1,201 commits

Author	SHA1	Message	Date
chenyu	4dbe6a2ee7	remove _force_unique from Tensor init (#16277)	2026-05-20 14:13:05 -04:00
chenyu	890b731b1e	more prerequisuite test changed for deviceless const (#16264)	2026-05-19 15:43:45 -04:00
ttomsa	aa1e59ab97	X86 with Ops.INS (#14873) ... * draft * cleanup test_encodings * cleanup test_isel * model flag state and support rematerialization * woops * add vbroadcastss instruction * don't fuse load if used multiple times in src * add movabs instruction and fix idiv * fixes * add x86 backend to tests * float16 fix * rm TwoAddress2nd * add BARRIER * test windows ci * yup isel fixes the mask stuff too and its beautiful * add cmoves to the spec * support storing imms * no TUPLE_ORDER, breaks tests * fix remaining seg faults * add float max * always fuse index * minor * fix DEFINE_VAR/SPECIAL and enable multithreading * linter * more linter * more * more * more * let's try this * perhaps * start new scheduler * more scheduling info * cleaner shuffle functions * fixup isel tests * skip bounds check when NOOPs exist * skip inf rewrite tests * fix const tag hack and add x86ops to _shape * fix * skip a few tests * func arg order independent from op value * x86 goes in own linearize * switch to PARAM * more * add min x86op and neg in decomps * do mulacc in isel * use def_reg in test_encodings * enable emulated int64 tests * how much does this fix * Ops becomes OpType * fix * rm noqa * rm machine scheduler stuff * and this * allow for extending enums and move X86Ops out of uop * fix imports * rm X86GroupOp from ops.py * spacing * tell mypy to shut up * more linter * add x86op test * allow set[X86Ops] in upat * move NOOPs to pre_isel_matcher and rm NOOP from spec * more asserts * also this * cleanup encode * simplify live range * fix idiv * add Ops.INS to x86 * more changes * more changes * more changes * fix * fix * fix * fix * print formatted assembly * fix 8bit idiv? * oops * enable float16 and unaligned vector load/store * actually no * move x86 tests * no more bool cast * fix * linter * linter * move X86Ops to x86.py * fix vpbroadcast * cleanups * linter * print correct reg names * canonical max * move max/min and add test * support float16 vector load/store * rm bad rewrite * vpsrldq can't access memory * regalloc takes renderer * enable vector load/store on all dtypes * more isel tests * rm this for now * a lot better * fix * fix * fix * deal with flags correctly * fix * enable gep noop rule * fix * fix * fix * add callee saved registers * use Ops.CONST instead of X86Ops.IMM * fix * enable TUPLE_ORDER * fix * rm x86 code in linearizer * fix * fix * fix * move isa rewrites to codegen * fix * fix * skip test_linearizer.py * skip more tests * fix * fix for idiv/mod changes * fix * don't use fmadd if it duplicates fused op * hacky * fix * cleanups * cleanups * fix --------- Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>	2026-05-19 12:42:54 -07:00
George Hotz	3f2d401464	all tests pass with NOOPT=1 (#16257) ... * all tests pass with NOOPT=1 * fix a few more * noopt 100% pass * noopt 100% pass	2026-05-18 20:39:51 -07:00
chenyu	754344087a	assign for deviceless const source (#16248)	2026-05-18 17:39:53 -04:00
chenyu	dcee90aa3f	remove requires_grad use in extra/examples (#16238) ... except the ones fed into optimizer	2026-05-16 18:40:26 -04:00
chenyu	8631b6f17d	remove use of requires_grad in test/ (#16237)	2026-05-16 17:21:07 -04:00
chenyu	0ddc50d050	do not gate backward on requires_grad (#16230) ... DETACH is filtered in _deepwalk. instead of None, it gets 0 grad now	2026-05-16 12:29:49 -04:00
chenyu	07a172dbbb	remove noop requires_grad_ calls (#16213)	2026-05-15 13:31:10 -04:00
chenyu	c6cf9e8f0c	remove test_svd_nonfull_5_5 (#16217) ... flaky, kinda overlap with test_svd_general	2026-05-15 13:10:02 -04:00
chenyu	409bb0c9ad	requires_grad cannot be None (#16212) ... final goal is to remove requires_grad, first change the default to True, and don't allow None	2026-05-15 02:01:04 -04:00
chenyu	a75c14f010	some setitem tests (#16209)	2026-05-14 22:36:25 -04:00
chenyu	ffa1aac7b1	gradient for STORE/AFTER ala clone (#16205)	2026-05-14 20:17:27 -04:00
chenyu	09096ea565	test_gradient_through_clone (#16203) ... backward through clone crashes now	2026-05-14 19:26:47 -04:00
b1tg	3c806ff406	clean up gguf (#16160)	2026-05-12 21:16:10 -07:00
chenyu	38d407fd58	simplify svd more (#16181) ... all the slowness is scheduling	2026-05-12 23:48:22 -04:00
chenyu	2172363be5	don't use Tensor indexing in svd (#16174) ... prepare mixin, also about 4X faster for 8x8 input	2026-05-12 21:56:19 -04:00
wozeparrot	a613bcfc6d	allow after on contiguous in spec (#16169) ... * feat: allow after on contiguous * feat: add test	2026-05-12 13:11:44 -07:00
chenyu	da3b7e89a4	atol in test_custom_kernel_multi_output_backward_interacting (#16166)	2026-05-12 14:42:12 -04:00
George Hotz	8294d105a7	Update the spec in spec.py to match the current state (#16132) ... * start work on specv2 * more spec * more spec * fix amd emulator * more spec * more * fix test_uop_graph * move those * spec=2 * skip those questionable tests * ptx fix * more spec=2 * store * allow custom function in tensor * spec 2 * fix beam search for tensor cores * delete the old specs * fix import	2026-05-11 20:07:47 -07:00
chenyu	3942a80f66	fix wrong kwargs passed into rands (#16149) ... working towards explicit args for these	2026-05-11 22:22:06 -04:00
chenyu	63c1f00b80	disable test_svd_general again (#16146) ... flaky on CI	2026-05-11 19:24:32 -04:00
chenyu	fbe8be0b8b	style cleanup to Tensor.qr and svd (#16142) ... * style cleanup to Tensor.qr and svd same kernels * more * enable	2026-05-11 17:16:59 -04:00
wozeparrot	4d1a9dca41	fix: don't copy precompiled custom kernel outputs (#16084)	2026-05-07 14:02:38 -07:00
nimlgen	5fa0016ffc	supports_exec_item -> supports_uop (#16033)	2026-05-05 22:41:13 +03:00
wozeparrot	419d525553	feat: handle multioutput kernel grads (#16028)	2026-05-02 22:31:45 -07:00
George Hotz	5f441ecffc	unify reduce + reduce_axis (#15973) ... * unify reduce + reduce_axis * fix all tests * lil cleanups	2026-04-29 10:29:56 -07:00
nimlgen	4164666c72	programinfo (#15942) ... * programinfo * fix * m * x * x * changes * x * fix * rm	2026-04-27 23:12:03 +03:00
nimlgen	96165ff0d1	validate_with_cpu as rewrite (#15938) ... * validate_with_cpu as rewrite * compil * x * linter * moved * fix	2026-04-26 19:58:53 +03:00
nimlgen	d3378010ee	schedule() -> schedule_linear() in tests (batch 1) (#15915) ... * schedule_with_vars -> linear_with_vars in tests * tests batch 1 * batch 2 * estimate_uop * simpler * rm	2026-04-24 23:40:53 +03:00
b1tg	af93a677ae	llm: glm 4.5 air (#15771) ... * llm: glm 4.5 air * clean * clean * remove gguf_size	2026-04-22 22:47:37 +08:00
Christopher Milan	99a0debd62	Device.count() (#15842)	2026-04-21 16:46:38 -04:00
chenyu	9192c93b7e	Tensor.invalid -> Tesnor.invalids (#15849) ... matches ones and zeros, and to not share name with UOp.invalid	2026-04-21 11:19:51 -04:00
nimlgen	01ac1c8c15	remove all run_schedule from tests (#15846)	2026-04-21 12:02:10 +03:00
Christopher Milan	1a8ba4cbd6	CPU renderers use arch (#15839)	2026-04-20 23:38:29 -04:00
George Hotz	5819c0abed	fix gc in gguf (#15820) ... * fix gc in gguf * fix mypy	2026-04-20 10:15:03 +08:00
George Hotz	67ed4c4eb3	move gguf stuff from nn/state.py to llm/gguf.py (#15783) ... * move gguf stuff from nn/state.py to llm/gguf.py * docs	2026-04-20 09:41:43 +08:00
Kartik Vashishta	a1696e8413	objc: fix _classmethods_ dispatch flag (#14854) ... * objc: fix _classmethods_ dispatch flag * test: add objc _classmethods_ regression	2026-04-20 09:35:03 +08:00
chenyu	5bdfd4883f	update test_assign (#15809) ... clean up old skips and update tests	2026-04-18 21:25:44 -04:00
Christopher Milan	6adf4c3cd9	MOCKGPU interfaces (#15796)	2026-04-17 21:56:29 -04:00
chenyu	8da308573f	update test_assign_changes_alt with clone (#15802)	2026-04-17 20:17:37 -04:00
qazal	9f2a578e26	unskip TestCall.test_call_gemm_uop [pr] (#15786)	2026-04-17 16:18:51 +03:00
George Hotz	e1d13bc4fe	add GGUF IQ4_XS support (#15766) ... * add GGUF IQ4_XS support * gguf 21 * gguf 21 * use plus * ggml_common autogen for constant arrays * fix * ggml_common in autogen * inline	2026-04-17 14:43:39 +08:00
George Hotz	a9b6cfece0	refactor llm into files (#15780) ... * refactor llm into files * chat.html * tokenizer cleanup * cleanup * tests	2026-04-17 12:33:11 +08:00
George Hotz	ec00cefa5b	llm is the only app (#15779) ... * tinygrad/llm is the only app * upd pyproject * claude refs * scoping * min diff	2026-04-17 10:44:48 +08:00
chenyu	f0c12a2004	another form of assign to itself (#15770)	2026-04-16 15:17:19 -04:00
chenyu	d147e2a549	update test_nested_after_contiguous_store (#15763) ... add kernel counts and some TODOs	2026-04-16 09:59:26 -04:00
George Hotz	f57380cbc2	simplify GatedDeltaNetBlock using two state tensors (#15704) ... * test double after * simpler ssm * no double test	2026-04-16 21:14:00 +08:00
George Hotz	d1cce7a476	put the ranges on store instead of after (#15759) ... * put the ranges on store instead of after * better assert * fix stuff * comment out slow rules i don't understand * simpler rule * closer * return false for store * fix loop * only a few schedule failures remain * remove stores to self * all tests pass locally * remove junk * regression test and fix * better test, bump broken torch count * bugfix with regression test * new fusion is better	2026-04-16 19:06:40 +08:00
George Hotz	d24466c844	CALL with return value is FUNCTION (#15758) ... * CALL with return value is FUNCTION (GPT try) * cleanups	2026-04-16 13:25:07 +08:00