable `main` branch: upstream restructured
# the repo on 2026-09-15 (dropped/renamed vocoder_streaming.safetensors and
# speaker_encoder.safetensors), which crash-looped this container on every
# boot since the unpinned load kept re-fetching the broken layout.
SOPRO_REVISION = os.environ.get("TTS_SOPRO_REVISION", "ceeeb86b5fd805e662a5a750d6053ac4f990a45a")
log.info("Loading Sopro on %s (revision=%s)", DEVICE, SOPRO_REVISION)
sopro_model = SoproTTS.from_pretrained(
"samuel-vitorino/sopro-v2-turbo", device=DEVICE, revision=SOPRO_REVISION
)
log.info("Sopro ready")
# References are just resampled/cropped tensors of the voice clip -- cheap to