up

Author: metascroy

examples/models/llama/export_llama_lib.py

@@ -353,8 +353,8 @@ def build_args_parser() -> argparse.ArgumentParser:
     parser.add_argument(
         "--coreml-quantize",
         default=None,
-        choices=["b4w"],
-        help="This option is only for coreml: Use coreml quantization, e.g. b4w (for blockwise 4 bit weight)",
+        choices=["b4w", "c4w"],
+        help="This option is only for coreml: Use coreml quantization, e.g. b4w (for blockwise 4 bit weight), c4w (for channelwise 4 bit weight)",
     )
     parser.add_argument(
         "--coreml-ios",

extension/llm/export/partitioner_lib.py

@@ -152,7 +152,7 @@ def _validate_ios_version() -> None:
     compile_specs = CoreMLBackend.generate_compile_specs(  # pyre-fixme[16]
         minimum_deployment_target=minimum_deployment_target,
         compute_precision=ct.precision(ct.precision.FLOAT16.value),
-        compute_units=coreml_compute_units,
+        compute_unit=coreml_compute_units,
         model_type=CoreMLBackend.MODEL_TYPE.MODEL,  # pyre-fixme[16]
         op_linear_quantizer_config=op_linear_quantizer_config,
     )