sync with latest upstream llama.cpp (v3.5.0)

.gitignore

@@ -2,6 +2,7 @@ node_modules
 .DS_Store
 .vscode
 .claude
+.npmrc
 /cache
 /build
 /build-compat

.npmignore

@@ -1,3 +1,4 @@
+.npmrc
 /docs
 /wasm
 /examples

CMakeLists.txt

@@ -3,6 +3,8 @@ project("wllama")
 
 option(WLLAMA_TEST_BACKEND "Build wllama with test-backend-ops included" OFF)
 
+set(MTMD_VIDEO OFF CACHE BOOL "" FORCE)
+
 set(CMAKE_THREAD_LIBS_INIT "-lpthread")
 set(CMAKE_HAVE_THREADS_LIBRARY 1)
 set(CMAKE_USE_WIN32_THREADS_INIT 0)
@@ -23,6 +25,7 @@ set(WLLAMA_LINK_OPTIONS
     -sMODULARIZE=0
     -sINITIAL_MEMORY=128MB
     -sMAXIMUM_MEMORY=4096MB
+    -sSTACK_SIZE=5MB
     -sALLOW_MEMORY_GROWTH=1
     -sFORCE_FILESYSTEM=1
     -sEXPORTED_FUNCTIONS=_main,_wllama_malloc,_wllama_start,_wllama_action,_wllama_exit,_wllama_debug

compat/.npmignore

@@ -0,0 +1,15 @@
+.npmrc
+/docs
+/wasm
+/examples
+/llama.cpp
+/models
+/build
+/build-*
+/cache
+/compat
+/.github
+/.vscode
+/.claude
+a.out.js
+a.out.wasm

compat/package.json

@@ -1,12 +1,11 @@
 {
   "name": "@wllama/wllama-compat",
-  "version": "3.4.1",
+  "version": "3.5.0",
   "description": "Optional package providing compatibility with older browsers for @wllama/wllama",
   "main": "index.js",
   "type": "module",
   "scripts": {
-    "upload": "npm publish --access public",
-    "prepublishOnly": "node ../scripts/check_package_size.js"
+    "upload": "npm publish --access public"
   },
   "repository": {
     "type": "git",

cpp/wllama-context.h

@@ -1143,6 +1143,20 @@ std::pair<long, std::vector<char>> common_remote_get_content(const std::string &
   throw std::runtime_error("common_remote_get_content is not implemented in wllama");
 }
 
+#if 1
+common_device_memory_data_vec common_get_device_memory_data(
+    const char *path_model,
+    const llama_model_params *mparams,
+    const llama_context_params *cparams,
+    std::vector<ggml_backend_dev_t> &devs,
+    uint32_t &hp_ngl,
+    uint32_t &hp_n_ctx_train,
+    uint32_t &hp_n_expert,
+    ggml_log_level log_level)
+{
+  throw std::runtime_error("common_get_device_memory_data is not implemented in wllama");
+}
+#else
 std::vector<llama_device_memory_data> common_get_device_memory_data(
     const char *path_model,
     const struct llama_model_params *mparams,
@@ -1155,3 +1169,4 @@ std::vector<llama_device_memory_data> common_get_device_memory_data(
 {
   throw std::runtime_error("common_get_device_memory_data is not implemented in wllama");
 }
+#endif

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@wllama/wllama",
-  "version": "3.4.1",
+  "version": "3.5.0",
   "description": "WebAssembly binding for llama.cpp - Enabling on-browser LLM inference",
   "main": "index.js",
   "type": "module",
@@ -18,16 +18,15 @@
     "build:tsup": "tsup src/index.ts --format cjs,esm --clean",
     "build:minified": "terser esm/index.js -o esm/index.min.js --compress --mangle --source-map",
     "build:typedef": "tsc --emitDeclarationOnly --declaration -p tsconfig.build.json",
-    "build": "npm run clean && npm run build:worker && npm run build:tsup && npm run build:minified && npm run build:typedef",
-    "postbuild": "./scripts/post_build.sh && npm run docs",
+    "build": "npm run clean && npm run build:worker && npm run build:tsup && npm run build:minified && npm run build:typedef && ./scripts/post_build.sh && npm run docs",
     "docs": "typedoc --tsconfig tsconfig.build.json src/index.ts",
-    "upload": "npm run format && npm run build && npm publish --access public && (cd compat && npm publish --access public)",
+    "upload": "npm run format && npm run build && node scripts/check_package_size.js && npm publish --access public && (cd compat && npm publish --access public)",
     "format": "prettier --write .",
     "test": "vitest",
+    "test:auto": "AUTO=1 vitest",
     "test:firefox": "BROWSER=firefox vitest",
     "test:safari": "BROWSER=safari vitest",
-    "test:wgpu": "WEBGPU=1 vitest",
-    "prepublishOnly": "node scripts/check_package_size.js"
+    "test:wgpu": "WEBGPU=1 vitest"
   },
   "repository": {
     "type": "git",

scripts/bisect_test.sh

@@ -0,0 +1,33 @@
+#!/bin/bash
+
+# Git bisect script for finding the llama.cpp commit that introduced the locale crash.
+# Run from wllama root: git bisect run ./scripts/bisect_test.sh
+#
+# Start (good): aa46bda89b9a8378ae76bb15fc2ce2f571f0983c  (wllama master's llama.cpp)
+# End   (bad):  dd4623a74                                  (current HEAD of submodule)
+
+set -e
+
+WLLAMA_ROOT="$( cd "$( dirname "${BASH_SOURCE[0]}" )/.." &> /dev/null && pwd )"
+cd "$WLLAMA_ROOT"
+
+CONTEXT_H="$WLLAMA_ROOT/cpp/wllama-context.h"
+FIT_H="$WLLAMA_ROOT/llama.cpp/common/fit.h"
+
+# Adjust #if 0 / #if 1 around common_get_device_memory_data stub based on
+# which signature fit.h declares at the current bisect commit.
+if grep -q "common_device_memory_data_vec" "$FIT_H" 2>/dev/null; then
+    # New signature (after d8a24ccee): enable the common_device_memory_data_vec block
+    sed -i.bak 's/^#if [01]$/\#if 1/' "$CONTEXT_H"
+else
+    # Old signature (before d8a24ccee): enable the std::vector<llama_device_memory_data> block
+    sed -i.bak 's/^#if [01]$/\#if 0/' "$CONTEXT_H"
+fi
+rm -f "${CONTEXT_H}.bak"
+
+# Build and run tests. Exit 125 = skip (build infra broken at this commit).
+rm -rf build
+
+SKIP_COMPAT=1 npm run build:wasm 2>&1 || exit 125
+npm run build 2>&1 || exit 125
+AUTO=1 npm run test 2>&1

scripts/docker-compose.yml

@@ -12,6 +12,7 @@ services:
     user: "${D_UID}:${D_GID}"
     environment:
       - WLLAMA_TEST_BACKEND=${WLLAMA_TEST_BACKEND:-}
+      - SKIP_COMPAT=${SKIP_COMPAT:-}
     volumes:
       - ..:/source:Z
     entrypoint: /bin/bash
@@ -49,23 +50,27 @@ services:
         cp build/wllama.js   src/wasm
         cp build/wllama.wasm src/wasm
 
-        # build compat package
-        mkdir -p build-compat
-        cd build-compat
-        emcmake cmake .. -DWLLAMA_COMPAT=ON -DLLAMA_WASM_MEM64=OFF -DGGML_WEBGPU=ON -DGGML_WEBGPU_JSPI=OFF -DEMDAWNWEBGPU_DIR="$${EMDAWNWEBGPU_DIR}" $${CMAKE_EXTRA_FLAGS}
-        emmake make wllama -j
+        # build compat package (skip if SKIP_COMPAT=1)
+        if [ -z "${SKIP_COMPAT}" ]; then
+          mkdir -p build-compat
+          cd build-compat
+          emcmake cmake .. -DWLLAMA_COMPAT=ON -DLLAMA_WASM_MEM64=OFF -DGGML_WEBGPU=ON -DGGML_WEBGPU_JSPI=OFF -DEMDAWNWEBGPU_DIR="$${EMDAWNWEBGPU_DIR}" $${CMAKE_EXTRA_FLAGS}
+          emmake make wllama -j
 
-        # go back to root
-        cd ..
+          # go back to root
+          cd ..
 
-        # copy output files (compat build)
-        mkdir -p compat/wasm
-        cp build-compat/wllama.js   compat/wasm
-        cp build-compat/wllama.wasm compat/wasm
+          # copy output files (compat build)
+          mkdir -p compat/wasm
+          cp build-compat/wllama.js   compat/wasm
+          cp build-compat/wllama.wasm compat/wasm
+        fi
 
         # generate source map
         node scripts/build_source_map.js
 
         # list files (optional, for verification)
         ls -lh build
-        ls -lh build-compat
+        if [ -d build-compat ]; then
+          ls -lh build-compat
+        fi

src/wasm-from-cdn.ts

@@ -2,12 +2,12 @@
 // Do not edit this file directly
 
 const WasmFromCDN = {
-  default: 'https://cdn.jsdelivr.net/npm/@wllama/wllama@3.4.1/src/wasm/wllama.wasm',
+  default: 'https://cdn.jsdelivr.net/npm/@wllama/wllama@3.5.0/src/wasm/wllama.wasm',
 };
 
 export const WasmCompatFromCDN = {
-  worker: 'https://cdn.jsdelivr.net/npm/@wllama/wllama-compat@3.4.1/wasm/wllama.js',
-  wasm: 'https://cdn.jsdelivr.net/npm/@wllama/wllama-compat@3.4.1/wasm/wllama.wasm',
+  worker: 'https://cdn.jsdelivr.net/npm/@wllama/wllama-compat@3.5.0/wasm/wllama.js',
+  wasm: 'https://cdn.jsdelivr.net/npm/@wllama/wllama-compat@3.5.0/wasm/wllama.wasm',
 };
 
 export default WasmFromCDN;