dragonpilot

mirror of https://github.com/dragonpilot/dragonpilot.git synced 2026-02-22 02:23:53 +08:00

Files

HaraldSchafer cb0b7375b7 Rocket Launcher Model (#25963 )

* 1456d261-d232-4654-8885-4d9fde883894/440 6b7d7cec-ead8-40f3-86cc-86d52c9b03fe/300

* compute only 9 tokens: 1456d261-d232-4654-8885-4d9fde883894/440 6b7d7cec-ead8-40f3-86cc-86d52c9b03fe/300

* tinygrad: cleanup gather

* 1456d261-d232-4654-8885-4d9fde883894/440 6b7d7cec-ead8-40f3-86cc-86d52c9b03fe/700

* empty commit for tests

* bump tinygrad

* dont use tinygrad matmul for now

* bump tinygrad

* 1456d261-d232-4654-8885-4d9fde883894/440 e63ab895-2222-4abd-a9a5-af86bb70e260/700

* float16 1456d261-d232-4654-8885-4d9fde883894/440 e63ab895-2222-4abd-a9a5-af86bb70e260/700

* increase steer rate cost

* Revert "increase steer rate cost"

This reverts commit 74ce9ab9be7ef17ecfec931f96851b12f37f2336.

* fork tinygrad

* empty commit for tests

* basics

* Kinda works

* new lat

* new tuning

* Move LATMPCN so scons compiles

* Update long weights

* Add tinygrad optim

* Update model ref

* update weights

* Update ref

* Try

* Error message for field ignore

* update model regf

* ref commit

* Fix onnx test

Co-authored-by: Yassine Yousfi <yyousfi1@binghamton.edu>

2022-10-05 21:43:38 -07:00

include

thneed saves 45% of a core (#1512 )

2020-05-15 13:53:01 -07:00

__init__.py

model: mse err from 0.02-> 0.000056 (#23891 )

2022-03-02 20:52:17 -08:00

lib.py

Mypy fixes for --check-untyped-defs (#24372 )

2022-05-12 14:08:11 +02:00

README

thneed saves 45% of a core (#1512 )

2020-05-15 13:53:01 -07:00

serialize.cc

modeld: delete unused SNPE stuff after move to tinygrad (#25635 )

2022-09-01 18:28:20 -07:00

thneed_common.cc

Rocket Launcher Model (#25963 )

2022-10-05 21:43:38 -07:00

thneed_pc.cc

modeld: delete unused SNPE stuff after move to tinygrad (#25635 )

2022-09-01 18:28:20 -07:00

thneed_qcom2.cc

modeld: delete unused SNPE stuff after move to tinygrad (#25635 )

2022-09-01 18:28:20 -07:00

thneed.h

modeld: delete unused SNPE stuff after move to tinygrad (#25635 )

2022-09-01 18:28:20 -07:00

README

thneed is an SNPE accelerator. I know SNPE is already an accelerator, but sometimes things need to go even faster..

It runs on the local device, and caches a single model run. Then it replays it, but fast.

thneed slices through abstraction layers like a fish.

You need a thneed.