7f59912916
The model code was spread across eight crates that had grown into each other:
ggml and cuda and mlx each owned part of a tensor runtime, llama and tts and
voice2 each owned part of a model, and libs/diffusion owned everything else.
They are now one tree with an explicit shape:
libs/ai/cuda — kernels and launch surface
libs/ai/metal — Metal shaders and the shim
libs/ai/llm — the language-model runtime (sessions, lanes, contexts,
the CUDA and Metal executors, the compiled Metal path)
libs/ai/models/ — common, flux, h3, music, paint, speech, stems, vision
libs/diffusion is not deleted but demoted: what remains is the VALIDATOR
crate — several dozen `*_validate.rs` oracles that check a native
implementation against a reference, which is where they belong now that the
implementations live next door.
The functional work inside the move is mostly in the LLM runtime: N lanes that
draft while one verify batch serves all of them, per-slot prefill over a shared
folded attention arena, speculation that survives batching, and a scheduler
that reports rather than publishes. And in the CUDA build: a machine without
usable CUDA must still LINK (and say so), the default kernel arch is the
building machine's GPU, `NO_CUDA` forces the stub even where the toolkit
exists, and kernels compile in parallel with progress.
libs/video_flow is new here: classical optical flow estimation and the `mkfl`
motion-field payload — a flow field measured from a clip without a model,
which is what drives free-rate bounce-looping playback and the uprez/tween
enhance pipe.
97 lines
4.6 KiB
Text
97 lines
4.6 KiB
Text
Third-party notices for the Makepad AI inference stack (`libs/ai`).
|
|
This file is the MIT attribution surface for vendored and ported sources.
|
|
Makepad's own code in this tree remains under the repository MIT license
|
|
(see `/LICENSE`). Publishing this repository is distribution of these
|
|
portions; the copyright notices and permission notices below must be
|
|
retained.
|
|
|
|
========================================================================
|
|
ggml / llama.cpp
|
|
========================================================================
|
|
|
|
License: MIT
|
|
Copyright (c) 2023-2026 The ggml authors
|
|
Upstream: https://github.com/ggml-org/llama.cpp
|
|
Upstream LICENSE: MIT License, "Copyright (c) 2023-2026 The ggml authors"
|
|
|
|
Substantial portions of the following are derived from ggml / llama.cpp
|
|
(quantized GEMM, flash-attention, K-quant layouts, Metal kernels, GGUF
|
|
container details, and related device code). Some files are close
|
|
transcriptions of a pinned llama.cpp revision; others are reimplementations
|
|
that still retain ggml block layouts, kernel structure, or constants.
|
|
|
|
libs/ai/cuda/kernels/llm/fattn/ CUDA flash-attn, MMQ, MMVQ, vec-dot
|
|
libs/ai/cuda/kernels/llm/kernels.cu K-quant decode / MMVQ launch path
|
|
libs/ai/cuda/kernels/kquants.cu K-quant / Q4_0 / related dequant
|
|
libs/ai/metal/shaders/ggml/ vendored ggml-metal snapshot
|
|
libs/ai/loader/src/formats/gguf.rs GGUF reader (moved from llama)
|
|
libs/ai/loader/src/quant.rs GGUF dtype / block metadata
|
|
|
|
Pinned Metal snapshot (see that directory's LICENSE):
|
|
commit 364c77f4ca2737e3287652e0e8a8c6dce3231bba
|
|
|
|
CUDA kernel comments name the llama.cpp revision they were ported from
|
|
(for example b10430 / commit 4c1a0af40d88c7fbb3b15c85bf2e8016d1d5b64c).
|
|
That pin is documentation of provenance, not a claim that every later
|
|
edit is still byte-identical to that commit.
|
|
|
|
The MIT permission notice for ggml / llama.cpp:
|
|
|
|
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.
|
|
|
|
========================================================================
|
|
Apple MLX (steel quantized GEMM)
|
|
========================================================================
|
|
|
|
License: MIT
|
|
Copyright © 2023-2025 Apple Inc.
|
|
Upstream: https://github.com/ml-explore/mlx
|
|
Vendored snapshot: v0.31.2
|
|
Location: libs/ai/metal/shaders/mlx_qmm/
|
|
Full license text: libs/ai/metal/shaders/mlx_qmm/LICENSE
|
|
|
|
Only the steel GEMM + quantized.h pieces needed for one instantiation
|
|
are vendored. Do not treat that directory as a full MLX Metal port.
|
|
|
|
========================================================================
|
|
Tokenizer tables bundled with Flux text encoders
|
|
========================================================================
|
|
|
|
libs/ai/models/flux/assets/clip_merges_utf8.txt
|
|
CLIP BPE merges table. CLIP (OpenAI) is MIT-licensed.
|
|
Copyright (c) 2021 OpenAI.
|
|
|
|
libs/ai/models/flux/assets/t5_tokenizer.json
|
|
T5 SentencePiece / HuggingFace tokenizer dump.
|
|
T5 (Google) is Apache License 2.0.
|
|
https://github.com/google-research/text-to-text-transfer-transformer
|
|
|
|
These files are data tables required to run the corresponding checkpoints.
|
|
They are not model weights.
|
|
|
|
========================================================================
|
|
What this NOTICE does not cover
|
|
========================================================================
|
|
|
|
Model *weights* that a user downloads (FLUX, Hunyuan, Qwen, MiniMax, …)
|
|
are licensed separately by their authors. This stack does not host or
|
|
redistribute those weights. The Asset UI requires an explicit per-model
|
|
acknowledgement of the registry license record before a model is cleared
|
|
for download or generation. See libs/asset/ai/registry.json `license`
|
|
objects and the Hunyuan paint `LicenseAcknowledgement` gate.
|