toy 0.8.0

data/vendor/ggml/include/gguf.h

@@ -0,0 +1,204 @@
+// This file contains functionality related to "GGUF" files, the binary file format used by ggml.
+// GGUF files have the following structure:
+//
+// 1. File magic "GGUF" (4 bytes).
+// 2. File version (uint32_t).
+// 3. Number of ggml tensors in file (int64_t).
+// 4. Number of key-value-pairs in file (int64_t).
+// 5. For each KV pair:
+//   1. The key (string).
+//   2. The value type (gguf_type).
+//   3a. If the value type is GGUF_TYPE_ARRAY:
+//     1. The type of the array (gguf_type).
+//     2. The number of elements in the array (uint64_t).
+//     3. The binary representation of each element in the array.
+//   3b. Otherwise:
+//     1. The binary representation of the value.
+// 6. For each ggml tensor:
+//   1. The tensor name (string).
+//   2. The number of dimensions of the tensor (uint32_t).
+//   3. For each dimension:
+//     1. The size of the tensor in the dimension (int64_t).
+//   4. The tensor data type (ggml_type).
+//   5. The tensor data offset in the tensor data binary blob (uint64_t).
+// 7. The tensor data binary blob (optional, aligned).
+//
+// Strings are serialized as the string length (uint64_t) followed by the C string without the null terminator.
+// All enums are stored as int32_t.
+// All bool values are stored as int8_t.
+// If the special key "general.alignment" (uint32_t) is defined it is used for alignment,
+//   otherwise GGUF_DEFAULT_ALIGNMENT is used.
+//
+// Module maintainer: Johannes Gäßler (@JohannesGaessler, johannesg@5d6.de)
+
+#pragma once
+
+#include "ggml.h"
+
+#include <stdbool.h>
+#include <stdint.h>
+
+#define GGUF_MAGIC   "GGUF"
+#define GGUF_VERSION 3
+
+#define GGUF_KEY_GENERAL_ALIGNMENT "general.alignment"
+
+#define GGUF_DEFAULT_ALIGNMENT 32
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+    // types that can be stored as GGUF KV data
+    enum gguf_type {
+        GGUF_TYPE_UINT8   = 0,
+        GGUF_TYPE_INT8    = 1,
+        GGUF_TYPE_UINT16  = 2,
+        GGUF_TYPE_INT16   = 3,
+        GGUF_TYPE_UINT32  = 4,
+        GGUF_TYPE_INT32   = 5,
+        GGUF_TYPE_FLOAT32 = 6,
+        GGUF_TYPE_BOOL    = 7,
+        GGUF_TYPE_STRING  = 8,
+        GGUF_TYPE_ARRAY   = 9,
+        GGUF_TYPE_UINT64  = 10,
+        GGUF_TYPE_INT64   = 11,
+        GGUF_TYPE_FLOAT64 = 12,
+        GGUF_TYPE_COUNT,       // marks the end of the enum
+    };
+
+    struct gguf_context;
+
+    struct gguf_init_params {
+        bool no_alloc;
+
+        // if not NULL, create a ggml_context and allocate the tensor data in it
+        struct ggml_context ** ctx;
+    };
+
+    GGML_API struct gguf_context * gguf_init_empty(void);
+    GGML_API struct gguf_context * gguf_init_from_file_ptr(FILE * file, struct gguf_init_params params);
+    GGML_API struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params);
+    //GGML_API struct gguf_context * gguf_init_from_buffer(..);
+
+    GGML_API void gguf_free(struct gguf_context * ctx);
+
+    GGML_API const char * gguf_type_name(enum gguf_type type);
+
+    GGML_API uint32_t gguf_get_version    (const struct gguf_context * ctx);
+    GGML_API size_t   gguf_get_alignment  (const struct gguf_context * ctx);
+    GGML_API size_t   gguf_get_data_offset(const struct gguf_context * ctx);
+
+    GGML_API int64_t      gguf_get_n_kv(const struct gguf_context * ctx);
+    GGML_API int64_t      gguf_find_key(const struct gguf_context * ctx, const char * key); // returns -1 if key is not found
+    GGML_API const char * gguf_get_key (const struct gguf_context * ctx, int64_t key_id);
+
+    GGML_API enum gguf_type gguf_get_kv_type (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API enum gguf_type gguf_get_arr_type(const struct gguf_context * ctx, int64_t key_id);
+
+    // will abort if the wrong type is used for the key
+    GGML_API uint8_t      gguf_get_val_u8  (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API int8_t       gguf_get_val_i8  (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API uint16_t     gguf_get_val_u16 (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API int16_t      gguf_get_val_i16 (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API uint32_t     gguf_get_val_u32 (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API int32_t      gguf_get_val_i32 (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API float        gguf_get_val_f32 (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API uint64_t     gguf_get_val_u64 (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API int64_t      gguf_get_val_i64 (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API double       gguf_get_val_f64 (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API bool         gguf_get_val_bool(const struct gguf_context * ctx, int64_t key_id);
+    GGML_API const char * gguf_get_val_str (const struct gguf_context * ctx, int64_t key_id);
+    GGML_API const void * gguf_get_val_data(const struct gguf_context * ctx, int64_t key_id);
+    GGML_API size_t       gguf_get_arr_n   (const struct gguf_context * ctx, int64_t key_id);
+
+    // get raw pointer to the first element of the array with the given key_id
+    // for bool arrays, note that they are always stored as int8 on all platforms (usually this makes no difference)
+    GGML_API const void * gguf_get_arr_data(const struct gguf_context * ctx, int64_t key_id);
+
+    // get ith C string from array with given key_id
+    GGML_API const char * gguf_get_arr_str (const struct gguf_context * ctx, int64_t key_id, size_t i);
+
+    GGML_API int64_t        gguf_get_n_tensors    (const struct gguf_context * ctx);
+    GGML_API int64_t        gguf_find_tensor      (const struct gguf_context * ctx, const char * name); // returns -1 if the tensor is not found
+    GGML_API size_t         gguf_get_tensor_offset(const struct gguf_context * ctx, int64_t tensor_id);
+    GGML_API const char *   gguf_get_tensor_name  (const struct gguf_context * ctx, int64_t tensor_id);
+    GGML_API enum ggml_type gguf_get_tensor_type  (const struct gguf_context * ctx, int64_t tensor_id);
+    GGML_API size_t         gguf_get_tensor_size  (const struct gguf_context * ctx, int64_t tensor_id);
+
+    // removes key if it exists, returns id that the key had prior to removal (-1 if it didn't exist)
+    GGML_API int64_t gguf_remove_key(struct gguf_context * ctx, const char * key);
+
+    // overrides an existing KV pair or adds a new one, the new KV pair is always at the back
+    GGML_API void gguf_set_val_u8  (struct gguf_context * ctx, const char * key, uint8_t      val);
+    GGML_API void gguf_set_val_i8  (struct gguf_context * ctx, const char * key, int8_t       val);
+    GGML_API void gguf_set_val_u16 (struct gguf_context * ctx, const char * key, uint16_t     val);
+    GGML_API void gguf_set_val_i16 (struct gguf_context * ctx, const char * key, int16_t      val);
+    GGML_API void gguf_set_val_u32 (struct gguf_context * ctx, const char * key, uint32_t     val);
+    GGML_API void gguf_set_val_i32 (struct gguf_context * ctx, const char * key, int32_t      val);
+    GGML_API void gguf_set_val_f32 (struct gguf_context * ctx, const char * key, float        val);
+    GGML_API void gguf_set_val_u64 (struct gguf_context * ctx, const char * key, uint64_t     val);
+    GGML_API void gguf_set_val_i64 (struct gguf_context * ctx, const char * key, int64_t      val);
+    GGML_API void gguf_set_val_f64 (struct gguf_context * ctx, const char * key, double       val);
+    GGML_API void gguf_set_val_bool(struct gguf_context * ctx, const char * key, bool         val);
+    GGML_API void gguf_set_val_str (struct gguf_context * ctx, const char * key, const char * val);
+
+    // creates a new array with n elements of the given type and copies the corresponding number of bytes from data
+    GGML_API void gguf_set_arr_data(struct gguf_context * ctx, const char * key, enum gguf_type type, const void * data, size_t n);
+
+    // creates a new array with n strings and copies the corresponding strings from data
+    GGML_API void gguf_set_arr_str (struct gguf_context * ctx, const char * key, const char ** data, size_t n);
+
+    // set or add KV pairs from another context
+    GGML_API void gguf_set_kv(struct gguf_context * ctx, const struct gguf_context * src);
+
+    // add tensor to GGUF context, tensor name must be unique
+    GGML_API void gguf_add_tensor(struct gguf_context * ctx, const struct ggml_tensor * tensor);
+
+    // after changing a tensor's type, the offsets of all tensors with higher indices are immediately recalculated
+    //   in such a way that the tensor data remains as one contiguous block (except for padding)
+    GGML_API void gguf_set_tensor_type(struct gguf_context * ctx, const char * name, enum ggml_type type);
+
+    // assumes that at least gguf_get_tensor_size bytes can be read from data
+    GGML_API void gguf_set_tensor_data(struct gguf_context * ctx, const char * name, const void * data);
+
+    // writing gguf files can be done in 3 ways:
+    //
+    // - write the entire gguf_context to a binary file in a single pass:
+    //
+    //   gguf_write_to_file(ctx, fname, /*only_meta =*/ false);
+    //
+    // - write only the meta data to a file, then re-open the file and append the tensor data:
+    //
+    //   gguf_write_to_file(ctx, fname, /*only_meta =*/ true);
+    //   FILE * f = fopen(fname, "ab");
+    //   fwrite(f, ...); // write tensor data
+    //   fclose(f);
+    //
+    // - first prepare a file with a placeholder for the meta data, write the tensor data, then write the meta data:
+    //
+    //   FILE * f = fopen(fname, "wb");
+    //   const size_t size_meta = gguf_get_meta_size(ctx);
+    //   fseek(f, size_meta, SEEK_SET);
+    //   fwrite(f, ...); // write tensor data
+    //   void * data = malloc(size_meta);
+    //   gguf_get_meta_data(ctx, data);
+    //   rewind(f);
+    //   fwrite(data, 1, data, f);
+    //   free(data);
+    //   fclose(f);
+    //
+
+    // write the entire context to a binary file
+    GGML_API bool gguf_write_to_file_ptr(const struct gguf_context * ctx, FILE * file, bool only_meta);
+    GGML_API bool gguf_write_to_file(const struct gguf_context * ctx, const char * fname, bool only_meta);
+
+    // get the size in bytes of the meta data (header, kv pairs, tensor info) including padding
+    GGML_API size_t gguf_get_meta_size(const struct gguf_context * ctx);
+
+    // writes the meta data to pointer "data"
+    GGML_API void   gguf_get_meta_data(const struct gguf_context * ctx, void * data);
+
+#ifdef  __cplusplus
+}
+#endif

data/vendor/ggml/requirements.txt

@@ -0,0 +1,12 @@
+accelerate==0.19.0
+numpy~=1.26.4; python_version < "3.13"
+numpy~=2.1.0; python_version >= "3.13"
+sentencepiece>=0.1.98,<0.3.0
+torchvision~=0.21.0
+transformers==5.5.1
+gguf>=0.1.0
+keras==3.10.0
+tensorflow==2.20.0
+
+--extra-index-url https://download.pytorch.org/whl/cpu
+torch~=2.6.0

data/vendor/ggml/scripts/gen-authors.sh

@@ -0,0 +1,9 @@
+#!/usr/bin/env bash
+
+printf "# date: $(date)\n" > AUTHORS
+printf "# this file is auto-generated by scripts/gen-authors.sh\n\n" >> AUTHORS
+
+git log --format='%an <%ae>' --reverse --date=short master | awk '!seen[$0]++' | sort >> AUTHORS
+
+# if necessary, update your name here. for example: jdoe -> John Doe
+sed -i '' 's/^jdoe/John Doe/g' AUTHORS

data/vendor/ggml/scripts/release.sh

@@ -0,0 +1,296 @@
+#!/bin/bash
+#
+# Automated release script for ggml.
+#
+# Note: Sync from llama.cpp should be done separately via PR process
+# prior to running this script.
+#
+# Usage:
+#   ./scripts/release.sh prepare [major|minor|patch] [--dry-run]
+#   ./scripts/release.sh finalize [--dry-run]
+#
+# Two-stage release process:
+#
+# Stage 1 - Prepare:
+# $ ./scripts/release.sh prepare minor
+# This creates a release candidate branch with version bump and removes -dev suffix.
+# The branch should then be manually pushed and a PR created, reviewed, and merged.
+#
+# Stage 2 - Finalize:
+# $ ./scripts/release.sh finalize
+# After the RC PR is merged, this reads the current version from CMakeLists.txt,
+# creates the release tag, and prepares the next development cycle.
+#
+# Prepare stage:
+# 1. Creates release candidate branch
+# 2. Updates version and removes -dev suffix
+# 3. Commits the version bump
+#
+# Finalize stage:
+# 1. Reads current release version from CMakeLists.txt
+# 2. Creates signed git tag on master
+# 3. Adds -dev suffix back for next development cycle
+# 4. Creates branch and commit for development version
+#
+
+set -e
+
+if [ ! -f "CMakeLists.txt" ] || [ ! -d "scripts" ]; then
+    echo "Error: Must be run from ggml root directory"
+    exit 1
+fi
+
+# Parse command line arguments
+COMMAND=""
+VERSION_TYPE=""
+DRY_RUN=false
+
+# First argument should be the command
+if [ $# -eq 0 ]; then
+    echo "Error: Missing command"
+    echo "Usage: $0 prepare [major|minor|patch] [--dry-run]"
+    echo "       $0 finalize [--dry-run]"
+    exit 1
+fi
+
+COMMAND="$1"
+shift
+
+# Parse remaining arguments
+for arg in "$@"; do
+    case $arg in
+        --dry-run)
+            DRY_RUN=true
+            ;;
+        major|minor|patch)
+            if [ "$COMMAND" = "prepare" ]; then
+                VERSION_TYPE="$arg"
+            else
+                echo "Error: Version type only valid for 'prepare' command"
+                exit 1
+            fi
+            ;;
+        *)
+            echo "Error: Unknown argument '$arg'"
+            echo "Usage: $0 prepare [major|minor|patch] [--dry-run]"
+            echo "       $0 finalize [--dry-run]"
+            exit 1
+            ;;
+    esac
+done
+
+# Validate command
+if [[ ! "$COMMAND" =~ ^(prepare|finalize)$ ]]; then
+    echo "Error: Command must be 'prepare' or 'finalize'"
+    echo "Usage: $0 prepare [major|minor|patch] [--dry-run]"
+    echo "       $0 finalize [--dry-run]"
+    exit 1
+fi
+
+# For prepare command, default to patch if no version type specified
+if [ "$COMMAND" = "prepare" ]; then
+    VERSION_TYPE="${VERSION_TYPE:-patch}"
+    if [[ ! "$VERSION_TYPE" =~ ^(major|minor|patch)$ ]]; then
+        echo "Error: Version type must be 'major', 'minor', or 'patch'"
+        echo "Usage: $0 prepare [major|minor|patch] [--dry-run]"
+        exit 1
+    fi
+fi
+
+# Common validation functions
+check_git_status() {
+    # Check for uncommitted changes (skip in dry-run)
+    if [ "$DRY_RUN" = false ] && ! git diff-index --quiet HEAD --; then
+        echo "Error: You have uncommitted changes. Please commit or stash them first."
+        exit 1
+    fi
+}
+
+check_master_branch() {
+    # Ensure we're on master branch
+    CURRENT_BRANCH=$(git branch --show-current)
+    if [ "$CURRENT_BRANCH" != "master" ]; then
+        if [ "$DRY_RUN" = true ]; then
+            echo "[dry run] Warning: Not on master branch (currently on: $CURRENT_BRANCH). Continuing with dry-run..."
+            echo ""
+        else
+            echo "Error: Must be on master branch. Currently on: $CURRENT_BRANCH"
+            exit 1
+        fi
+    fi
+}
+
+check_master_up_to_date() {
+    # Check if we have the latest from master (skip in dry-run)
+    if [ "$DRY_RUN" = false ]; then
+        echo "Checking if local master is up-to-date with remote..."
+        git fetch origin master
+        LOCAL=$(git rev-parse HEAD)
+        REMOTE=$(git rev-parse origin/master)
+
+        if [ "$LOCAL" != "$REMOTE" ]; then
+            echo "Error: Your local master branch is not up-to-date with origin/master."
+            echo "Please run 'git pull origin master' first."
+            exit 1
+        fi
+        echo "✓ Local master is up-to-date with remote"
+        echo ""
+    elif [ "$(git branch --show-current)" = "master" ]; then
+        echo "[dry run] Warning: Dry-run mode - not checking if master is up-to-date with remote"
+        echo ""
+    fi
+}
+
+prepare_release() {
+    if [ "$DRY_RUN" = true ]; then
+        echo "[dry-run] Preparing release (no changes will be made)"
+    else
+        echo "Starting release preparation..."
+    fi
+    echo ""
+
+    check_git_status
+    check_master_branch
+    check_master_up_to_date
+
+    # Extract current version from CMakeLists.txt
+    echo "Step 1: Reading current version..."
+    MAJOR=$(grep "set(GGML_VERSION_MAJOR" CMakeLists.txt | sed 's/.*MAJOR \([0-9]*\).*/\1/')
+    MINOR=$(grep "set(GGML_VERSION_MINOR" CMakeLists.txt | sed 's/.*MINOR \([0-9]*\).*/\1/')
+    PATCH=$(grep "set(GGML_VERSION_PATCH" CMakeLists.txt | sed 's/.*PATCH \([0-9]*\).*/\1/')
+
+    echo "Current version: $MAJOR.$MINOR.$PATCH"
+
+    # Calculate new version
+    case $VERSION_TYPE in
+        major)
+            NEW_MAJOR=$((MAJOR + 1))
+            NEW_MINOR=0
+            NEW_PATCH=0
+            ;;
+        minor)
+            NEW_MAJOR=$MAJOR
+            NEW_MINOR=$((MINOR + 1))
+            NEW_PATCH=0
+            ;;
+        patch)
+            NEW_MAJOR=$MAJOR
+            NEW_MINOR=$MINOR
+            NEW_PATCH=$((PATCH + 1))
+            ;;
+    esac
+
+    NEW_VERSION="$NEW_MAJOR.$NEW_MINOR.$NEW_PATCH"
+    RC_BRANCH="ggml-rc-v$NEW_VERSION"
+    echo "New release version: $NEW_VERSION"
+    echo "Release candidate branch: $RC_BRANCH"
+    echo ""
+
+    # Create release candidate branch
+    echo "Step 2: Creating release candidate branch..."
+    if [ "$DRY_RUN" = true ]; then
+        echo "  [dry-run] Would create branch: $RC_BRANCH"
+    else
+        git checkout -b "$RC_BRANCH"
+        echo "✓ Created and switched to branch: $RC_BRANCH"
+    fi
+    echo ""
+
+    # Update CMakeLists.txt for release
+    echo "Step 3: Updating version in CMakeLists.txt..."
+    if [ "$DRY_RUN" = true ]; then
+        echo "  [dry-run] Would update GGML_VERSION_MAJOR to $NEW_MAJOR"
+        echo "  [dry-run] Would update GGML_VERSION_MINOR to $NEW_MINOR"
+        echo "  [dry-run] Would update GGML_VERSION_PATCH to $NEW_PATCH"
+    else
+        sed -i'' -e "s/set(GGML_VERSION_MAJOR [0-9]*)/set(GGML_VERSION_MAJOR $NEW_MAJOR)/" CMakeLists.txt
+        sed -i'' -e "s/set(GGML_VERSION_MINOR [0-9]*)/set(GGML_VERSION_MINOR $NEW_MINOR)/" CMakeLists.txt
+        sed -i'' -e "s/set(GGML_VERSION_PATCH [0-9]*)/set(GGML_VERSION_PATCH $NEW_PATCH)/" CMakeLists.txt
+    fi
+    echo ""
+
+    # Commit version bump
+    echo "Step 4: Committing version bump..."
+    if [ "$DRY_RUN" = true ]; then
+        echo "  [dry-run] Would commit: 'ggml : bump version to $NEW_VERSION'"
+    else
+        git add CMakeLists.txt
+        git commit -m "ggml : bump version to $NEW_VERSION"
+    fi
+    echo ""
+
+    echo ""
+    if [ "$DRY_RUN" = true ]; then
+        echo "[dry-run] Summary (no changes were made):"
+        echo "  • Would have created branch: $RC_BRANCH"
+        echo "  • Would have updated version to: $NEW_VERSION"
+    else
+        echo "Release preparation completed!"
+        echo "Summary:"
+        echo "  • Created branch: $RC_BRANCH"
+        echo "  • Updated version to: $NEW_VERSION"
+        echo ""
+        echo "Next steps:"
+        echo "  • Push branch to remote: git push origin $RC_BRANCH"
+        echo "  • Create a Pull Request from $RC_BRANCH to master"
+        echo "  • After PR is merged, run: ./scripts/release.sh finalize"
+    fi
+}
+
+finalize_release() {
+    if [ "$DRY_RUN" = true ]; then
+        echo "[dry-run] Finalizing release (no changes will be made)"
+    else
+        echo "Starting release finalization..."
+    fi
+    echo ""
+
+    check_git_status
+    check_master_branch
+    check_master_up_to_date
+
+    # Read current version from CMakeLists.txt
+    echo "Step 1: Reading current release version..."
+    MAJOR=$(grep "set(GGML_VERSION_MAJOR" CMakeLists.txt | sed 's/.*MAJOR \([0-9]*\).*/\1/')
+    MINOR=$(grep "set(GGML_VERSION_MINOR" CMakeLists.txt | sed 's/.*MINOR \([0-9]*\).*/\1/')
+    PATCH=$(grep "set(GGML_VERSION_PATCH" CMakeLists.txt | sed 's/.*PATCH \([0-9]*\).*/\1/')
+
+    RELEASE_VERSION="$MAJOR.$MINOR.$PATCH"
+    echo "Release version: $RELEASE_VERSION"
+    echo ""
+
+    # Create git tag
+    echo "Step 2: Creating signed git tag..."
+    if [ "$DRY_RUN" = true ]; then
+        echo "  [dry-run] Would create signed tag: v$RELEASE_VERSION with message 'Release version $RELEASE_VERSION'"
+    else
+        git tag -s "v$RELEASE_VERSION" -m "Release version $RELEASE_VERSION"
+        echo "✓ Created signed tag: v$RELEASE_VERSION"
+    fi
+    echo ""
+
+
+    echo ""
+    if [ "$DRY_RUN" = true ]; then
+        echo "[dry-run] Summary (no changes were made):"
+        echo "  • Would have created tag: v$RELEASE_VERSION"
+    else
+        echo "Release finalization completed!"
+        echo "Summary:"
+        echo "  • Created signed tag: v$RELEASE_VERSION"
+        echo ""
+        echo "Next steps:"
+        echo "  • Push tag to remote: git push origin v$RELEASE_VERSION"
+        echo "  • The release is now complete!"
+    fi
+}
+
+# Execute the appropriate command
+case $COMMAND in
+    prepare)
+        prepare_release
+        ;;
+    finalize)
+        finalize_release
+        ;;
+esac

data/vendor/ggml/scripts/sync-llama-am.sh

@@ -0,0 +1,167 @@
+#!/bin/bash
+#
+# Synchronize llama.cpp changes to ggml
+#
+# Usage:
+#
+#   $ cd /path/to/ggml
+#   $ ./scripts/sync-llama-am.sh -skip hash0,hash1,hash2... -C 3
+#
+
+set -e
+
+sd=$(dirname $0)
+cd $sd/../
+
+SRC_GGML=$(pwd)
+SRC_LLAMA=$(cd ../llama.cpp; pwd)
+
+if [ ! -d $SRC_LLAMA ]; then
+    echo "llama.cpp not found at $SRC_LLAMA"
+    exit 1
+fi
+
+lc=$(cat $SRC_GGML/scripts/sync-llama.last)
+echo "Syncing llama.cpp changes since commit $lc"
+
+to_skip=""
+
+# context for git patches in number of lines
+ctx="8"
+
+while [ "$1" != "" ]; do
+    case $1 in
+        -skip )
+            shift
+            to_skip=$1
+            ;;
+        -C )
+            shift
+            ctx=$1
+            ;;
+    esac
+    shift
+done
+
+cd $SRC_LLAMA
+
+git log --oneline $lc..HEAD
+git log --oneline $lc..HEAD --reverse | grep -v "(ggml/[0-9]*)" | grep -v "(whisper/[0-9]*)" | cut -d' ' -f1 > $SRC_GGML/llama-commits
+
+if [ ! -s $SRC_GGML/llama-commits ]; then
+    rm -v $SRC_GGML/llama-commits
+    echo "No new commits"
+    exit 0
+fi
+
+if [ -f $SRC_GGML/llama-src.patch ]; then
+    rm -v $SRC_GGML/llama-src.patch
+fi
+
+while read c; do
+    if [ -n "$to_skip" ]; then
+        if [[ $to_skip == *"$c"* ]]; then
+            echo "Skipping $c"
+            continue
+        fi
+    fi
+
+    git format-patch -U${ctx} -k $c~1..$c --stdout -- \
+        ggml/CMakeLists.txt \
+        ggml/src/CMakeLists.txt \
+        ggml/cmake/BuildTypes.cmake \
+        ggml/cmake/GitVars.cmake \
+        ggml/cmake/common.cmake \
+        ggml/cmake/ggml-config.cmake.in \
+        ggml/src/ggml-cpu/cmake/FindSIMD.cmake \
+        ggml/src/ggml* \
+        ggml/src/gguf* \
+        ggml/include/ggml*.h \
+        ggml/include/gguf*.h \
+        tests/test-opt.cpp \
+        tests/test-quantize-fns.cpp \
+        tests/test-quantize-perf.cpp \
+        tests/test-backend-ops.cpp \
+        LICENSE \
+        scripts/gen-authors.sh \
+        >> $SRC_GGML/llama-src.patch
+done < $SRC_GGML/llama-commits
+
+rm -v $SRC_GGML/llama-commits
+
+# delete files if empty
+if [ ! -s $SRC_GGML/llama-src.patch ]; then
+    rm -v $SRC_GGML/llama-src.patch
+fi
+
+cd $SRC_GGML
+
+if [ -f $SRC_GGML/llama-src.patch ]; then
+    # replace PR numbers
+    #
+    # Subject: some text (#1234)
+    # Subject: some text (llama/1234)
+    cat llama-src.patch | sed -e 's/^Subject: \(.*\) (#\([0-9]*\))/Subject: \1 (llama\/\2)/' > llama-src.patch.tmp
+    mv llama-src.patch.tmp llama-src.patch
+
+    cat llama-src.patch | sed -e 's/^\(.*\) (#\([0-9]*\))$/\1 (llama\/\2)/' > llama-src.patch.tmp
+    mv llama-src.patch.tmp llama-src.patch
+
+    # replace filenames:
+    #
+    # ggml/CMakelists.txt       -> CMakeLists.txt
+    # ggml/src/CMakelists.txt   -> src/CMakeLists.txt
+    #
+    # ggml/cmake/BuildTypes.cmake            -> cmake/BuildTypes.cmake
+    # ggml/cmake/GitVars.cmake               -> cmake/GitVars.cmake
+    # ggml/cmake/common.cmake                -> cmake/common.cmake
+    # ggml/cmake/ggml-config.cmake.in        -> cmake/ggml-config.cmake.in
+    # ggml/src/ggml-cpu/cmake/FindSIMD.cmake -> src/ggml-cpu/cmake/FindSIMD.cmake
+    #
+    # ggml/src/ggml* -> src/ggml*
+    # ggml/src/gguf* -> src/gguf*
+    #
+    # ggml/include/ggml*.h -> include/ggml*.h
+    # ggml/include/gguf*.h -> include/gguf*.h
+    #
+    # tests/test-opt.cpp           -> tests/test-opt.cpp
+    # tests/test-quantize-fns.cpp  -> tests/test-quantize-fns.cpp
+    # tests/test-quantize-perf.cpp -> tests/test-quantize-perf.cpp
+    # tests/test-backend-ops.cpp   -> tests/test-backend-ops.cpp
+    #
+    # LICENSE                -> LICENSE
+    # scripts/gen-authors.sh -> scripts/gen-authors.sh
+
+    cat llama-src.patch | sed -E \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/CMakeLists\.txt/\1CMakeLists.txt/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/src\/CMakeLists\.txt/\1src\/CMakeLists.txt/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/cmake\/BuildTypes\.cmake/\1cmake\/BuildTypes\.cmake/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/cmake\/GitVars\.cmake/\1cmake\/GitVars\.cmake/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/cmake\/common\.cmake/\1cmake\/common\.cmake/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/cmake\/ggml-config\.cmake\.in/\1cmake\/ggml-config\.cmake\.in/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/src\/ggml-cpu\/cmake\/FindSIMD\.cmake/\1src\/ggml-cpu\/cmake\/FindSIMD\.cmake/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/src\/ggml(.*)/\1src\/ggml\2/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/src\/gguf(.*)/\1src\/gguf\2/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/include\/ggml(.*)\.h/\1include\/ggml\2.h/g' \
+        -e 's/(^[[:space:]]| [ab]\/)ggml\/include\/gguf(.*)\.h/\1include\/gguf\2.h/g' \
+        -e 's/(^[[:space:]]| [ab]\/)tests\/test-opt\.cpp/\1tests\/test-opt.cpp/g' \
+        -e 's/(^[[:space:]]| [ab]\/)tests\/test-quantize-fns\.cpp/\1tests\/test-quantize-fns.cpp/g' \
+        -e 's/(^[[:space:]]| [ab]\/)tests\/test-quantize-perf\.cpp/\1tests\/test-quantize-perf.cpp/g' \
+        -e 's/(^[[:space:]]| [ab]\/)tests\/test-backend-ops\.cpp/\1tests\/test-backend-ops.cpp/g' \
+        -e 's/(^[[:space:]]| [ab]\/)LICENSE/\1LICENSE/g' \
+        -e 's/(^[[:space:]]| [ab]\/)scripts\/gen-authors\.sh/\1scripts\/gen-authors.sh/g' \
+        > llama-src.patch.tmp
+    mv llama-src.patch.tmp llama-src.patch
+
+    git am -C${ctx} llama-src.patch
+
+    rm -v $SRC_GGML/llama-src.patch
+fi
+
+# update last commit
+cd $SRC_LLAMA
+git log -1 --format=%H > $SRC_GGML/scripts/sync-llama.last
+
+echo "Done"
+
+exit 0