From 0ba2ea45be856475bbc9cd642692ff6626245ff2 Mon Sep 17 00:00:00 2001 From: Brian Hetherman Date: Sat, 5 Sep 2026 13:40:59 -0400 Subject: [PATCH] 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 --- mtp-relaxed-decoding/entrypoint.sh | 3 ++- open-webui | 2 +- spotify-voice-assistant | 2 +- 3 files changed, 4 insertions(+), 3 deletions(-) diff --git a/mtp-relaxed-decoding/entrypoint.sh b/mtp-relaxed-decoding/entrypoint.sh index b373ef1..8b23cef 100644 --- a/mtp-relaxed-decoding/entrypoint.sh +++ b/mtp-relaxed-decoding/entrypoint.sh @@ -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 diff --git a/open-webui b/open-webui index 6f583de..966140c 160000 --- a/open-webui +++ b/open-webui @@ -1 +1 @@ -Subproject commit 6f583deff99138a19ef77cdd5ebc0dd4a4cd15cd +Subproject commit 966140c3efac42429c1c48fa1774cd1dace1ceb2 diff --git a/spotify-voice-assistant b/spotify-voice-assistant index 7249919..2ed1ca1 160000 --- a/spotify-voice-assistant +++ b/spotify-voice-assistant @@ -1 +1 @@ -Subproject commit 72499190068f3db057d55013a45846bdccf11fc8 +Subproject commit 2ed1ca1e999599954bb762c6d357077e8151300a