From f93096f2bf6f9a46efaff74752edc65477cb661e Mon Sep 17 00:00:00 2001
From: Frank <frank@dankware>
Date: Tue, 11 Aug 2026 00:57:04 -0700
Subject: [PATCH] feat: frank-default-small = HauhauCS Qwen3.6-35B-A3B
 uncensored Q6_K_P + DFlash on dankserver

llama.cpp service replacing retired vllm-awq-qwen36 on port 7770 with alias
frank-default-small. Wrapper sources frank-default-small.env for tunables:
3 parallel slots at 262144 ctx, draft-dflash n-max 7, KV K=q8_0 V=f16 on GPU,
all MoE experts on CPU (--n-cpu-moe 40), non-expert tensors on GPU, mmproj
vision + jinja. Unit: frank-default-small.service (ExecStart in dankserver
repo clone /home/dank/vLLM/scripts).
---
 scripts/frank-default-small.env     | 18 ++++++++
 scripts/frank-default-small.service | 14 ++++++
 scripts/serve-llama-frank-small.sh  | 67 +++++++++++++++++++++++++++++
 3 files changed, 99 insertions(+)
 create mode 100644 scripts/frank-default-small.env
 create mode 100644 scripts/frank-default-small.service
 create mode 100755 scripts/serve-llama-frank-small.sh

diff --git a/scripts/frank-default-small.env b/scripts/frank-default-small.env
new file mode 100644
index 0000000..444aa79
--- /dev/null
+++ b/scripts/frank-default-small.env
@@ -0,0 +1,18 @@
+# Tunables for frank-default-small (HauhauCS Q6_K_P + DFlash draft, dankserver)
+# Runtime: llama-server b10108; RTX 3090 24GB; Ryzen 7 7700X; 62GB RAM.
+MODEL_DIR=/home/dank/models
+ALIAS=frank-default-small
+HOST=0.0.0.0
+PORT=7770
+CTX_SIZE=262144
+PARALLEL=3
+SPEC_TYPE=draft-dflash
+SPEC_N_MAX=7
+CACHE_TYPE_K=q8_0
+CACHE_TYPE_V=f16
+# All 40 layers keep their MoE expert weights on CPU (targeted expert offload).
+N_CPU_MOE=40
+# Every non-expert tensor (attention, norms, embedding, LM head) goes to GPU.
+N_GPU_LAYERS=all
+THREADS=8
+THREADS_HTTP=8
\ No newline at end of file
diff --git a/scripts/frank-default-small.service b/scripts/frank-default-small.service
new file mode 100644
index 0000000..770ebb4
--- /dev/null
+++ b/scripts/frank-default-small.service
@@ -0,0 +1,14 @@
+[Unit]
+Description=llama.cpp DFlash inference (frank-default-small)
+After=network.target
+
+[Service]
+Type=simple
+User=dank
+Group=dank
+ExecStart=/home/dank/vLLM/scripts/serve-llama-frank-small.sh
+Restart=on-failure
+RestartSec=10
+
+[Install]
+WantedBy=multi-user.target
\ No newline at end of file
diff --git a/scripts/serve-llama-frank-small.sh b/scripts/serve-llama-frank-small.sh
new file mode 100755
index 0000000..5fe8969
--- /dev/null
+++ b/scripts/serve-llama-frank-small.sh
@@ -0,0 +1,67 @@
+#!/usr/bin/env bash
+# Serve HauhauCS Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive:Q6_K_P
+# with a DFlash draft model on dankserver (RTX 3090, MoE experts on CPU).
+# Runtime: upstream llama.cpp llama-server (b10108+).
+#
+# Tunables come from frank-default-small.env in this directory unless the
+# FRANK_SMALL_ENV_FILE environment variable overrides that path.
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+ENV_FILE="${FRANK_SMALL_ENV_FILE:-$SCRIPT_DIR/frank-default-small.env}"
+if [[ -f "$ENV_FILE" ]]; then
+    set -a
+    # shellcheck disable=SC1090
+    source "$ENV_FILE"
+    set +a
+fi
+
+# --- defaults (used when the env file does not set them) -----------------
+: "${MODEL_DIR:=/home/dank/models}"
+: "${MODEL_FILE:=Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Q6_K_P.gguf}"
+: "${DRAFT_FILE:=Qwen3.6-35B-A3B-DFlash-Q8_0.gguf}"
+: "${MMPROJ_FILE:=mmproj-Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-f16.gguf}"
+: "${ALIAS:=frank-default-small}"
+: "${HOST:=0.0.0.0}"
+: "${PORT:=7770}"
+: "${CTX_SIZE:=262144}"
+: "${PARALLEL:=3}"
+: "${SPEC_TYPE:=draft-dflash}"
+: "${SPEC_N_MAX:=7}"
+: "${CACHE_TYPE_K:=q8_0}"
+: "${CACHE_TYPE_V:=f16}"
+: "${N_CPU_MOE:=40}"     # all 40 layers: MoE expert weights stay on CPU
+: "${N_GPU_LAYERS:=all}" # every non-expert tensor goes to GPU
+: "${THREADS:=8}"        # 7700X physical cores
+: "${THREADS_HTTP:=8}"
+
+LLAMA_SERVER="${LLAMA_SERVER:-/home/dank/llama.cpp/build/bin/llama-server}"
+MODEL="$MODEL_DIR/$MODEL_FILE"
+DRAFT="$MODEL_DIR/$DRAFT_FILE"
+MMPROJ="$MODEL_DIR/$MMPROJ_FILE"
+
+for f in "$MODEL" "$DRAFT" "$MMPROJ"; do
+    if [[ ! -f "$f" ]]; then
+        echo "missing required file: $f" >&2
+        exit 1
+    fi
+done
+
+exec "$LLAMA_SERVER" \
+    --model "$MODEL" \
+    --model-draft "$DRAFT" \
+    --mmproj "$MMPROJ" \
+    --jinja \
+    --spec-type "$SPEC_TYPE" \
+    --spec-draft-n-max "$SPEC_N_MAX" \
+    --cache-type-k "$CACHE_TYPE_K" \
+    --cache-type-v "$CACHE_TYPE_V" \
+    --ctx-size "$CTX_SIZE" \
+    --parallel "$PARALLEL" \
+    --n-cpu-moe "$N_CPU_MOE" \
+    --gpu-layers "$N_GPU_LAYERS" \
+    --threads "$THREADS" \
+    --threads-http "$THREADS_HTTP" \
+    --alias "$ALIAS" \
+    --host "$HOST" \
+    --port "$PORT"
\ No newline at end of file