From 6febea216ea5f8176e7d044a4d8cfff185fbc47a Mon Sep 17 00:00:00 2001
From: pyr0ball <pyroballpcs@gmail.com>
Date: Fri, 27 Feb 2026 06:18:34 -0800
Subject: [PATCH] feat: inject DUAL_GPU_MODE sub-profile in Makefile; update
 manage.sh help

---
 Makefile  | 4 ++++
 manage.sh | 5 ++++-
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/Makefile b/Makefile
index b606fb9..efa8502 100644
--- a/Makefile
+++ b/Makefile
@@ -23,6 +23,7 @@ COMPOSE ?= $(shell \
 # compose.override.yml. We must include it explicitly when present.
 OVERRIDE_FILE := $(wildcard compose.override.yml)
 COMPOSE_OVERRIDE := $(if $(OVERRIDE_FILE),-f compose.override.yml,)
+DUAL_GPU_MODE ?= $(shell grep -m1 '^DUAL_GPU_MODE=' .env 2>/dev/null | cut -d= -f2 || echo ollama)
 
 COMPOSE_FILES := -f compose.yml $(COMPOSE_OVERRIDE)
 ifneq (,$(findstring podman,$(COMPOSE)))
@@ -34,6 +35,9 @@ else
     COMPOSE_FILES := -f compose.yml $(COMPOSE_OVERRIDE) -f compose.gpu.yml
   endif
 endif
+ifeq ($(PROFILE),dual-gpu)
+  COMPOSE_FILES += --profile dual-gpu-$(DUAL_GPU_MODE)
+endif
 
 # 'remote' means base services only — no services are tagged 'remote' in compose.yml,
 # so --profile remote is a no-op with Docker and a fatal error on old podman-compose.
diff --git a/manage.sh b/manage.sh
index 1d2ee5e..3e24197 100755
--- a/manage.sh
+++ b/manage.sh
@@ -42,7 +42,10 @@ usage() {
     echo "    remote       API-only, no local inference (default)"
     echo "    cpu          Local Ollama inference on CPU"
     echo "    single-gpu   Ollama + Vision on GPU 0"
-    echo "    dual-gpu     Ollama + Vision + vLLM on GPU 0+1"
+    echo "    dual-gpu     Ollama + Vision on GPU 0; GPU 1 set by DUAL_GPU_MODE"
+    echo "                 DUAL_GPU_MODE=ollama  (default) ollama_research on GPU 1"
+    echo "                 DUAL_GPU_MODE=vllm             vllm on GPU 1"
+    echo "                 DUAL_GPU_MODE=mixed            both on GPU 1 (VRAM-split)"
     echo ""
     echo "  Examples:"
     echo "    ./manage.sh start"