jwfromm commented on a change in pull request #7823:
URL: https://github.com/apache/tvm/pull/7823#discussion_r615326897
##########
File path: python/tvm/driver/tvmc/autotuner.py
##########
@@ -228,24 +240,137 @@ def drive_tune(args):
args: argparse.Namespace
Arguments from command line parser.
"""
- # extra arguments validation before importing the model, so that obvious
errors
- # are pointed in advance.
- if args.rpc_tracker:
- parsed_url = urlparse("//%s" % args.rpc_tracker)
+ tvmc_model = frontends.load_model(args.FILE, args.model_format,
shape_dict=args.input_shapes)
+ tvmc_model.tuning_records = args.tuning_records
+ # Specify hardware parameters, although they'll only be used if
autoscheduling.
+ hardware_params = auto_scheduler.HardwareParams(
+ args.num_cores,
+ args.vector_unit_bytes,
+ args.cache_line_bytes,
+ args.max_shared_memory_per_block,
+ args.max_local_memory_per_block,
+ args.max_threads_per_block,
+ args.max_vthread_extent,
+ args.warp_size,
+ args.target,
+ args.target_host,
+ )
+
+ tune_model(
Review comment:
Definitely an option for follow up work. I don't want to introduce any
json objects in this PR as it would dramatically change TVMC from the command
line.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
