Uploaded using `kernel-builder`.

.gitattributes

@@ -59,3 +59,6 @@ build/torch210-cxx11-cpu-x86_64-linux/_quantization_bitsandbytes_cpu_e09894a.abi
 build/torch211-cxx11-cpu-x86_64-linux/_quantization_bitsandbytes_cpu_e09894a.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch210-cxx11-cpu-x86_64-linux/_quantization_bitsandbytes_cpu_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch211-cxx11-cpu-x86_64-linux/_quantization_bitsandbytes_cpu_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch210-cxx11-cpu-x86_64-linux/_quantization_bitsandbytes_cpu_817f262.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch211-cxx11-cpu-x86_64-linux/_quantization_bitsandbytes_cpu_817f262.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch212-cxx11-cpu-x86_64-linux/_quantization_bitsandbytes_cpu_817f262.abi3.so filter=lfs diff=lfs merge=lfs -text

build/torch210-cxx11-cpu-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _quantization_bitsandbytes_cpu_86f75d9
-ops = torch.ops._quantization_bitsandbytes_cpu_86f75d9
+from . import _quantization_bitsandbytes_cpu_817f262
+ops = torch.ops._quantization_bitsandbytes_cpu_817f262
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_quantization_bitsandbytes_cpu_86f75d9::{op_name}"
+    return f"_quantization_bitsandbytes_cpu_817f262::{op_name}"

build/torch210-cxx11-cpu-x86_64-linux/{_quantization_bitsandbytes_cpu_86f75d9.abi3.so → _quantization_bitsandbytes_cpu_817f262.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea6acbfbebf83a26d5c02ff66d1fb790ec954fedec47550e1fe8cecba317cac8
+oid sha256:ead6baa630df2338c3a3b09107d51e7af1e5014e4ee2d4960d1a43a360e85b1f
 size 1993840

build/torch210-cxx11-cpu-x86_64-linux/metadata.json

@@ -1,6 +1,6 @@
 {
   "name": "quantization-bitsandbytes",
-  "id": "_quantization_bitsandbytes_cpu_86f75d9",
+  "id": "_quantization_bitsandbytes_cpu_817f262",
   "version": 1,
   "license": "MIT",
   "python-depends": [],

build/torch211-cxx11-cpu-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _quantization_bitsandbytes_cpu_86f75d9
-ops = torch.ops._quantization_bitsandbytes_cpu_86f75d9
+from . import _quantization_bitsandbytes_cpu_817f262
+ops = torch.ops._quantization_bitsandbytes_cpu_817f262
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_quantization_bitsandbytes_cpu_86f75d9::{op_name}"
+    return f"_quantization_bitsandbytes_cpu_817f262::{op_name}"

build/torch211-cxx11-cpu-x86_64-linux/{_quantization_bitsandbytes_cpu_86f75d9.abi3.so → _quantization_bitsandbytes_cpu_817f262.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5663ca01b1c628a724bdbf25274202a5277a3ce55077092c6e8d6cd626b06e8c
+oid sha256:d040bd66e1feb75b566ff987f9ac4baf00520ee9252608f5427f96dd458d59fc
 size 1993840

build/torch211-cxx11-cpu-x86_64-linux/metadata.json

@@ -1,6 +1,6 @@
 {
   "name": "quantization-bitsandbytes",
-  "id": "_quantization_bitsandbytes_cpu_86f75d9",
+  "id": "_quantization_bitsandbytes_cpu_817f262",
   "version": 1,
   "license": "MIT",
   "python-depends": [],

build/torch212-cxx11-cpu-x86_64-linux/__init__.py

@@ -0,0 +1,3 @@
+from .custom_ops import gemm_4bit_forward
+
+__all__ = ["gemm_4bit_forward"]

build/torch212-cxx11-cpu-x86_64-linux/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _quantization_bitsandbytes_cpu_817f262
+ops = torch.ops._quantization_bitsandbytes_cpu_817f262
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_quantization_bitsandbytes_cpu_817f262::{op_name}"

build/torch212-cxx11-cpu-x86_64-linux/_quantization_bitsandbytes_cpu_817f262.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba7d51e9f9a7f540f80697163c3caffdae7c2f036904cee93e6d9406f28a4741
+size 1993856

build/torch212-cxx11-cpu-x86_64-linux/custom_ops.py

@@ -0,0 +1,19 @@
+import torch
+from ._ops import ops
+
+def gemm_4bit_forward(
+    input: torch.Tensor,
+    weight: torch.Tensor,
+    absmax: torch.Tensor,
+    blocksize: int,
+    quant_type: int,
+) -> torch.Tensor:
+    original_dtype = input.dtype
+    if original_dtype != torch.bfloat16:
+        input = input.to(torch.bfloat16)
+
+    output = ops.gemm_4bit_forward(input, weight, absmax, blocksize, quant_type)
+    if original_dtype != torch.bfloat16:
+        output = output.to(original_dtype)
+
+    return output

build/torch212-cxx11-cpu-x86_64-linux/metadata.json

@@ -0,0 +1,10 @@
+{
+  "name": "quantization-bitsandbytes",
+  "id": "_quantization_bitsandbytes_cpu_817f262",
+  "version": 1,
+  "license": "MIT",
+  "python-depends": [],
+  "backend": {
+    "type": "cpu"
+  }
+}

build/torch212-cxx11-cpu-x86_64-linux/quantization_bitsandbytes/__init__.py

@@ -0,0 +1,26 @@
+import ctypes
+import importlib.util
+import sys
+from pathlib import Path
+from types import ModuleType
+
+
+def _import_from_path(file_path: Path) -> ModuleType:
+    # We cannot use the module name as-is, after adding it to `sys.modules`,
+    # it would also be used for other imports. So, we make a module name that
+    # depends on the path for it to be unique using the hex-encoded hash of
+    # the path.
+    path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
+    module_name = path_hash
+    spec = importlib.util.spec_from_file_location(module_name, file_path)
+    if spec is None:
+        raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
+    module = importlib.util.module_from_spec(spec)
+    if module is None:
+        raise ImportError(f"Cannot load module {module_name} from spec")
+    sys.modules[module_name] = module
+    spec.loader.exec_module(module)  # type: ignore
+    return module
+
+
+globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))