AstrAI

803 Commits 1 Branch 14 Tags

3 Commits

Include renames

Author	SHA1	Message	Date
ViperEkura	bf239d194c	refactor: dedupe fp8 kernel helpers and trim comments - merge the quantize launchers into one Tiled template; extract shared cvt_fp8/publish_amax helpers and replace the dtype x format ladder with two-level template dispatch - fold the gemm interior/generic operand loads into one kInterior template and the fast/generic async loads into load_async<kFast>; Policy carries the smem budget - compress kernel comments to the load-bearing invariants, dropping measured-number essays; Policy signature and kernel code unchanged Benchmark: NVIDIA L20, 1.2B model train step fwd+bwd+CE - M=8192: fp8 532.2 -> 530.4 ms (1.26x, noise); tests/extension 65 passed, quantize layouts byte-exact, NT routing diff 0.0	2026-08-28 16:45:21 +08:00
ViperEkura	8a353117ea	perf: transpose-quantize backward operands to route all gemms nt - quantize gains out_layout (0 row-major / 1 transposed / 2 single-read dual-write); modes 1/2 run a new 32x32 smem-tile transpose kernel - backward feeds g8/w8T and g8T/x8T to trans_b=True gemms, dropping the NN-swap and TT crosswise kernels from training; fp8 weights keep the swap fallback - a 64x64 tile variant tied on the real step mix and was reverted; noted in the kernel header Benchmark: NVIDIA L20, 1.2B model, full train step fwd+bwd+CE - M=8192: fp8 551.8 -> 532.2 ms, 1.21x -> 1.26x vs bf16; M=2048 0.90x -> 0.95x - kernel-level grad_x +3.7..12.4%, grad_w +13.8..20.8%; layouts byte-exact, fp8 tests 36/36	2026-08-28 16:12:15 +08:00
ViperEkura	057c0d33df	refactor: split quantize into multi-type primitive - split quantize into quantize.cuh, templated on input type (bf16/fp16/fp32) - rename pybind entry quantize_bf16 to quantize; validate the fmt enum - fix fp8x2 packing: one 32-bit word packs two pairs (halves were dropped) - drop the dead OutFp8 template param; GEMM output is always bf16 - fp8_state.reset() restores recipe/format defaults too (test state leak) - rewrite tests for the two-primitive API with fp32-domain amax references	2026-08-25 20:07:40 +08:00

Author

SHA1

Message

Date

ViperEkura

bf239d194c

refactor: dedupe fp8 kernel helpers and trim comments

- merge the quantize launchers into one Tiled template; extract shared cvt_fp8/publish_amax helpers and replace the dtype x format ladder with two-level template dispatch
- fold the gemm interior/generic operand loads into one kInterior template and the fast/generic async loads into load_async<kFast>; Policy carries the smem budget
- compress kernel comments to the load-bearing invariants, dropping measured-number essays; Policy signature and kernel code unchanged

Benchmark: NVIDIA L20, 1.2B model train step fwd+bwd+CE
- M=8192: fp8 532.2 -> 530.4 ms (1.26x, noise); tests/extension 65 passed, quantize layouts byte-exact, NT routing diff 0.0

2026-08-28 16:45:21 +08:00

ViperEkura

8a353117ea

perf: transpose-quantize backward operands to route all gemms nt

- quantize gains out_layout (0 row-major / 1 transposed / 2 single-read dual-write); modes 1/2 run a new 32x32 smem-tile transpose kernel
- backward feeds g8/w8T and g8T/x8T to trans_b=True gemms, dropping the NN-swap and TT crosswise kernels from training; fp8 weights keep the swap fallback
- a 64x64 tile variant tied on the real step mix and was reverted; noted in the kernel header

Benchmark: NVIDIA L20, 1.2B model, full train step fwd+bwd+CE
- M=8192: fp8 551.8 -> 532.2 ms, 1.21x -> 1.26x vs bf16; M=2048 0.90x -> 0.95x
- kernel-level grad_x +3.7..12.4%, grad_w +13.8..20.8%; layouts byte-exact, fp8 tests 36/36

2026-08-28 16:12:15 +08:00

ViperEkura

057c0d33df

refactor: split quantize into multi-type primitive

- split quantize into quantize.cuh, templated on input type (bf16/fp16/fp32)
- rename pybind entry quantize_bf16 to quantize; validate the fmt enum
- fix fp8x2 packing: one 32-bit word packs two pairs (halves were dropped)
- drop the dead OutFp8 template param; GEMM output is always bf16
- fp8_state.reset() restores recipe/format defaults too (test state leak)
- rewrite tests for the two-primitive API with fp32-domain amax references

2026-08-25 20:07:40 +08:00