Commit Graph

Select branches

Hide Pull Requests

abstractions_4

add_local_buffer

after_qol

algebraic_upat

all_bv_no_idx

all_sink

amd_isel

amd_refac

amd_smu_reset_patch

amd_sqtt

amd_uop

amd_uop_cleanups

assign_copy

assign_try_3

before2

better_spec

better_syn

bitcast_spec

broadcast_shape_2

bt_ei

buf_heur

bump_nv_2404

cache_reshape

call_inline

call_is_kernels

callify

cast_vec_as_cast

cat_mop

changes_from_rangify

check_clsetarg

check_dims

cifar_30

cifar_30_clean

clean_load

clean_spec

clone_tg

codegen_try_2

contig_on_after

continue_mat

deepsek

default_late_allreduce

delete_kernel_py

delete_ones

delete_slow_rangeify

delta_attention

delta_attention_scan

demote_op

disable_rules

dsp_search

dsp_search_merged

dtype_shape

earlier_gater

emu_cdna

emulate_ctx

end_all_ranges

endrange

er_prioity

execution_unit

failing_assign

fancy_usb

fast_llm

fix_fa_fwd_clang_22

fix_fuse

fix_llvm_cast

fix_range_merging_sd

fix_rangeify_tests

fix_rc

fix_sched_copy

fix_slice_store

fix_src_spec

fixup_fix_kernel

full_spec

fun_w_egraphs

gate-on-load-store

gate_rewrite

gemma4_gpt

gen_pdf_fast

gh-pages

gl_dims

gldims

glm_flash

grad_outerworld

grok_x86_novec

group_div_rules

hcq2_vram_copies

heur_4096

heur_group

hk_gemm

http_threading

hwtest_fixes

id_store

if_range_cleanups

ignore_oob

image_idx_xy

image_in_mc

image_no_vec

image_refactors_stack

index_to_slice

invalid_try_3

jinja_template

kernel4

kernel_is_call

kernelless

kimi_slop

kitten_matmul

late_Locals

late_add_load

late_allreduce

late_codegen

late_decomp_is_late

late_lower_idx

late_view

late_weakint

lds_double_buffering

less_wmma_arg

lil_dg_change

lil_rangeify_td

llama_trainer

llm_262k_minimal

llm_flash_attn

llm_qwen

llm_serve_fixes

load_view_late

local_cache

locals_have_device

loop_end_op

loop_is_rng

lowerer_hacks

mac_pytest

manual_kernel_3

master

mega_lowerer

min_new_lin

min_outer_range

minor_views

mixins7

mlperf_training_v4.1

mlperf_training_v5.0

mlperf_training_v6.0

modern_scan

more_apps

more_early_comps

more_folding_rules

more_kernelcount

more_pcode

more_rangeify_pm_tests

more_ren_clean

more_uop_gc_test

more_uop_mm

more_vmap

move_devec

move_dtype_decomps

move_expander

move_gates_to_load_store

move_mop_cleanup

move_schedule

move_simp_views

moveleftright

mt_typing

multi_range

multioutput

new_ck

new_devec

new_invalid_and_image

new_linearizer

new_shape

new_test

no_assign_2

no_buf_0

no_decimals

no_late_simplify_marg

no_merge_ends

no_merge_views

no_mstack_in_allreduce

no_one_ranges

no_sip_usbgpu

no_tsink_base

no_tuplize

no_user_ck_contig

no_vec_x86_4

no_weakint_variable

noop_index

novecimg

num_batches_tracked_shape

olmo3_support

one_compiler

one_hcopt

only_assign_buffers

only_loop_backends

only_reg_emu

only_reg_emu2

outer_range

outerworld_work

pad_invalid_work

pad_mixin

param_call_is_function

parse_sqtt_more

pci_iommu

port_nir

postrange_hcopts

precompile_backward

prefill_qwen_fast

print_special_ops

ps_clean

pyremu_try

python_prg

python_speed

q4k_support

qcom-openpilot-13ms

qcom_830

qcom_mmapeak

qcom_speed_codex

qol_string

qwen36_27b_amd_900

qwen36_27b_amd_minimal

qwen36_amd_bugfixes

qwen_mergable

r_cleanups

r_new_opt

ranged_call

rangeify_2

rangeify_const_folding

rdna3_vibes

rdna4_asm

rdna4_gemm

real_pyrender

recurse_sub

reg_coalese

relu_pattern

remove_assign

remove_cast_hack

remove_ck_contig

remove_define_var

remove_forced_re

remove_on_stack

remove_programspec

remove_pylint

remove_slow_tests

remove_vec_2

remove_vec_3

renderers_use_shapes_2

replace_if_with_range

reprocess_node_experiments

reshape_trait

resnet-ci

revert-11547-fix_mismatch_red

revert-11715-revert-11711-amd_llvm_tc_cast

revert-11809-no_ki_in_gpudims

revert-11811-better_div_nesting

revert-11876-memplan_vbig

revert-12104-simplify_valid_in_symbolic

revert-12308-bufferize_early

revert-12731-jitlink_init_check

revert-12813-after_cleanups

revert-13046-better_cleanup_arange_buffers

revert-13311-prof_link_sqtt

revert-13583-rocr_desc_2

revert-14734-tload

revert-17141-full_rewrite_to_sink

revert-17408-_broadcasted-weak

revert-9438-simple_matmul_np_rand

revert_wrong_ck_fix

rewrite_rangeify2

rtoposort

scan_assign

scan_op_8

sched_cache

sched_lin

sd_load_simple

self_type

set_testpath

shrink_in_render

simp_spec

simple_gidxs

simple_pool

simple_priority

simple_rangeify

simpler-sd-compile

simpler_end

simpler_fusion

simpler_image

simpler_postrange

simpler_reduce

slice_like_shrink

spec_3

spec_cleanup

spec_ren

speed_one_linear

split_buf_idx

split_decomp_pass

split_sub

stack_0

store_is_void

string_ranges

support_opts_in_contig

swizzle_in_kernel

sym_llm

sym_work

symbolic-chunk-split

szdiff_bot_test

tc_is_heur

tensor_graph_single_sink

test_amd_crash

test_fa

test_from_postopt

test_hgemm

test_unit

tiny_scan

tmr

tool_call_support

topoprop

ttomsa-new_linearizer

tuplegtuple

two_stage_remove

typed_checks

typing_work

unused_permute_arg

uop_prg

upd_test4

update_benchmark

update_benchmark_search

update_mlperf

user_failure

vecless_x86_glm

verify_sink_dims_clean

vf_ish

view_after_const

view_in_codegen

viz_cbroadcast

warp_fun

wgpu-f16-packed

working_postopt

x86_moves

x86_numel

v0.10.0

v0.10.1

v0.10.2

v0.10.3

v0.11.0

v0.12.0

v0.13.0

v0.4.0

v0.5.0

v0.6.0

v0.7.0

v0.8.0

v0.9.0

v0.9.1

v0.9.2

Mono Color

• b1060ca708 don't promote dtype in _pad_constant [pr] (#17125) chenyuandGitHub 2026-07-21 18:13:50 -04:00
• b1a72299ab more _broadcast_to cleanup [PR] (#17124) chenyuandGitHub 2026-07-21 17:46:04 -04:00
• 8fa5993923 replace pytest-timeout with SIGABRT hook (#17122) sirhcmandGitHub 2026-07-21 17:42:58 -04:00
• f41e4a758f drop Ops.WAIT (#17121) nimlgenandGitHub 2026-07-22 00:42:45 +03:00
• f19a2ad771 single where mixin [pr] (#17118) chenyuandGitHub 2026-07-21 17:38:36 -04:00
• 787b2f2db2 hcq2: use ins for hcq ir (#17120) nimlgenandGitHub 2026-07-21 23:42:34 +03:00
• ef37830d13 allow freeing buffers when pickling/unpickling (#16799) Armand du Parc LocmariaandGitHub 2026-07-21 13:11:23 -07:00
• 5244d3cd2a fix test_u32_to_f16 (#17119) chenyuandGitHub 2026-07-21 16:06:05 -04:00
• b764599d87 add Ops.LOOP + conditional Ops.END (kimi) (#17117) 3 people 2026-07-21 22:54:14 +03:00
• ab917666f0 llm: fix expert routing race and improve locality geohot 2026-07-21 18:12:09 +00:00
• ac20b5e984 add Ops.LOOP + conditional Ops.END (kimi) loop_end_op geohot 2026-07-21 10:50:38 -07:00
• 46b82d4755 don't auto cast cond for WHERE (#17115) chenyuandGitHub 2026-07-21 13:00:07 -04:00
• 76dade5a11 implicit broadcast gradient based on shape only [pr] (#17114) chenyuandGitHub 2026-07-21 12:51:54 -04:00
• 150d5a4aea llm: increase recurrent prefill batch geohot 2026-07-21 16:35:04 +00:00
• e69df9e8b6 llm: parallelize long decode reduction geohot 2026-07-21 15:59:46 +00:00
• f64f96ec59 broadcast_axes [PR] (#17112) chenyuandGitHub 2026-07-21 11:53:12 -04:00
• 7cb77733d8 llm: fix short flash decode wave merge geohot 2026-07-21 15:22:42 +00:00
• f5c9330441 llm: reduce tail decode softmax overhead geohot 2026-07-21 10:58:54 +00:00
• 8c2d598285 llm: prewarm decode expert route inputs geohot 2026-07-21 10:39:51 +00:00
• 10df19afc3 hcq: pass only compiled kernel globals geohot 2026-07-21 10:39:51 +00:00
• a964134597 llm: speed decode at long context geohot 2026-07-21 09:12:27 +00:00
• 2357c1e955 llm: precompute expert route indices geohot 2026-07-21 08:59:03 +00:00
• 3d2636e592 llm: reduce tail decode attention overhead geohot 2026-07-21 08:02:54 +00:00
• 5d360f1bea llm: use two waves for IQ3 prefill geohot 2026-07-21 07:17:16 +00:00
• d5793c85bd llm: stabilize flash decode wave merge geohot 2026-07-21 06:03:59 +00:00
• 711308bc41 llm: preserve prompt checkpoints and speed long decode geohot 2026-07-21 05:39:17 +00:00
• 7b05caf5c5 viz: do not crash on sym_infer err (#17109) qazalandGitHub 2026-07-21 12:04:49 +08:00
• 34bcc5ad63 logcumsumexp mask is bool (#17108) chenyuandGitHub 2026-07-20 22:42:58 -04:00
• 40f0d4af14 clean up _broadcasted [PR] (#16974) chenyuandGitHub 2026-07-20 22:28:19 -04:00
• 2864036e8e correct more spelling of coalesce (#17107) chenyuandGitHub 2026-07-20 21:55:37 -04:00
• f3a5337825 correct spelling of coalesce (#17103) sirhcmandGitHub 2026-07-20 21:44:04 -04:00
• 95f5c85bf3 some realize and corealize for slow tests (#17106) chenyuandGitHub 2026-07-20 21:43:23 -04:00
• 2a81616492 update rules for INDEX mops (#17105) George HotzandGitHub 2026-07-20 18:43:00 -07:00
• 83ba9dd90f llm: finish one GPU serving optimizations geohot 2026-07-21 00:30:21 +00:00
• 13ca9bd8a6 remove dtypes.index again (#17104) chenyuandGitHub 2026-07-20 20:30:04 -04:00
• 6eb1379f01 Merge branch 'master' into modern_scan modern_scan George HotzandGitHub 2026-07-20 17:26:28 -07:00
• 636a43722d add END and GROUP to addrspace (#17102) George HotzandGitHub 2026-07-20 17:26:02 -07:00
• 980748ccfc add multiple_of to ParamArg (#17101) sirhcmandGitHub 2026-07-20 20:11:54 -04:00
• 8fa5b55b5e experiment with new scans geohot 2026-07-20 16:22:09 -07:00
• 6b28df0f0e llm: reuse recurrent state dot products geohot 2026-07-20 16:45:40 +00:00
• d00bdb6790 llm: avoid max context prompt allocation geohot 2026-07-20 14:51:33 +00:00
• 317a6b0a3e llm: speed up Qwen3.6 serving geohot 2026-07-20 05:06:59 +00:00
• f7ce7f330d llm: minor fixes + tests (#17099) George HotzandGitHub 2026-07-20 14:31:20 -07:00
• 02bb555162 error llm_serve_fixes geohot 2026-07-20 14:26:07 -07:00
• 1156623444 llm: minor fixes + tests geohot 2026-07-20 14:21:30 -07:00
• 4b8db13e01 rdna int8 wmma (#17098) chenyuandGitHub 2026-07-20 17:00:50 -04:00
• b1cbd1a43f pytest: use timeout_method signal (#17094) sirhcmandGitHub 2026-07-20 15:19:24 -04:00
• dbb0f6067e clean up ALU rules in spec.py (#17095) chenyuandGitHub 2026-07-20 15:18:48 -04:00
• 8481eba866 allow-unsafe-pr-checkout for szdiff.yml (#17096) chenyuandGitHub 2026-07-20 15:08:54 -04:00
• 2b96d64496 hcq2: tiny opts and fixes (#17092) nimlgenandGitHub 2026-07-20 18:46:52 +03:00
• ef77963cfd derivative of logsumexp is independent of max (#17088) Pol Puigdemont PlanaandGitHub 2026-07-20 15:52:16 +02:00
• abba2aebda llama: correct fused qkv shape assert (#17086) qazalandGitHub 2026-07-20 14:51:21 +08:00
• 1cf8f2f68c llama: inplace amax update (#17064) qazalandGitHub 2026-07-20 14:05:41 +08:00
• 6b98cb351a minor bugfixes llm_flash_attn geohot 2026-07-20 04:47:52 +00:00
• d8cea461e4 1661/155 geohot 2026-07-19 23:26:02 +00:00
• ac3f56a1a2 more shift tests (#17083) chenyuandGitHub 2026-07-19 16:05:13 -04:00
• 4d50b92812 fast and usable geohot 2026-07-19 17:20:39 +00:00
• 89117d8b9e use real shift in l2i decomp [pr] (#17080) chenyuandGitHub 2026-07-19 13:15:06 -04:00
• 9970a0aad0 fix Tensor << Tensor for x86 (#17082) chenyuandGitHub 2026-07-19 12:31:05 -04:00
• 093012c610 faster geohot 2026-07-19 14:32:17 +00:00
• 0146a30125 improve cast to unsign min_max [pr] (#17078) chenyuandGitHub 2026-07-18 21:58:41 -04:00
• b53cd35cff llm: make tokenizer fast (kimi) (#17077) George HotzandGitHub 2026-07-18 17:31:59 -07:00
• 82debb4557 only allow x86_64 target arch on X86Renderer (#17076) Rick WierengaandGitHub 2026-07-18 16:51:01 -07:00
• 2008e44840 qwen 3.6 geohot 2026-07-18 15:07:20 +00:00
• ee290b3e39 optim: mxfp8 zero 1 allgathers in fp8 (#17073) wozeparrotandGitHub 2026-07-18 07:44:50 -07:00
• 232529ce88 hcq2: simpler sync (#17069) nimlgenandGitHub 2026-07-18 16:27:44 +03:00
• 24d8681be7 viz: better sidebar collapse ux (#17072) qazalandGitHub 2026-07-18 17:07:58 +08:00
• 47629f4bcf more weak dtype materialization raise (#17071) chenyuandGitHub 2026-07-17 23:15:14 -04:00
• 7a95708926 optimize openpilot vision on qcom qcom-openpilot-13ms geohot 2026-07-18 00:48:58 +00:00
• 10d207a6f4 work qcom_speed_codex geohot 2026-07-18 00:43:44 +00:00
• ba168bd79e tokenizer cache geohot 2026-07-17 20:15:30 +00:00
• c2837daaea slow tokenizer geohot 2026-07-17 20:10:20 +00:00
• f315df29a0 no weak Tensor from and to real buffer (#17067) chenyuandGitHub 2026-07-17 16:09:10 -04:00
• 81148c7a37 real speed geohot 2026-07-17 19:58:54 +00:00
• 7151a8ad9a bugfix geohot 2026-07-17 19:02:59 +00:00
• 8a5fdf1e67 add flash attention to llm (codex slop) geohot 2026-07-17 18:46:34 +00:00
• 86a6ad8ed2 llm: split cli.py into serve.py with the HTTP server (#17065) George HotzandGitHub 2026-07-17 10:45:37 -07:00
• 3ee2baf71d llm: add tool calling support (kimi) (#17061) George HotzandGitHub 2026-07-17 10:20:01 -07:00
• 7dd3422c63 llama: replace two stage amax with atomics (#17063) qazalandGitHub 2026-07-17 18:27:10 +08:00
• a836c3822a gptoss: 3d mx block scale (#17062) wozeparrotandGitHub 2026-07-16 23:30:24 -07:00
• 6f1176ea90 benchmarks: test usbgpu copy speeds on comma (#17060) sirhcmandGitHub 2026-07-17 02:02:21 -04:00
• b162ab15da pad invalid work (glm) pad_invalid_work geohot 2026-07-16 19:05:49 -07:00
• 46172bb7c7 llm: add optional jinja template support (kimi) (#17058) George HotzandGitHub 2026-07-16 19:02:58 -07:00
• 6d8e8b6468 more crap to fallback jinja_template geohot 2026-07-16 18:56:01 -07:00
• 9a36e54708 lil geohot 2026-07-16 18:52:24 -07:00
• 551cc76f63 fix tests geohot 2026-07-16 18:47:26 -07:00
• 0275e7c0f0 add jinja template support (kimi) geohot 2026-07-16 18:39:36 -07:00
• 7b19732a7a gpt fixes tool_call_support geohot 2026-07-17 00:55:50 +00:00
• ee19fd0b6a fixes geohot 2026-07-16 17:45:26 -07:00
• ce5ae31f5d work geohot 2026-07-16 17:30:14 -07:00
• 22bdc7b6c2 rm that geohot 2026-07-16 17:18:34 -07:00
• e13e6b3752 add tool calling support to llm geohot 2026-07-16 17:09:35 -07:00
• 88826a6f35 no weak dtype for randn_like either (#17055) chenyuandGitHub 2026-07-16 18:29:12 -04:00
• 3bfd62e915 fix 0 size tolist to match numpy (#17054) chenyuandGitHub 2026-07-16 17:42:52 -04:00
• 709babb97c system: remove sibling functions of PCIDevice (#17052) nimlgenandGitHub 2026-07-17 00:15:32 +03:00
• d8b83daac6 set tc_upcast_axes to None when done with it (#17053) George HotzandGitHub 2026-07-16 14:15:21 -07:00
• c74149c973 avoid repeated parsing and toposort in _valid_priority [PR] (#17049) stylishvoidandGitHub 2026-07-16 22:24:08 +02:00
• 6fa0b2b19e materialize weak dtype casts to default (#17051) chenyuandGitHub 2026-07-16 16:12:33 -04:00
• 4d8c3d3fc9 add test_hgemm to test_tiny (#17050) George HotzandGitHub 2026-07-16 13:12:10 -07:00
• 96c71378da dsp skip test_hgemm geohot 2026-07-16 13:05:55 -07:00