Uploaded using `kernel-builder`.

.gitattributes

@@ -59,3 +59,6 @@ build/torch210-cxx11-cpu-x86_64-linux/_quantization_gptq_cpu_e8436b5.abi3.so fil
 build/torch211-cxx11-cpu-x86_64-linux/_quantization_gptq_cpu_e8436b5.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch210-cxx11-cpu-x86_64-linux/_quantization_gptq_cpu_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch211-cxx11-cpu-x86_64-linux/_quantization_gptq_cpu_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch210-cxx11-cpu-x86_64-linux/_quantization_gptq_cpu_1890667.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch211-cxx11-cpu-x86_64-linux/_quantization_gptq_cpu_1890667.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch212-cxx11-cpu-x86_64-linux/_quantization_gptq_cpu_1890667.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_gptq_cpu_86f75d9
-ops = torch.ops._quantization_gptq_cpu_86f75d9
+from . import _quantization_gptq_cpu_1890667
+ops = torch.ops._quantization_gptq_cpu_1890667
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_quantization_gptq_cpu_86f75d9::{op_name}"
+    return f"_quantization_gptq_cpu_1890667::{op_name}"

build/torch210-cxx11-cpu-x86_64-linux/{_quantization_gptq_cpu_86f75d9.abi3.so → _quantization_gptq_cpu_1890667.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba8c854c1421802095068d357ae3ef600ef0d5394603f390704b4d52656a54b7
+oid sha256:0c2ac9bc5f6ea69904ab77c50726ab1dd66338c435aaa41aa65f651d4ecf5d03
 size 1950576

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

@@ -1,6 +1,6 @@
 {
   "name": "quantization-gptq",
-  "id": "_quantization_gptq_cpu_86f75d9",
+  "id": "_quantization_gptq_cpu_1890667",
   "version": 1,
   "license": "MIT",
   "python-depends": [],

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

@@ -1,9 +1,9 @@
 import torch
-from . import _quantization_gptq_cpu_86f75d9
-ops = torch.ops._quantization_gptq_cpu_86f75d9
+from . import _quantization_gptq_cpu_1890667
+ops = torch.ops._quantization_gptq_cpu_1890667
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_quantization_gptq_cpu_86f75d9::{op_name}"
+    return f"_quantization_gptq_cpu_1890667::{op_name}"

build/torch211-cxx11-cpu-x86_64-linux/{_quantization_gptq_cpu_86f75d9.abi3.so → _quantization_gptq_cpu_1890667.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc97f05d868196fe5d339e0b1083b7177e32db58c7b5acaafe17069c44cefca7
+oid sha256:f2657ba7ffa5dee2dc8f244c8581c4c4bed8c1c78e88f096ee943a2f1fdcc176
 size 1950576

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

@@ -1,6 +1,6 @@
 {
   "name": "quantization-gptq",
-  "id": "_quantization_gptq_cpu_86f75d9",
+  "id": "_quantization_gptq_cpu_1890667",
   "version": 1,
   "license": "MIT",
   "python-depends": [],

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

@@ -0,0 +1,3 @@
+from .custom_ops import gemm_int4_forward
+
+__all__ = ["gemm_int4_forward"]

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

@@ -0,0 +1,9 @@
+import torch
+from . import _quantization_gptq_cpu_1890667
+ops = torch.ops._quantization_gptq_cpu_1890667
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_quantization_gptq_cpu_1890667::{op_name}"

build/torch212-cxx11-cpu-x86_64-linux/_quantization_gptq_cpu_1890667.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31aaa63bb5827037929107bacad206fb7d8b69314fc8627e2a98543184095cc2
+size 1950584

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

@@ -0,0 +1,19 @@
+import torch
+from ._ops import ops
+
+def gemm_int4_forward(
+    input: torch.Tensor,
+    weight: torch.Tensor,
+    zeros: torch.Tensor,
+    absmax: torch.Tensor,
+    blocksize: int,
+) -> torch.Tensor:
+    original_dtype = input.dtype
+    if original_dtype != torch.bfloat16:
+        input = input.to(torch.bfloat16)
+
+    output = ops.gemm_int4_forward(input, weight, zeros, absmax, blocksize)
+    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-gptq",
+  "id": "_quantization_gptq_cpu_1890667",
+  "version": 1,
+  "license": "MIT",
+  "python-depends": [],
+  "backend": {
+    "type": "cpu"
+  }
+}

build/torch212-cxx11-cpu-x86_64-linux/quantization_gptq/__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")))