Add llamacpp container

Services/llamacpp-swap/Scripts/startBaseMedium.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Report descrittivi: 0.6 ok; 0.55 più stabile
+TEMP=${BASE_TEMP:-0.6}
+exec /app/llama-server $BASE_MEDIUM_MODEL \
+  -c $BASE_CONTEXT_SIZE -ngl $BASE_GPU_LAYERS -n $BASE_MAX_TOKENS \
+  --temp $TEMP --top-p 0.9 --top-k 40 --repeat-penalty 1.1 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 600 --host 0.0.0.0 --port 8092 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startBaseMini.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+
+exec /app/llama-server $BASE_MINI_MODEL \
+  -c 4096 -n 128 \
+  --temp 0.2 --top-p 0.9 --top-k 40 --repeat-penalty 1.05 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 600 --host 0.0.0.0 --port 8091 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startBaseTop.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+
+exec /app/llama-server $BASE_TOP_MODEL \
+  -c $BASE_CONTEXT_SIZE -ngl $BASE_GPU_LAYERS -n $BASE_MAX_TOKENS \
+  --temp 0.5 --top-p 0.9 --top-k 40 --repeat-penalty 1.1 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 900 --host 0.0.0.0 --port 8093 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startChat.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Report descrittivi: 0.6 ok; 0.55 più stabile
+TEMP=${GENERAL_TEMP:-0.6}
+exec /app/llama-server $CHAT_MODEL \
+  -c $GENERAL_CONTEXT_SIZE -ngl $GENERAL_GPU_LAYERS -n $GENERAL_MAX_TOKENS \
+  --temp $TEMP --top-p 0.9 --top-k 40 --repeat-penalty 1.1 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 600 --host 0.0.0.0 --port 8093 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startCoder.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+
+exec /app/llama-server $CODER_MODEL \
+  -c $CODER_CONTEXT_SIZE -n $CODER_MAX_TOKENS \
+  --temp 0.3 --top-p 0.9 --top-k 40 --repeat-penalty 1.05 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 600 --host 0.0.0.0 --port 8094 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startCoderMedium.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+
+# Prefer Q6 + mmap; fallback to no-mmap only if explicitly requested
+EXTRA=""
+if [[ "$FORCE_NO_MMAP_CODER" == "1" ]]; then EXTRA="--no-mmap"; fi
+exec /app/llama-server $CODER_MEDIUM_MODEL \
+  -c $CODER_CONTEXT_SIZE -ngl $CODER_GPU_LAYERS -n $CODER_MAX_TOKENS \
+  --temp 0.5 --top-p 0.9 --top-k 40 --repeat-penalty 1.1 $EXTRA \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 900 --host 0.0.0.0 --port 8095 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startCoderMini.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+
+exec /app/llama-server $CODER_MINI_MODEL \
+  -c 4096 -n 256 \
+  --temp 0.3 --top-p 0.9 --top-k 40 --repeat-penalty 1.05 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --timeout 600 --host 0.0.0.0 --port 8094 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startCoderTop.sh

@@ -0,0 +1,32 @@
+#!/bin/bash
+
+# Large model: auto-select mmap based on GPU layers
+# <= 45 layers: use mmap (less VRAM usage, faster startup)
+# > 45 layers: disable mmap (avoids SVM limits)
+LAYERS=${CODER_TOP_GPU_LAYERS:-55}
+MMAP_OPT=""
+if [ "$LAYERS" -gt 45 ]; then
+  MMAP_OPT="--no-mmap"
+  echo "Using --no-mmap (layers=$LAYERS > 45)"
+else
+  echo "Using mmap (layers=$LAYERS <= 45)"
+fi
+exec /app/llama-server $CODER_TOP_MODEL \
+  -c $CODER_CONTEXT_SIZE -ngl $LAYERS -n $CODER_MAX_TOKENS \
+  --temp 0.45 --top-p 0.9 --top-k 40 --repeat-penalty 1.12 \
+  $MMAP_OPT \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 1200 --host 0.0.0.0 --port 8096 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startEmbedding.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+exec /app/llama-server $EMBEDDING_MODEL \
+  --embeddings --pooling mean \
+  --flash-attn auto --threads -1 --threads-http -1 \
+  --timeout 600 --host 0.0.0.0 --port 8096 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startEmbeddingFast.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+exec /app/llama-server $EMBEDDING_FAST_MODEL \
+  --embeddings --pooling mean \
+  --flash-attn auto --threads -1 --threads-http -1 \
+  --timeout 600 --host 0.0.0.0 --port 8095 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startEmbeddingMedium.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+exec /app/llama-server $EMBEDDING_MEDIUM_MODEL \
+  --embeddings --pooling mean \
+  --flash-attn auto --threads -1 --threads-http -1 \
+  --timeout 600 --host 0.0.0.0 --port 8098 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startEmbeddingMini.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+exec /app/llama-server $EMBEDDING_MINI_MODEL \
+  --embeddings --pooling mean \
+  --flash-attn auto --threads -1 --threads-http -1 \
+  --timeout 600 --host 0.0.0.0 --port 8097 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startEmbeddingTop.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+exec /app/llama-server $EMBEDDING_TOP_MODEL \
+  --embeddings --pooling mean \
+  --flash-attn auto --threads -1 --threads-http -1 \
+  --timeout 600 --host 0.0.0.0 --port 8099 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startGeneral.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Report descrittivi: 0.6 ok; 0.55 più stabile
+TEMP=${GENERAL_TEMP:-0.6}
+exec /app/llama-server $GENERAL_MODEL \
+  -c $GENERAL_CONTEXT_SIZE -ngl $GENERAL_GPU_LAYERS -n $GENERAL_MAX_TOKENS \
+  --temp $TEMP --top-p 0.9 --top-k 40 --repeat-penalty 1.1 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 600 --host 0.0.0.0 --port 8092 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/startGeneralFast.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+
+exec /app/llama-server $GENERAL_FAST_MODEL \
+  -c $GENERAL_CONTEXT_SIZE -n 128 \
+  --temp 0.6 --top-p 0.9 --top-k 40 --repeat-penalty 1.05 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 600 --host 0.0.0.0 --port 8091 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID

Services/llamacpp-swap/Scripts/stopBaseMedium.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8092" || true

Services/llamacpp-swap/Scripts/stopBaseMini.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8091" || true

Services/llamacpp-swap/Scripts/stopBaseTop.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8093" || true

Services/llamacpp-swap/Scripts/stopChat.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+# Stop llama-server per DuckAi-Chat
+pkill -f "llama-server.*--port 8093"

Services/llamacpp-swap/Scripts/stopCoder.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+# Stop llama-server per DuckAi-Coder
+pkill -f "llama-server.*--port 8094"

Services/llamacpp-swap/Scripts/stopCoderMedium.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8095" || true

Services/llamacpp-swap/Scripts/stopCoderMini.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8094" || true

Services/llamacpp-swap/Scripts/stopCoderTop.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8096" || true

Services/llamacpp-swap/Scripts/stopEmbedding.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+# Stop llama-server per DuckAi-Embedding
+pkill -f "llama-server.*--port 8096"

Services/llamacpp-swap/Scripts/stopEmbeddingFast.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+# Stop llama-server per DuckAi-EmbeddingFast
+pkill -f "llama-server.*--port 8095"

Services/llamacpp-swap/Scripts/stopEmbeddingMedium.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8098" || true

Services/llamacpp-swap/Scripts/stopEmbeddingMini.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8097" || true

Services/llamacpp-swap/Scripts/stopEmbeddingTop.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+pkill -f "llama-server.*8099" || true

Services/llamacpp-swap/Scripts/stopGeneral.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+# Stop llama-server per DuckAi-General
+pkill -f "llama-server.*--port 8092"

Services/llamacpp-swap/Scripts/stopGeneralFast.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+# Stop llama-server per DuckAi-GeneralFast
+pkill -f "llama-server.*--port 8091"