Merge branch 'master' of 192.168.1.126:TheCrazyInsanity/nixos-configv3

flake.lock

@@ -1,5 +1,85 @@
 {
   "nodes": {
+    "blueprint": {
+      "inputs": {
+        "nixpkgs": [
+          "llm-agents",
+          "nixpkgs"
+        ],
+        "systems": [
+          "llm-agents",
+          "systems"
+        ]
+      },
+      "locked": {
+        "lastModified": 1771437256,
+        "narHash": "sha256-bLqwib+rtyBRRVBWhMuBXPCL/OThfokA+j6+uH7jDGU=",
+        "owner": "numtide",
+        "repo": "blueprint",
+        "rev": "06ee7190dc2620ea98af9eb225aa9627b68b0e33",
+        "type": "github"
+      },
+      "original": {
+        "owner": "numtide",
+        "repo": "blueprint",
+        "type": "github"
+      }
+    },
+    "bun2nix": {
+      "inputs": {
+        "flake-parts": [
+          "llm-agents",
+          "flake-parts"
+        ],
+        "import-tree": "import-tree",
+        "nixpkgs": [
+          "llm-agents",
+          "nixpkgs"
+        ],
+        "systems": [
+          "llm-agents",
+          "systems"
+        ],
+        "treefmt-nix": [
+          "llm-agents",
+          "treefmt-nix"
+        ]
+      },
+      "locked": {
+        "lastModified": 1770895533,
+        "narHash": "sha256-v3QaK9ugy9bN9RXDnjw0i2OifKmz2NnKM82agtqm/UY=",
+        "owner": "nix-community",
+        "repo": "bun2nix",
+        "rev": "c843f477b15f51151f8c6bcc886954699440a6e1",
+        "type": "github"
+      },
+      "original": {
+        "owner": "nix-community",
+        "repo": "bun2nix",
+        "type": "github"
+      }
+    },
+    "flake-parts": {
+      "inputs": {
+        "nixpkgs-lib": [
+          "llm-agents",
+          "nixpkgs"
+        ]
+      },
+      "locked": {
+        "lastModified": 1772408722,
+        "narHash": "sha256-rHuJtdcOjK7rAHpHphUb1iCvgkU3GpfvicLMwwnfMT0=",
+        "owner": "hercules-ci",
+        "repo": "flake-parts",
+        "rev": "f20dc5d9b8027381c474144ecabc9034d6a839a3",
+        "type": "github"
+      },
+      "original": {
+        "owner": "hercules-ci",
+        "repo": "flake-parts",
+        "type": "github"
+      }
+    },
     "home-manager": {
       "inputs": {
         "nixpkgs": [
@@ -7,11 +87,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1772060133,
-        "narHash": "sha256-VuyRptb8v1lVGMlLp4/1vRX3Efwec0CN0S6mKmDPzLg=",
+        "lastModified": 1772845525,
+        "narHash": "sha256-Dp5Ir2u4jJDGCgeMRviHvEQDe+U37hMxp6RSNOoMMPc=",
         "owner": "nix-community",
         "repo": "home-manager",
-        "rev": "ce9b6e52500a0ea0ec48f0bbf6d7a3e431d9dfa4",
+        "rev": "27b93804fbef1544cb07718d3f0a451f4c4cd6c0",
         "type": "github"
       },
       "original": {
@@ -61,6 +141,44 @@
         "type": "github"
       }
     },
+    "import-tree": {
+      "locked": {
+        "lastModified": 1763762820,
+        "narHash": "sha256-ZvYKbFib3AEwiNMLsejb/CWs/OL/srFQ8AogkebEPF0=",
+        "owner": "vic",
+        "repo": "import-tree",
+        "rev": "3c23749d8013ec6daa1d7255057590e9ca726646",
+        "type": "github"
+      },
+      "original": {
+        "owner": "vic",
+        "repo": "import-tree",
+        "type": "github"
+      }
+    },
+    "llm-agents": {
+      "inputs": {
+        "blueprint": "blueprint",
+        "bun2nix": "bun2nix",
+        "flake-parts": "flake-parts",
+        "nixpkgs": "nixpkgs_2",
+        "systems": "systems",
+        "treefmt-nix": "treefmt-nix"
+      },
+      "locked": {
+        "lastModified": 1773322058,
+        "narHash": "sha256-xYQ32BrphBupOi+rTm3XF3URFEmH0kHYVRi58fMER0I=",
+        "owner": "Qumulo",
+        "repo": "llm-agents",
+        "rev": "efdf1f01a8474c3c6c2d0f95a66c97b1baec4dc7",
+        "type": "github"
+      },
+      "original": {
+        "owner": "Qumulo",
+        "repo": "llm-agents",
+        "type": "github"
+      }
+    },
     "nixpkgs": {
       "locked": {
         "lastModified": 1768564909,
@@ -79,11 +197,27 @@
     },
     "nixpkgs_2": {
       "locked": {
-        "lastModified": 1771848320,
-        "narHash": "sha256-0MAd+0mun3K/Ns8JATeHT1sX28faLII5hVLq0L3BdZU=",
+        "lastModified": 1773201692,
+        "narHash": "sha256-NXrKzNMniu4Oam2kAFvqJ3GB2kAvlAFIriTAheaY8hw=",
+        "owner": "NixOS",
+        "repo": "nixpkgs",
+        "rev": "b6067cc0127d4db9c26c79e4de0513e58d0c40c9",
+        "type": "github"
+      },
+      "original": {
+        "owner": "NixOS",
+        "ref": "nixpkgs-unstable",
+        "repo": "nixpkgs",
+        "type": "github"
+      }
+    },
+    "nixpkgs_3": {
+      "locked": {
+        "lastModified": 1772773019,
+        "narHash": "sha256-E1bxHxNKfDoQUuvriG71+f+s/NT0qWkImXsYZNFFfCs=",
         "owner": "nixos",
         "repo": "nixpkgs",
-        "rev": "2fc6539b481e1d2569f25f8799236694180c0993",
+        "rev": "aca4d95fce4914b3892661bcb80b8087293536c6",
         "type": "github"
       },
       "original": {
@@ -120,9 +254,46 @@
       "inputs": {
         "home-manager": "home-manager",
         "impermanence": "impermanence",
-        "nixpkgs": "nixpkgs",
+        "llm-agents": "llm-agents",
+        "nixpkgs": "nixpkgs_3",
         "plasma-manager": "plasma-manager"
       }
+    },
+    "systems": {
+      "locked": {
+        "lastModified": 1681028828,
+        "narHash": "sha256-Vy1rq5AaRuLzOxct8nz4T6wlgyUR7zLU309k9mBC768=",
+        "owner": "nix-systems",
+        "repo": "default",
+        "rev": "da67096a3b9bf56a91d16901293e51ba5b49a27e",
+        "type": "github"
+      },
+      "original": {
+        "owner": "nix-systems",
+        "repo": "default",
+        "type": "github"
+      }
+    },
+    "treefmt-nix": {
+      "inputs": {
+        "nixpkgs": [
+          "llm-agents",
+          "nixpkgs"
+        ]
+      },
+      "locked": {
+        "lastModified": 1773297127,
+        "narHash": "sha256-6E/yhXP7Oy/NbXtf1ktzmU8SdVqJQ09HC/48ebEGBpk=",
+        "owner": "numtide",
+        "repo": "treefmt-nix",
+        "rev": "71b125cd05fbfd78cab3e070b73544abe24c5016",
+        "type": "github"
+      },
+      "original": {
+        "owner": "numtide",
+        "repo": "treefmt-nix",
+        "type": "github"
+      }
     }
   },
   "root": "root",

flake.nix

@@ -17,6 +17,7 @@
       inputs.nixpkgs.follows = "nixpkgs";
       inputs.home-manager.follows = "home-manager";
     };
+    llm-agents.url = "github:Qumulo/llm-agents";
     # nix-flatpak.url = "github:gmodena/nix-flatpak/?ref=v0.6.0";
   };
 
@@ -26,7 +27,6 @@
       nixpkgs,
       home-manager,
       impermanence,
-      plasma-manager,
       # nix-flatpak,
     }@inputs:
     let

global/default.nix

@@ -33,13 +33,13 @@
         blasSupport = true;
       }).overrideAttrs
         (oldAttrs: rec {
-          version = "8184";
+          version = "8209";
           src = pkgs.fetchFromGitHub {
-            owner = "aagit";
+            owner = "ggml-org";
             repo = "llama.cpp";
-            # tag = "b${version}";
-            rev = "6ebf2e0d00d31acfc1a1fa9662e9a7d38bd07bf7"; # https://github.com/ggml-org/llama.cpp/pull/19970
-            hash = "sha256-xryajW0Cs1d+WDijspMTW21FDaZP9Grkb+uErMQCQ48=";
+            tag = "b${version}";
+            # rev = "6ebf2e0d00d31acfc1a1fa9662e9a7d38bd07bf7"; # https://github.com/ggml-org/llama.cpp/pull/19970
+            hash = "sha256-7z9mQZ/hgNS+doLCVPtax+FBhr6dEfmR9wZJTwtl/pM=";
             leaveDotGit = true;
             postFetch = ''
               git -C "$out" rev-parse --short HEAD > $out/COMMIT
@@ -54,7 +54,30 @@
           # for reproducible builds). We sacrifice portability for faster CPU layers.
           cmakeFlags = (oldAttrs.cmakeFlags or [ ]) ++ [
             "-DGGML_NATIVE=ON"
-            "-DCMAKE_CUDA_ARCHITECTURES=86" # RTX 3090 - needed since sandbox has no GPU
+            "-DGGML_LTO=ON" # Link Time Optimization for overall binary speed
+            "-DCMAKE_CUDA_ARCHITECTURES=86" # RTX 3090
+            "-DGGML_CUDA=ON"
+            "-DGGML_CUDA_FA=ON" # FlashAttention kernels (accelerated attention)
+            "-DGGML_CUDA_FA_ALL_QUANTS=ON" # Support for all KV cache quant types in FA
+            "-DGGML_CUDA_GRAPHS=ON" # CUDA Graphs for lower overhead inference
+            "-DGGML_CUDA_FORCE_CUBLAS=ON" # cuBLAS optimized prompt processing for Ampere+
+            "-DGGML_CUDA_PEER_MAX_BATCH_SIZE=256" # Increased for multi-GPU efficiency (split mode)
+            "-DGGML_CUDA_COMPRESSION_MODE=speed" # Fast binary loading (CUDA 12.8+)
+            "-DGGML_OPENMP=ON" # Optimal multi-threading on CPU
+            "-DGGML_LLAMAFILE=ON" # Use llamafile sgemm for faster CPU layers
+            "-DGGML_CPU_REPACK=ON" # Optimize Q4_0 quant handling
+            "-DGGML_AVX=ON"
+            "-DGGML_AVX2=ON"
+            "-DGGML_FMA=ON"
+            "-DGGML_F16C=ON"
+            "-DGGML_AVX512=ON" # Intel AVX-512 extensions
+            "-DGGML_AVX512_VNNI=ON" # Vector Neural Network Instructions
+            "-DGGML_AVX512_BF16=ON" # Bfloat16 support
+            "-DGGML_AVX_VNNI=ON" # VNNI for processors without AVX-512
+            "-DGGML_AMX_TILE=ON" # Intel Advanced Matrix Extensions (Sapphire Rapids+)
+            "-DGGML_AMX_INT8=ON"
+            "-DGGML_AMX_BF16=ON"
+            "-DGGML_BLAS=ON" # Uses internal BLAS provided by Nix (blasSupport=true works)
           ];
 
           # Disable Nix's NIX_ENFORCE_NO_NATIVE which strips -march=native flags
@@ -178,7 +201,7 @@
     qpwgraph
     libimobiledevice
     ifuse
-    neofetch
+    fastfetch
     gimp # Despite the fact it falls under creative an image editor is too important to leave out.
     zip
     xz

machines/homepc/configuration.nix

@@ -6,6 +6,7 @@
   config,
   lib,
   pkgs,
+  inputs,
   ...
 }:
 {
@@ -42,11 +43,14 @@
   # TODO: honestly, while I currently only have one pc that can run local AI, might change in the future.
   # And this config is getting a bit complicated for a single pc config
   # Should be moved to it's own shit
-  environment.systemPackages = with pkgs; [
-    ollama-cuda
-    opencode
-    llama-cpp
-    llama-swap
+  environment.systemPackages = [
+    pkgs.ollama-cuda
+    pkgs.opencode
+    pkgs.llama-cpp
+    pkgs.llama-swap
+
+    inputs.llm-agents.packages.${pkgs.stdenv.hostPlatform.system}.pi
+    inputs.llm-agents.packages.${pkgs.stdenv.hostPlatform.system}.rtk
   ];
 
   services.ollama = {
@@ -87,32 +91,35 @@
   environment.etc."llama-swap/config.yaml".text = ''
     models:
       "Qwen3.5-35B-A3B-GGUF":
-        cmd: llama-server --port ''${PORT} -hf unsloth/Qwen3.5-35B-A3B-GGUF:Q4_K_S --ctx-size 128000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -np 1  --cpu-moe --fit-target 256 --ubatch-size 1024 -fa on  --slots --slot-save-path /home/laythe/llamapcache  --jinja -kvu --no-mmproj --swa-checkpoints 32
+        cmd: llama-server --port ''${PORT} -hf unsloth/Qwen3.5-35B-A3B-GGUF:UD-IQ4_XS --ctx-size 128000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -np 1 --fit-target 256 --ubatch-size 1024 -fa on  --slots --slot-save-path /home/laythe/llamapcache --jinja -kvu --no-mmproj --swa-checkpoints 32 --no-kv-offload
         ttl: 2400
       "Qwen3-1.7B-GGUF":
         cmd: llama-server --port ''${PORT} -hf unsloth/Qwen3-1.7B-GGUF:Q4_K_S --ctx-size 32768 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu
-        ttl: 120
+        ttl: 300
       "Qwen3-8B-GGUF":
         cmd: llama-server --port ''${PORT} -hf unsloth/Qwen3-8B-GGUF:Q4_K_S --ctx-size 32768 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1
-        ttl: 120
+        ttl: 300
       "Qwen3-4B-Claude-Opus-Distill":
         cmd: llama-server --port ''${PORT} -hf TeichAI/Qwen3-4B-Thinking-2507-Claude-4.5-Opus-High-Reasoning-Distill-GGUF:Q4_K_M --ctx-size 32768 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1 --fit-target 256
-        ttl: 120
+        ttl: 300
       "Qwen3.5-9B-Thinking":
-        cmd: "llama-server --port ''${PORT} -hf unsloth/Qwen3.5-9B-GGUF:Q4_K_M --ctx-size 32000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1 --fit-target 256 --chat-template-kwargs '{\"enable_thinking\": true}' --no-mmproj"
-        ttl: 120
+        cmd: "llama-server --port ''${PORT} -hf unsloth/Qwen3.5-9B-GGUF:Q4_K_M --ctx-size 128000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1 --fit-target 256 --chat-template-kwargs '{\"enable_thinking\": true}' --no-mmproj --no-kv-offload"
+        ttl: 300
+      "Qwen3.5-9B-Claude-Opus-Distill":
+        cmd: "llama-server --port ''${PORT} -hf Jackrong/Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled-GGUF:Q4_K_S --ctx-size 32000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1 --fit-target 512 --chat-template-kwargs '{\"enable_thinking\": true}' --no-mmproj"
+        ttl: 300
       "Qwen3.5-4B-Thinking":
         cmd: "llama-server --port ''${PORT} -hf unsloth/Qwen3.5-4B-GGUF:Q4_K_M --ctx-size 64000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1 --fit-target 256 --chat-template-kwargs '{\"enable_thinking\": true}'"
-        ttl: 120
+        ttl: 300
       "Qwen3.5-9B-Non-Thinking":
         cmd: "llama-server --port ''${PORT} -hf unsloth/Qwen3.5-9B-GGUF:Q4_K_M --ctx-size 32000 --temp 0.7 --top-p 0.8 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1 --fit-target 256 --no-mmproj"
-        ttl: 120
+        ttl: 300
       "Qwen3.5-4B-Non-Thinking":
         cmd: "llama-server --port ''${PORT} -hf unsloth/Qwen3.5-4B-GGUF:Q4_K_M --ctx-size 64000 --temp 0.7 --top-p 0.8 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1 --fit-target 256"
-        ttl: 120
+        ttl: 300
       "Qwen3.5-0.8B-Non-Thinking":
         cmd: "llama-server --port ''${PORT} -hf unsloth/Qwen3.5-0.8B-GGUF:Q4_K_M --ctx-size 64000 --temp 0.7 --top-p 0.8 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1 --fit-target 256"
-        ttl: 120'';
+        ttl: 300'';
 
   # Set your time zone.
   # time.timeZone = "Europe/Amsterdam";

modules/gaming/default.nix

@@ -5,6 +5,7 @@
     remotePlay.openFirewall = true; # Open ports in the firewall for Steam Remote Play
     dedicatedServer.openFirewall = true; # Open ports in the firewall for Source Dedicated Server
     localNetworkGameTransfers.openFirewall = true; # Open ports in the firewall for Steam Local Network Game Transfers
+    protontricks.enable = true;
   };
 
   environment.systemPackages = with pkgs; [