Merge pull request #23 from outis1one/claude/gpu-setup-research-c78YT

Claude/gpu setup research c78 yt

laptop_full_setup.sh

@@ -66,6 +66,9 @@ elif [[ "$TOTAL_VRAM" -ge 14 ]]; then
 elif [[ "$TOTAL_VRAM" -ge 8 ]]; then
     CHAT_MODEL="qwen3.5:9b";  CODE_MODEL="qwen3.5:9b"
     CTX=32768;  GPU_TIER="${TOTAL_VRAM}GB VRAM — 9B dense, 32K context"
+elif [[ "$TOTAL_VRAM" -ge 6 ]]; then
+    CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
+    CTX=32768;  GPU_TIER="${TOTAL_VRAM}GB VRAM — 4B + KV quant, 32K context (3.5GB free for cache)"
 elif [[ "$TOTAL_VRAM" -ge 4 ]]; then
     CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
     CTX=16384;  GPU_TIER="${TOTAL_VRAM}GB VRAM — 4B models, 16K context"
@@ -463,23 +466,32 @@ if $INSTALL_AI; then
     echo ""
     echo "  Origin preference:"
     echo "  1) Western-only      — Codestral (Mistral 🇫🇷) · Phi4 (Microsoft 🇺🇸) · Mistral 7B"
-    echo "  2) Performance-first — Qwen2.5 · Qwen2.5-Coder (Chinese, top benchmarks)"
-    echo "  3) Mixed             — Western for chat/reasoning, Qwen for coding only"
+    echo "  2) Performance-first — Qwen 3.5 (Feb 2026, top benchmarks, vision+code)"
+    echo "  3) Mixed             — Western for chat, Qwen 3.5 for coding"
     echo "  4) Custom            — enter model names manually"
     echo ""
-    read -rp "  Choice [1]: " MODEL_PREF
-    MODEL_PREF="${MODEL_PREF:-1}"
+    read -rp "  Choice [2]: " MODEL_PREF
+    MODEL_PREF="${MODEL_PREF:-2}"
 
-    # Q4 approximate sizes in GB: 7B=4, 13-15B=9, 22B=13, 32B=19, 70-72B=41
+    # Q4_K_M approximate weights in GB: 4B=2.5, 7B=4, 9B=5.5, 14B=9, 22B=13, 27B=17, 35B-MoE=12, 70B=41
 
     # VRAM-based recommendation (soft — shown as suggestion only)
-    if   [[ "$VRAM_GB" -ge 20 ]]; then REC_TIER="32B"
-    elif [[ "$VRAM_GB" -ge 12 ]]; then REC_TIER="22B"
-    elif [[ "$VRAM_GB" -ge  6 ]]; then REC_TIER="14B"
-    else                                REC_TIER="7B"
-    fi
-    # performance pref has no 22B tier
-    [[ "$MODEL_PREF" == "2" && "$REC_TIER" == "22B" ]] && REC_TIER="32B"
+    case "$MODEL_PREF" in
+        2)  # Performance (Qwen 3.5)
+            if   [[ "$TOTAL_VRAM" -ge 28 ]]; then REC_TIER="27B"
+            elif [[ "$TOTAL_VRAM" -ge 14 ]]; then REC_TIER="35B"
+            elif [[ "$TOTAL_VRAM" -ge  8 ]]; then REC_TIER="9B"
+            else                                   REC_TIER="4B"
+            fi
+            ;;
+        *)  # Western / Mixed
+            if   [[ "$TOTAL_VRAM" -ge 40 ]]; then REC_TIER="70B"
+            elif [[ "$TOTAL_VRAM" -ge 12 ]]; then REC_TIER="35B"
+            elif [[ "$TOTAL_VRAM" -ge  8 ]]; then REC_TIER="14B"
+            else                                   REC_TIER="7B"
+            fi
+            ;;
+    esac
 
     echo ""
     echo "  Size tier — estimated speed on your ${VRAM_GB}GB GPU:"
@@ -1391,6 +1403,14 @@ if $INSTALL_AI; then
         echo "    Via MCP:      gitea_github_sync(mode='all')"
     }
 fi
+if $INSTALL_AI; then
+    echo ""
+    echo -e "  ${YELLOW}Recommended Open WebUI Functions (install from Admin → Functions → ＋):${NC}"
+    echo "    Context tracker:    https://openwebui.com/f/centrisic/context_tracker"
+    echo "      → Shows tokens used vs available, progress bar, context % remaining"
+    echo "    Context compaction: https://openwebui.com/f/projectmoon/checkpoint_summarization_filter"
+    echo "      → Auto-summarizes old messages when context fills up (like Claude)"
+fi
 if $SVC_KIWIX && [[ "$ZIM_CHOICE" == "3" ]]; then
     echo -e "  ${YELLOW}ZIM downloads:${NC}      ./kiwix_download.sh  (not started)"
 elif $SVC_KIWIX; then

local-ai-setup.sh

@@ -42,6 +42,9 @@ elif [[ "$TOTAL_VRAM" -ge 14 ]]; then
 elif [[ "$TOTAL_VRAM" -ge 8 ]]; then
     CHAT_MODEL="qwen3.5:9b";  CODE_MODEL="qwen3.5:9b"
     CTX=32768;  TIER="${TOTAL_VRAM}GB — 9B dense, 32K context"
+elif [[ "$TOTAL_VRAM" -ge 6 ]]; then
+    CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
+    CTX=32768;  TIER="${TOTAL_VRAM}GB — 4B + KV quant, 32K context"
 elif [[ "$TOTAL_VRAM" -ge 4 ]]; then
     CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
     CTX=16384;  TIER="${TOTAL_VRAM}GB — 4B models, 16K context"
@@ -898,7 +901,13 @@ echo ""
 echo -e "  ${YELLOW}Aider CLI:${NC}  cd your-repo && $BASE/aider.sh <files>"
 echo    "  (or open browser UI above — both use your local code model)"
 echo ""
-echo -e "  ${YELLOW}Save Claude usage:${NC} use local 14B for boilerplate, docs,"
-echo    "  simple fixes. Use Claude Sonnet 4.6 for hard bugs,"
+echo -e "  ${YELLOW}Recommended Open WebUI Functions${NC} (install from Admin → Functions → ＋):"
+echo    "    Context tracker:    https://openwebui.com/f/centrisic/context_tracker"
+echo    "      → Shows tokens used vs available, progress bar, context % remaining"
+echo    "    Context compaction: https://openwebui.com/f/projectmoon/checkpoint_summarization_filter"
+echo    "      → Auto-summarizes old messages when context fills up (like Claude)"
+echo ""
+echo -e "  ${YELLOW}Save Claude usage:${NC} use local models for boilerplate, docs,"
+echo    "  simple fixes. Use Claude for hard bugs,"
 echo    "  multi-file refactoring, architecture decisions."
 echo ""

ubuntu-post-install.sh

@@ -1125,6 +1125,9 @@ elif [[ "$TOTAL_VRAM" -ge 14 ]]; then
 elif [[ "$TOTAL_VRAM" -ge 8 ]]; then
     CHAT_MODEL="qwen3.5:9b";  CODE_MODEL="qwen3.5:9b"
     CTX=32768;  TIER="${TOTAL_VRAM}GB — 9B dense, 32K context"
+elif [[ "$TOTAL_VRAM" -ge 6 ]]; then
+    CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
+    CTX=32768;  TIER="${TOTAL_VRAM}GB — 4B + KV quant, 32K context"
 elif [[ "$TOTAL_VRAM" -ge 4 ]]; then
     CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
     CTX=16384;  TIER="${TOTAL_VRAM}GB — 4B models, 16K context"