Closed ylc2001 closed 10 months ago
Are there any docs elaborating on the model building process in https://github.com/mlc-ai/mlc-llm/blob/main/build.py ?
I'm learning about tvm these days and I'm courious about what exactly did you do to transform & build the model in build.py. Especially the part where tir functions are optimized for different targets in function build():
if target_kind != "cpu": db = utils.get_database(args.db_path) # pylint: disable=invalid-name with db, tvm.target.Target("apple/m1-gpu-restricted"): if args.target_kind == "android": mod_deploy = mlc_llm.dispatch.DispatchTIROperatorAdreno()( # pylint: disable=not-callable mod_deploy ) mod_deploy = relax.transform.MetaScheduleApplyDatabase()(mod_deploy) mod_deploy = ( mlc_llm.dispatch.DispatchTIROperator( # pylint: disable=not-callable args.model_category )(mod_deploy) ) mod_deploy = tvm.tir.transform.DefaultGPUSchedule()(mod_deploy) mod_deploy = mlc_llm.transform.LiftTIRGlobalBufferAlloc()(mod_deploy) mod_deploy = tvm.tir.transform.ForceNarrowIndexToInt32()(mod_deploy)
This looks quite different from the mlc course https://mlc.ai/
same
It is now available at llm.mlc.ai
❓ General Question
Are there any docs elaborating on the model building process in https://github.com/mlc-ai/mlc-llm/blob/main/build.py ?
I'm learning about tvm these days and I'm courious about what exactly did you do to transform & build the model in build.py. Especially the part where tir functions are optimized for different targets in function build():
This looks quite different from the mlc course https://mlc.ai/