stop tracking build/, like the other ggml kernels

`build/` is already gitignored here, and `ggml-attn` and `ggml-gated-delta-net` never tracked it: the
artifacts a client downloads live on the kernel repo's `v1` branch, which `build-and-upload` writes.
Keeping a second copy in git only creates two things to keep in step, and they had already drifted --
what shipped was stamped 09da59d, what was committed here 1f05408.

.gitattributes

@@ -33,5 +33,3 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
-# Every built artifact, whatever hash the build stamps into its name.
-build/**/*.abi3.so filter=lfs diff=lfs merge=lfs -text

build/torch212-metal-aarch64-darwin/__init__.py

@@ -1,78 +0,0 @@
-"""Compute directly on the packed blocks of a GGUF checkpoint.
-
-A GGUF weight is stored as blocks of 32 or 256 values sharing a scale. These kernels read that
-layout as it is, so a quantized model can be loaded and run without ever materializing a dense
-copy — which is the whole memory saving.
-
-Ported from llama.cpp's ggml-cuda; `vendor/UPSTREAM` pins the revision. The ops are named for what
-they do rather than for a backend: each backend registers its own implementation of the same
-schema, so calls dispatch on the tensor's device.
-"""
-
-import torch
-
-from ._ops import add_op_namespace_prefix, ops
-
-
-__all__ = ["GEMV_TYPES", "MAX_GEMV_ROWS", "dequantize", "get_rows", "mul_mat_vec"]
-
-# Upstream's MMVQ_MAX_BATCH_SIZE: `mul_mat_vec` has no implementation beyond this many rows, so a
-# caller with more (prefill) dequantizes and uses an ordinary matmul.
-MAX_GEMV_ROWS = 8
-
-# ggml type ids `mul_mat_vec` implements, as this build reports them -- it is backend-specific, and
-# a type routed into a gemv that has no kernel for it is a fault, not a fallback. CUDA covers
-# Q4_0/Q4_1/Q5_0/Q5_1/Q8_0, the K quants, the IQ quants, MXFP4/NVFP4 and Q1_0/Q2_0; Metal is the
-# same minus NVFP4/Q1_0/Q2_0, which its metallib has no kernels for. `dequantize` covers more, so
-# check this before choosing the fused path.
-try:
-    GEMV_TYPES = frozenset(ops.gemv_types())
-except AttributeError:
-    # A build published before `gemv_types` existed; those are CUDA-only, so its list is theirs.
-    GEMV_TYPES = frozenset({2, 3, 6, 7, 8, 10, 11, 12, 13, 14, 16, 17, 18, 19, 20, 21, 22, 23, 29, 39, 40, 41, 42})
-
-
-def get_rows(
-    blocks: torch.Tensor, indices: torch.Tensor, ggml_type: int, cols: int, dtype: torch.dtype
-) -> torch.Tensor:
-    """The rows `indices` names, unpacked: `(rows, bytes_per_row)` uint8 -> `(len(indices), cols)`.
-
-    ggml's `get_rows`, which dequantizes as it gathers -- so reading a few rows out of a large table
-    never touches the rest.
-    """
-    return ops.get_rows(blocks, indices, ggml_type, cols, dtype)
-
-
-def dequantize(
-    blocks: torch.Tensor, ggml_type: int, rows: int, cols: int, dtype: torch.dtype
-) -> torch.Tensor:
-    """`(rows, bytes_per_row)` uint8 blocks -> `(rows, cols)` values of `dtype`."""
-    return ops.dequantize(blocks, ggml_type, rows, cols, dtype)
-
-
-def mul_mat_vec(
-    blocks: torch.Tensor, x: torch.Tensor, ggml_type: int, out_features: int
-) -> torch.Tensor:
-    """Fused dequantize-gemv: `x @ blocks.T` without unpacking `blocks`.
-
-    `x` is `(rows, in_features)` and `rows` must be at most `MAX_GEMV_ROWS`. The result is f32
-    whatever `x`'s dtype was, since the kernel writes an f32 destination; cast it at the call site,
-    where the cast can be fused into whatever consumes it.
-    """
-    return ops.mul_mat_vec(blocks, x, ggml_type, out_features)
-
-
-# Without these, torch.compile cannot trace the ops and breaks the graph at every call.
-@torch.library.register_fake(add_op_namespace_prefix("get_rows"))
-def _get_rows_fake(blocks, indices, ggml_type, cols, dtype):
-    return blocks.new_empty((indices.numel(), cols), dtype=dtype)
-
-
-@torch.library.register_fake(add_op_namespace_prefix("dequantize"))
-def _dequantize_fake(blocks, ggml_type, rows, cols, dtype):
-    return blocks.new_empty((rows, cols), dtype=dtype)
-
-
-@torch.library.register_fake(add_op_namespace_prefix("mul_mat_vec"))
-def _mul_mat_vec_fake(blocks, x, ggml_type, out_features):
-    return x.new_empty((x.shape[0], out_features), dtype=torch.float32)

build/torch212-metal-aarch64-darwin/_ggml_quantization_metal_09da59d.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f3ec2c0bd762a1cb5855006a816cdb22f18c11e969006d17131460e35e577ebb
-size 817976

build/torch212-metal-aarch64-darwin/_ops.py

@@ -1,9 +0,0 @@
-import torch
-from . import _ggml_quantization_metal_09da59d
-ops = torch.ops._ggml_quantization_metal_09da59d
-
-def add_op_namespace_prefix(op_name: str):
-    """
-    Prefix op by namespace.
-    """
-    return f"_ggml_quantization_metal_09da59d::{op_name}"

build/torch212-metal-aarch64-darwin/metadata.json

@@ -1,29 +0,0 @@
-{
-  "name": "ggml-quantization",
-  "id": "_ggml_quantization_metal_09da59d",
-  "version": 1,
-  "license": "MIT",
-  "python-depends": [],
-  "backend": {
-    "type": "metal"
-  },
-  "digest": {
-    "algorithm": "sha256",
-    "files": {
-      "__init__.py": "68gDnPDp55W5ZSVOVBPJqML9O/YHq9mzcdCvNs4rnEg=",
-      "_ggml_quantization_metal_09da59d.abi3.so": "8+wsC9dioctYVQBqgWzbIvGMEelpAG0XExRg415Xfrs=",
-      "_ops.py": "KPScFwQZRvXLLvmzyYkY4+0hIJB+X5ZJ9opUxecMnZs="
-    }
-  },
-  "provenance": {
-    "kernel-builder": {
-      "version": "0.17.0-dev0",
-      "sha": "81f55ea30fd8f819dcf93a3c934dd584c895bd2f",
-      "dirty": false
-    },
-    "kernel": {
-      "sha": "09da59d51d3f73baa39a941eb07bf6c386223f82",
-      "dirty": false
-    }
-  }
-}

build/torch213-metal-aarch64-darwin/__init__.py

@@ -1,78 +0,0 @@
-"""Compute directly on the packed blocks of a GGUF checkpoint.
-
-A GGUF weight is stored as blocks of 32 or 256 values sharing a scale. These kernels read that
-layout as it is, so a quantized model can be loaded and run without ever materializing a dense
-copy — which is the whole memory saving.
-
-Ported from llama.cpp's ggml-cuda; `vendor/UPSTREAM` pins the revision. The ops are named for what
-they do rather than for a backend: each backend registers its own implementation of the same
-schema, so calls dispatch on the tensor's device.
-"""
-
-import torch
-
-from ._ops import add_op_namespace_prefix, ops
-
-
-__all__ = ["GEMV_TYPES", "MAX_GEMV_ROWS", "dequantize", "get_rows", "mul_mat_vec"]
-
-# Upstream's MMVQ_MAX_BATCH_SIZE: `mul_mat_vec` has no implementation beyond this many rows, so a
-# caller with more (prefill) dequantizes and uses an ordinary matmul.
-MAX_GEMV_ROWS = 8
-
-# ggml type ids `mul_mat_vec` implements, as this build reports them -- it is backend-specific, and
-# a type routed into a gemv that has no kernel for it is a fault, not a fallback. CUDA covers
-# Q4_0/Q4_1/Q5_0/Q5_1/Q8_0, the K quants, the IQ quants, MXFP4/NVFP4 and Q1_0/Q2_0; Metal is the
-# same minus NVFP4/Q1_0/Q2_0, which its metallib has no kernels for. `dequantize` covers more, so
-# check this before choosing the fused path.
-try:
-    GEMV_TYPES = frozenset(ops.gemv_types())
-except AttributeError:
-    # A build published before `gemv_types` existed; those are CUDA-only, so its list is theirs.
-    GEMV_TYPES = frozenset({2, 3, 6, 7, 8, 10, 11, 12, 13, 14, 16, 17, 18, 19, 20, 21, 22, 23, 29, 39, 40, 41, 42})
-
-
-def get_rows(
-    blocks: torch.Tensor, indices: torch.Tensor, ggml_type: int, cols: int, dtype: torch.dtype
-) -> torch.Tensor:
-    """The rows `indices` names, unpacked: `(rows, bytes_per_row)` uint8 -> `(len(indices), cols)`.
-
-    ggml's `get_rows`, which dequantizes as it gathers -- so reading a few rows out of a large table
-    never touches the rest.
-    """
-    return ops.get_rows(blocks, indices, ggml_type, cols, dtype)
-
-
-def dequantize(
-    blocks: torch.Tensor, ggml_type: int, rows: int, cols: int, dtype: torch.dtype
-) -> torch.Tensor:
-    """`(rows, bytes_per_row)` uint8 blocks -> `(rows, cols)` values of `dtype`."""
-    return ops.dequantize(blocks, ggml_type, rows, cols, dtype)
-
-
-def mul_mat_vec(
-    blocks: torch.Tensor, x: torch.Tensor, ggml_type: int, out_features: int
-) -> torch.Tensor:
-    """Fused dequantize-gemv: `x @ blocks.T` without unpacking `blocks`.
-
-    `x` is `(rows, in_features)` and `rows` must be at most `MAX_GEMV_ROWS`. The result is f32
-    whatever `x`'s dtype was, since the kernel writes an f32 destination; cast it at the call site,
-    where the cast can be fused into whatever consumes it.
-    """
-    return ops.mul_mat_vec(blocks, x, ggml_type, out_features)
-
-
-# Without these, torch.compile cannot trace the ops and breaks the graph at every call.
-@torch.library.register_fake(add_op_namespace_prefix("get_rows"))
-def _get_rows_fake(blocks, indices, ggml_type, cols, dtype):
-    return blocks.new_empty((indices.numel(), cols), dtype=dtype)
-
-
-@torch.library.register_fake(add_op_namespace_prefix("dequantize"))
-def _dequantize_fake(blocks, ggml_type, rows, cols, dtype):
-    return blocks.new_empty((rows, cols), dtype=dtype)
-
-
-@torch.library.register_fake(add_op_namespace_prefix("mul_mat_vec"))
-def _mul_mat_vec_fake(blocks, x, ggml_type, out_features):
-    return x.new_empty((x.shape[0], out_features), dtype=torch.float32)

build/torch213-metal-aarch64-darwin/_ggml_quantization_metal_09da59d.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:391156ef4b8ec8f557c1ecb59a89d716a289e41e1c3853c6f0566330aafe196a
-size 817976

build/torch213-metal-aarch64-darwin/_ops.py

@@ -1,9 +0,0 @@
-import torch
-from . import _ggml_quantization_metal_09da59d
-ops = torch.ops._ggml_quantization_metal_09da59d
-
-def add_op_namespace_prefix(op_name: str):
-    """
-    Prefix op by namespace.
-    """
-    return f"_ggml_quantization_metal_09da59d::{op_name}"

build/torch213-metal-aarch64-darwin/metadata.json

@@ -1,29 +0,0 @@
-{
-  "name": "ggml-quantization",
-  "id": "_ggml_quantization_metal_09da59d",
-  "version": 1,
-  "license": "MIT",
-  "python-depends": [],
-  "backend": {
-    "type": "metal"
-  },
-  "digest": {
-    "algorithm": "sha256",
-    "files": {
-      "__init__.py": "68gDnPDp55W5ZSVOVBPJqML9O/YHq9mzcdCvNs4rnEg=",
-      "_ggml_quantization_metal_09da59d.abi3.so": "ORFW70uOyPVXwey1monXFqKJ5B4cOFPG8FZjMKr+GWo=",
-      "_ops.py": "KPScFwQZRvXLLvmzyYkY4+0hIJB+X5ZJ9opUxecMnZs="
-    }
-  },
-  "provenance": {
-    "kernel-builder": {
-      "version": "0.17.0-dev0",
-      "sha": "81f55ea30fd8f819dcf93a3c934dd584c895bd2f",
-      "dirty": false
-    },
-    "kernel": {
-      "sha": "09da59d51d3f73baa39a941eb07bf6c386223f82",
-      "dirty": false
-    }
-  }
-}