mtfyun/ds4
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
4f806387f71d3f0e8517650d5cd13b559003cdfd
ds4/cuda/mmq/test
T
History
Salvatore Sanfilippo 45f54e9d11 Use aligned expert vector kernels for small Spark batches 
Reuse the existing fused IQ2 gate/up and Q2 down paths for up to eight resident DeepSeek rows on GB10. Keep GLM, streaming and multi-GPU dispatch unchanged. Test overlapping experts, distinct gate/up weights, invalid IDs and exact agreement with separate rows; CUDA memory, race and synchronization checks pass.
2026-09-06 00:40:24 +02:00
..
d2r_stubs.cu
bench: finalize gfx1151 prefill result
2026-08-29 21:51:45 +01:00
iq2_host_tables.h
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_gemm_dense_q8_d2r.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_gemm_lib.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_indexer_score.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_iq2_aligned.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_kv_reencode_idem.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_m2_comp.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_m2_hc.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_m2_q2k.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_m2_qkv.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_m2_router.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_mm_ids.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_q8_aligned.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
proto_q8_warp8.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
test_f16_gemm_lt.cu
cuda: vendored mmq prefill tier (raw entries) from batched-serving fork
2026-08-02 00:47:10 +02:00
test_iq2_aligned_entry.cu
Use aligned expert vector kernels for small Spark batches
2026-09-06 00:40:24 +02:00
test_mmq_parity.cu
Fix ROCm MMQ scratch lifetime and tail reads
2026-09-01 03:26:23 +02:00
test_mmq_soa_tiles.cu
rocm: add gfx1151 MMQ kernel lab
2026-08-29 21:51:44 +01:00