compressed-tensors-nightly 0.6.0.20240923__py3-none-any.whl → 0.6.0.20240925__py3-none-any.whl

compressed_tensors/__init__.py

@@ -19,3 +19,4 @@ from .compressors import *
 from .config import *
 from .quantization import QuantizationConfig, QuantizationStatus
 from .utils import *
+from .version import *

compressed_tensors/compressors/model_compressor.py

@@ -22,6 +22,7 @@ from typing import Any, Dict, Optional, Union
 
 import torch
 import transformers
+import compressed_tensors
 from compressed_tensors.base import (
     COMPRESSION_CONFIG_NAME,
     QUANTIZATION_CONFIG_NAME,
@@ -368,6 +369,7 @@ class ModelCompressor:
             config_data[COMPRESSION_CONFIG_NAME][
                 SPARSITY_CONFIG_NAME
             ] = sparsity_config_data
+        config_data[COMPRESSION_CONFIG_NAME]["version"] = compressed_tensors.__version__
 
         with open(config_file_path, "w") as config_file:
             json.dump(config_data, config_file, indent=2, sort_keys=True)

compressed_tensors/quantization/quant_scheme.py

@@ -108,7 +108,7 @@ def is_preset_scheme(name: str) -> bool:
 UNQUANTIZED = dict()
 
 # 8 bit integer weights and 8 bit activations quantization
-W8A8 = dict(
+INT8_W8A8 = dict(
     weights=QuantizationArgs(
         num_bits=8,
         type=QuantizationType.INT,
@@ -149,7 +149,7 @@ W4A16 = dict(
 )
 
 # 4 bit integer weights and 8 bit activations quantization
-W4A8 = dict(
+INT8_W4A8 = dict(
     weights=QuantizationArgs(
         num_bits=4,
         type=QuantizationType.INT,
@@ -210,8 +210,9 @@ PRESET_SCHEMES = {
     "W8A16": W8A16,
     "W4A16": W4A16,
     # Integer weight and activation schemes
-    "W8A8": W8A8,
-    "W4A8": W4A8,
+    "W8A8": INT8_W8A8,
+    "INT8": INT8_W8A8, # alias for W8A8
+    "W4A8": INT8_W4A8,
     # Float weight and activation schemes
     "FP8": FP8,
     "FP8_DYNAMIC": FP8_DYNAMIC,

{compressed_tensors_nightly-0.6.0.20240923.dist-info → compressed_tensors_nightly-0.6.0.20240925.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: compressed-tensors-nightly
-Version: 0.6.0.20240923
+Version: 0.6.0.20240925
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors_nightly-0.6.0.20240923.dist-info → compressed_tensors_nightly-0.6.0.20240925.dist-info}/RECORD

@@ -1,4 +1,4 @@
-compressed_tensors/__init__.py,sha256=SV1csvHUVCd8kHXz6UDZim1HZ_fAVG3vfk-j_4Bb6hY,789
+compressed_tensors/__init__.py,sha256=UtKmifNeBCSE2TZSAfduVNNzHY-3V7bLjZ7n7RuXLOE,812
 compressed_tensors/base.py,sha256=Mq4mfVQcJhNpha-BXzpOfpmFIdl01o09BJE7D2oQ_00,796
 compressed_tensors/version.py,sha256=83tBdwNu2sUhiLPvv6tRNh4Y7u70sZ1TFy3ydWctVL8,1586
 compressed_tensors/compressors/__init__.py,sha256=wmX4VnkUTS63xBwK5-6w8FP78bNZpcdcqvf2KOEC5E4,1133
@@ -6,7 +6,7 @@ compressed_tensors/compressors/base.py,sha256=NfVkhq6PRiq2cvAXaUXLoqC_nVYWdSrkE1
 compressed_tensors/compressors/dense.py,sha256=xcWECjcRY4INN6jC7vHx5wvUX3NmnKlxA9SVE1A6m2Q,1267
 compressed_tensors/compressors/helpers.py,sha256=k9avlkmeYj6vkOAvl-MgcixtP7ib24SCfhzZ-RusXfw,5403
 compressed_tensors/compressors/marlin_24.py,sha256=e7fGUyZbjUpA5VUMCPxqcYPGNiwoDKupHJaXWCoVKRw,9410
-compressed_tensors/compressors/model_compressor.py,sha256=gI6KKtH3eeWi2540Ayx-4bg9o8qjrvxlF4Gd_sqltGA,16678
+compressed_tensors/compressors/model_compressor.py,sha256=Wq-NbjtaVOEElDpcjEYun6QFvAIZee8ZAw_wbifuTDA,16793
 compressed_tensors/compressors/naive_quantized.py,sha256=z3h3ca5xKCN69mahutxcbzdv-OysiaxaM8P-Qum6zUQ,4823
 compressed_tensors/compressors/pack_quantized.py,sha256=27RVmJ2wg2dvCoawj407HSmKT3VPGJ6ujAMHlT26WlI,7571
 compressed_tensors/compressors/sparse_bitmask.py,sha256=kiDwBlFV0sJGLcIdDYxIiuF64ccgwDfqq1hWRQThYDc,8647
@@ -19,7 +19,7 @@ compressed_tensors/linear/compressed_linear.py,sha256=G0gEFfxLAUsgRcnfSV-PKz1ZBN
 compressed_tensors/quantization/__init__.py,sha256=83J5bPB7PavN2TfCoW7_vEDhfYpm4TDrqYO9vdSQ5bk,760
 compressed_tensors/quantization/quant_args.py,sha256=CmyVtjJeHlqCW-7R5Z7tIw6lXUrzCX6Y9bwgmMxEudY,8069
 compressed_tensors/quantization/quant_config.py,sha256=NpVu8YJ4Xw2pIQW_PGaNaml8kx1bUnxkvb0jBYWbKdE,9971
-compressed_tensors/quantization/quant_scheme.py,sha256=HmR1DcFZcjuqX7KHUYI0NFXsCIzJ8sxFGH6zhYuHmEs,5870
+compressed_tensors/quantization/quant_scheme.py,sha256=2ITawuNf76E1CDYBWrfpMP8tyZFykzwU99-eD-WggsM,5930
 compressed_tensors/quantization/lifecycle/__init__.py,sha256=MXE2E7GfIfRRfhrdGy2Og3AZOz5N59B0ZGFcsD89y6c,821
 compressed_tensors/quantization/lifecycle/apply.py,sha256=uftWFunr_CpCZM_qWfo2O1USXKB2qSYD1pBJsO8BuCU,15285
 compressed_tensors/quantization/lifecycle/calibration.py,sha256=PlS_EqCOPqJD3QKuLPXO9AOtDzXtQWvEBTynFv-FFVw,2698
@@ -45,8 +45,8 @@ compressed_tensors/utils/permutations_24.py,sha256=kx6fsfDHebx94zsSzhXGyCyuC9sVy
 compressed_tensors/utils/permute.py,sha256=V6tJLKo3Syccj-viv4F7ZKZgJeCB-hl-dK8RKI_kBwI,2355
 compressed_tensors/utils/safetensors_load.py,sha256=m08ANVuTBxQdoa6LufDgcNJ7wCLDJolyZljB8VEybAU,8578
 compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
-compressed_tensors_nightly-0.6.0.20240923.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors_nightly-0.6.0.20240923.dist-info/METADATA,sha256=on008p2guLeqpkI4kmHhJSX5BtnmbtrnFFRw2fK2Eec,6799
-compressed_tensors_nightly-0.6.0.20240923.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
-compressed_tensors_nightly-0.6.0.20240923.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors_nightly-0.6.0.20240923.dist-info/RECORD,,
+compressed_tensors_nightly-0.6.0.20240925.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors_nightly-0.6.0.20240925.dist-info/METADATA,sha256=AHeC-ko08CtK8_xQUnuNlWNQIhmDcKzDpihAiMBHjR8,6799
+compressed_tensors_nightly-0.6.0.20240925.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
+compressed_tensors_nightly-0.6.0.20240925.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors_nightly-0.6.0.20240925.dist-info/RECORD,,