gguf-cpp 0.0.5__tar.gz → 0.0.7__tar.gz

{gguf_cpp-0.0.5 → gguf_cpp-0.0.7}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.1
 Name: gguf-cpp
-Version: 0.0.5
-Summary: Unified GUI + engine for GGUF models: local LLM server, image generation and GGUF editing/quantization, all on one bundled gguf.cpp engine
+Version: 0.0.7
+Summary: Unified GUI + engine for GGUF models: local LLM server, image/video generation and GGUF editing/quantization, all on one bundled gguf.cpp engine
 Author-Email: gguf <info@gguf.us>
 License: MIT
 Classifier: License :: OSI Approved :: MIT License
@@ -46,6 +46,8 @@ gguf-cpp                 # unified GUI — Server / Diffuser / Editor panels
 python -m gguf_cpp       # same thing
 ```
 
+![screenshot](https://raw.githubusercontent.com/gguf-org/gguf-desktop/master/demo15.gif)
+
 Each panel also runs on its own, exactly like the standalone
 gguf-server / gguf-diffusion / gguf-editor packages did:
 
@@ -72,8 +74,10 @@ vendor/engine/           the unified gguf.cpp engine (one CMake build)
   src/ common/ mtmd/     GGUF LLM runtime
   app/                   the gguf-server HTTP server
   diffusion/             diffusion runtime + CLI
-  quantizer/             quantizer shared library (shares the kernels above)
+  quantizer/             quantizer shared library (its own quant kernels)
 src/gguf_cpp/            the Python package
   server/ diffuser/ editor/   the three panels (backend + web frontend each)
   gui.py static/         the unified 3-panel GUI shell
 ```
+
+![screenshot](https://raw.githubusercontent.com/gguf-org/gguf-desktop/master/pizza.jpg)

{gguf_cpp-0.0.5 → gguf_cpp-0.0.7}/README.md

@@ -28,6 +28,8 @@ gguf-cpp                 # unified GUI — Server / Diffuser / Editor panels
 python -m gguf_cpp       # same thing
 ```
 
+![screenshot](https://raw.githubusercontent.com/gguf-org/gguf-desktop/master/demo15.gif)
+
 Each panel also runs on its own, exactly like the standalone
 gguf-server / gguf-diffusion / gguf-editor packages did:
 
@@ -54,8 +56,10 @@ vendor/engine/           the unified gguf.cpp engine (one CMake build)
   src/ common/ mtmd/     GGUF LLM runtime
   app/                   the gguf-server HTTP server
   diffusion/             diffusion runtime + CLI
-  quantizer/             quantizer shared library (shares the kernels above)
+  quantizer/             quantizer shared library (its own quant kernels)
 src/gguf_cpp/            the Python package
   server/ diffuser/ editor/   the three panels (backend + web frontend each)
   gui.py static/         the unified 3-panel GUI shell
 ```
+
+![screenshot](https://raw.githubusercontent.com/gguf-org/gguf-desktop/master/pizza.jpg)

{gguf_cpp-0.0.5 → gguf_cpp-0.0.7}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "scikit_build_core.build"
 [project]
 name = "gguf-cpp"
 dynamic = ["version"]
-description = "Unified GUI + engine for GGUF models: local LLM server, image generation and GGUF editing/quantization, all on one bundled gguf.cpp engine"
+description = "Unified GUI + engine for GGUF models: local LLM server, image/video generation and GGUF editing/quantization, all on one bundled gguf.cpp engine"
 readme = "README.md"
 license = { text = "MIT" }
 authors = [{ name = "gguf", email = "info@gguf.us" }]

gguf_cpp-0.0.7/src/gguf_cpp/__init__.py

@@ -0,0 +1 @@
+__version__ = '0.0.7'

{gguf_cpp-0.0.5 → gguf_cpp-0.0.7}/src/gguf_cpp/editor/gguf.py

@@ -35,7 +35,7 @@ QUANTIZATION_NAME = {
     19: "IQ1_S", 20: "IQ4_NL", 21: "IQ3_S", 22: "IQ2_S",
     23: "IQ4_XS", 24: "I8", 25: "I16", 26: "I32", 27: "I64",
     28: "F64", 29: "IQ1_M", 30: "BF16", 34: "TQ1_0", 35: "TQ2_0",
-    39: "MXFP4", 40: "NVFP4", 41: "Q1_0",
+    39: "MXFP4", 40: "NVFP4", 41: "Q1_0", 42: "Q2_0",
 }
 
 # (block_size, type_size) per ggml_type id — mirrors ggml_blck_size /
@@ -48,12 +48,12 @@ GGML_TYPE_TRAITS = {
     20: (32, 18), 21: (256, 110), 22: (256, 82), 23: (256, 136),
     24: (1, 1), 25: (1, 2), 26: (1, 4), 27: (1, 8), 28: (1, 8),
     29: (256, 56), 30: (1, 2), 34: (256, 54), 35: (256, 66),
-    39: (32, 17), 40: (64, 36), 41: (128, 18),
+    39: (32, 17), 40: (64, 36), 41: (128, 18), 42: (64, 18),
 }
 
 # dtypes a user can pick for a new zero-filled tensor (order = UI order)
 ZERO_TENSOR_DTYPES = [0, 1, 30, 8, 2, 3, 6, 7, 15, 14, 13, 12, 11, 10,
-                      19, 29, 16, 17, 22, 18, 21, 23, 20, 39, 40]
+                      19, 29, 16, 17, 22, 18, 21, 23, 20, 39, 40, 41, 42]
 
 
 def quantization_name(dtype: int) -> str:

{gguf_cpp-0.0.5 → gguf_cpp-0.0.7}/src/gguf_cpp/editor/static/app.js

@@ -43,7 +43,7 @@ const GGML_TYPE_TRAITS = {
 };
 
 const ZERO_TENSOR_DTYPES = [0, 1, 30, 8, 2, 3, 6, 7, 15, 14, 13, 12, 11, 10,
-  19, 29, 16, 17, 22, 18, 21, 23, 20, 39, 40];
+  19, 29, 16, 17, 22, 18, 21, 23, 20, 39, 40, 41, 42];
 
 // Quantizer target types (name → ggml dtype id), same list as the desktop
 // editor / quantizer CLI.

{gguf_cpp-0.0.5 → gguf_cpp-0.0.7}/vendor/engine/CMakeLists.txt

@@ -12,7 +12,7 @@ project("gguf-cpp" C CXX)
 #   app/         the gguf-server HTTP server
 #   ui/          embedded web UI assets (optional, empty by default)
 #   diffusion/   the diffusion runtime + `diffusion` CLI (links shared ggml)
-#   quantizer/   the standalone quantizer shared library (shares ggml-quants)
+#   quantizer/   the standalone quantizer shared library (own quant kernels)
 #   thirdparty/  cpp-httplib, nlohmann/json, stb, miniaudio, subprocess.h
 #
 # There is no vendored llama.cpp checkout and no external ggml: everything is

gguf_cpp-0.0.7/vendor/engine/kernels/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 gguf.org and the ggml authors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

{gguf_cpp-0.0.5 → gguf_cpp-0.0.7}/vendor/engine/quantizer/CMakeLists.txt

@@ -1,12 +1,12 @@
 # Standalone GGUF quantizer, built as part of the unified gguf.cpp engine as
 # a shared library (the editor panel loads it via ctypes).
 #
-# The quant kernels are no longer a private copy: ggml-quants.c and every
-# ggml header come from the shared <engine>/kernels tree, so the quantizer
-# always agrees with the server/diffusion runtimes on quant formats (including
-# the fork's own types: q1_0, q2_0, mxfp4, nvfp4). The small qz-traits.c /
-# qz-init.cpp glue still stands in for the rest of ggml.c, keeping the
-# library free of the full ggml runtime.
+# The quant kernels are the quantizer's own (src/kernels/qz_*), written against
+# the GGUF on-disk formats rather than shared with the runtimes: encoding is a
+# different job from inference, and keeping the two apart means this library
+# pulls in nothing from the engine tree. The formats themselves still line up
+# with what the server and diffusion runtimes read, including the fork's own
+# types - q1_0, q2_0, mxfp4, nvfp4 - so anything produced here loads there.
 
 option(QUANTIZER_CUDA  "quantizer: CUDA support (NVIDIA GPUs)" OFF)
 option(QUANTIZER_HIP   "quantizer: HIP/ROCm support (AMD GPUs)" OFF)
@@ -23,7 +23,6 @@ if(MSVC)
 endif()
 
 set(QZ_DIR ${CMAKE_CURRENT_SOURCE_DIR})
-set(QZ_KERNELS_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../kernels)
 
 set(QUANTIZER_SOURCES
     ${QZ_DIR}/src/quantize.cpp
@@ -34,9 +33,13 @@ set(QUANTIZER_SOURCES
     ${QZ_DIR}/src/safetensors_io.h
     ${QZ_DIR}/src/device.cpp
     ${QZ_DIR}/src/device.h
-    ${QZ_KERNELS_DIR}/src/ggml-quants.c   # shared with the server/diffusion runtimes
-    ${QZ_DIR}/src/kernels/qz-traits.c
-    ${QZ_DIR}/src/kernels/qz-init.cpp
+    ${QZ_DIR}/src/kernels/qz_pack.c
+    ${QZ_DIR}/src/kernels/qz_super.c
+    ${QZ_DIR}/src/kernels/qz_lattice.c
+    ${QZ_DIR}/src/kernels/qz_decode.c
+    ${QZ_DIR}/src/kernels/qz_codebook.c
+    ${QZ_DIR}/src/kernels/qz_traits.c
+    ${QZ_DIR}/src/kernels/qz_init.cpp
 )
 
 if(QUANTIZER_CUDA)
@@ -63,11 +66,7 @@ if(QUANTIZER_METAL)
 endif()
 
 add_library(gguf_quantizer SHARED ${QUANTIZER_SOURCES})
-target_include_directories(gguf_quantizer PUBLIC
-    ${QZ_DIR}/src
-    ${QZ_KERNELS_DIR}/include
-    ${QZ_KERNELS_DIR}/src
-)
+target_include_directories(gguf_quantizer PUBLIC ${QZ_DIR}/src)
 target_compile_definitions(gguf_quantizer PRIVATE GGUF_QUANTIZER_BUILD PUBLIC GGUF_QUANTIZER_SHARED)
 
 if(QUANTIZER_CUDA)

{gguf_cpp-0.0.5 → gguf_cpp-0.0.7}/vendor/engine/quantizer/src/device.h

@@ -2,13 +2,13 @@
 
 // Internal device/accelerator abstraction.
 //
-// The CPU path (vendored ggml kernels, see src/kernels/) is always available
+// The CPU path (the kernels in src/kernels/) is always available
 // and needs no converter. Accelerator backends (CUDA/ROCm/Metal) implement
 // device_converter with fast f32 -> dst quantization kernels for a subset of
 // types; tensors with unsupported target types transparently fall back to the
 // CPU kernels.
 
-#include "ggml.h"
+#include "kernels/qz_quant.h"
 
 #include <memory>
 #include <string>
@@ -28,12 +28,12 @@ public:
     virtual const std::string & name() const = 0;
 
     // true if this device has a f32 -> dst_type quantization kernel
-    virtual bool supports(ggml_type dst_type) const = 0;
+    virtual bool supports(qz_type dst_type) const = 0;
 
     // quantize `nrows` rows of `n_per_row` f32 values into dst.
     // returns false and sets `error` on a runtime failure (the caller then
     // falls back to the CPU kernels for this tensor).
-    virtual bool convert(const float * src, ggml_type dst_type, void * dst, int64_t nrows,
+    virtual bool convert(const float * src, qz_type dst_type, void * dst, int64_t nrows,
                          int64_t n_per_row, std::string & error) = 0;
 };