George Hotz 47faa2d7b4 hotfix: llm kv cache uses clone instead of realize to avoid many realize

2026-03-04 19:07:03 +08:00

..

apps

hotfix: llm kv cache uses clone instead of realize to avoid many realize

2026-03-04 19:07:03 +08:00

codegen

simplify idx before counting backward_slice (#15117)

2026-03-03 23:53:50 -05:00

engine

move openpilot contiguous hacks to schedule (#15120)

2026-03-04 03:04:06 -05:00

mixin

move more to mixins (#14780)

2026-02-16 11:35:00 +08:00

nn

add contiguous_view_offset (#15084)

2026-03-02 18:05:04 +08:00

renderer

sqtt: fix rdna4 structs (#15111)

2026-03-03 23:32:14 +09:00

runtime

fix fa forward building with clang 22 (#15124)

2026-03-04 02:32:25 -08:00

schedule

set OPENPILOT_HACKS=1 to enable replace assign (#15123)

2026-03-04 05:26:04 -05:00

uop

allreduce is a function with LATE_ALLREDUCE=1 (#15119)

2026-03-04 15:17:58 +08:00

viz

no after removal (#15102)

2026-03-03 17:50:31 +08:00

__init__.py

start function and add walk rewrite (#14992)

2026-02-25 13:56:27 +08:00

device.py

feat: per device mem_used (#15100)

2026-03-03 01:31:28 -08:00

dtype.py

cleanup fp8 conversion helpers and fp8 edge-case tests (#14953)

2026-02-22 09:16:42 -05:00

function.py

add precompile to call (#15099)

2026-03-03 22:32:42 +08:00

gradient.py

functions for llama trainer (#15045)

2026-02-28 12:15:18 +08:00

helpers.py

set OPENPILOT_HACKS=1 to enable replace assign (#15123)

2026-03-04 05:26:04 -05:00

py.typed

add a single py.typed (#6083)

2024-08-14 17:31:46 -07:00

tensor.py

FLOAT16 logic for IMAGE==1 goes back to image_conv2d (#15105)

2026-03-03 05:37:57 -05:00