Skip to content

Actions: ORippler/llama.cpp

Actions

Check Pre-Tokenizer Hashes

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
pre-tokenizer-hashes.yml
28 workflow runs
28 workflow runs

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
CUDA: directly include cuda/iterator Check Pre-Tokenizer Hashes #28: Commit 8d0cd8b pushed by ORippler
1m 4s osimons/cuda_explicit_iterator_import
osimons/cuda_explicit_iterator_import
1m 4s
WIP fastdiv_u64 implementation in CUDA Check Pre-Tokenizer Hashes #27: Commit c93a762 pushed by ORippler
1m 6s osimons/fdiv_u64_poc
osimons/fdiv_u64_poc
1m 6s
CUDA: better coalesce data-access for contiguous concat Check Pre-Tokenizer Hashes #26: Commit a027d4f pushed by ORippler
1m 9s osimons/CUDA_faster_cont_concat
osimons/CUDA_faster_cont_concat
1m 9s
Initial WIP commit Check Pre-Tokenizer Hashes #25: Commit cd243c5 pushed by ORippler
1m 49s osimons/derived_tensor_poc
osimons/derived_tensor_poc
1m 49s
Optimize data-movement of load_tiles_nvfp4_producer Check Pre-Tokenizer Hashes #24: Commit fb42c25 pushed by ORippler
1m 0s osimons/warp-specialization-mmq-wip
osimons/warp-specialization-mmq-wip
1m 0s
CUDA: Limit DeviceSegmentedSort to immediate mode Check Pre-Tokenizer Hashes #23: Commit 51a86d5 pushed by ORippler
1m 17s osimons/cuda_fix_argsort_graph_capture
osimons/cuda_fix_argsort_graph_capture
1m 17s
CUDA: Fix CUB's argsort when nrows % block_size == 0 CCCL < 3.1 Check Pre-Tokenizer Hashes #22: Commit 28c3c54 pushed by ORippler
1m 3s osimons/cuda_fix_argsort_ceildiv_for_cccl_lt_3dot1
osimons/cuda_fix_argsort_ceildiv_for_cccl_lt_3dot1
1m 3s
WIP showing that data-access is the limitation in PP Check Pre-Tokenizer Hashes #21: Commit d7896a7 pushed by ORippler
53s osimons/gdx-pp-optimizations
osimons/gdx-pp-optimizations
53s
Apply code-formating Check Pre-Tokenizer Hashes #20: Commit e1cfcf8 pushed by ORippler
52s osimons/gated_delta_net_improvements
osimons/gated_delta_net_improvements
52s
CUBLAS: Use FP32 accumulation also for VOLTA TCs Check Pre-Tokenizer Hashes #19: Commit 4eee426 pushed by ORippler
1m 0s osimons/add_bf16_cublas_path
osimons/add_bf16_cublas_path
1m 0s
Add WIP implementation on cuda graph logic Check Pre-Tokenizer Hashes #18: Commit f39a785 pushed by ORippler
1m 20s osimons/update_cuda_graph_heuristics
osimons/update_cuda_graph_heuristics
1m 20s
Apply code-formater Check Pre-Tokenizer Hashes #17: Commit 28177e3 pushed by ORippler
3m 21s osimons/cuda_graphs_qwen3_next
osimons/cuda_graphs_qwen3_next
3m 21s
Do not mutate cgraph for fused ADDs Check Pre-Tokenizer Hashes #16: Commit 79f5eaf pushed by ORippler
1m 7s osimons/fix_multi_add
osimons/fix_multi_add
1m 7s
Bump consec_update threshold by 1 to enable CG on qwen3-next Check Pre-Tokenizer Hashes #15: Commit 2a59644 pushed by ORippler
1m 5s osimons/cuda_graphs_qwen3_next
osimons/cuda_graphs_qwen3_next
1m 5s
CUDA: Update CCCL-tag for 3.2 to final release from RC Check Pre-Tokenizer Hashes #14: Commit d8da435 pushed by ORippler
1m 8s master
master
1m 8s
Fix rope_norm Check Pre-Tokenizer Hashes #13: Commit 3fa5110 pushed by ORippler
57s osimons/cuda_fix_non-contig_rope
osimons/cuda_fix_non-contig_rope
57s
CUDA: Fix loop unrolling for BW in mul_mat_q_stream_k_fixup Check Pre-Tokenizer Hashes #12: Commit b056b05 pushed by ORippler
29s osimons/fix_bw_mmq_fixup_kernel
osimons/fix_bw_mmq_fixup_kernel
29s
CUDA: Fix builds for older CCCL versions by ifdefing strided_iterator Check Pre-Tokenizer Hashes #11: Commit f8b1071 pushed by ORippler
32s osimons/ifdef_strided_iterator
osimons/ifdef_strided_iterator
32s
Remove unnecessary include from top-k.cu Check Pre-Tokenizer Hashes #10: Commit 90c162a pushed by ORippler
29s osimons/cuda_iterator_in_argsort
osimons/cuda_iterator_in_argsort
29s
Use two_stage_warp_reduce also in softmax kernel, move smem out of it Check Pre-Tokenizer Hashes #9: Commit 1ebe58d pushed by ORippler
30s osimons/factor_out_two_stage_warp_reductions
osimons/factor_out_two_stage_warp_reductions
30s
Log all errors in compare_token_data Check Pre-Tokenizer Hashes #8: Commit da27c9b pushed by ORippler
30s osimons/gpu-sampling-equivalence-checks
osimons/gpu-sampling-equivalence-checks
30s
Log all errors in compare_token_data Check Pre-Tokenizer Hashes #7: Commit 7ee23c0 pushed by ORippler
28s osimons/gpu-sampling-equivalence-checks
osimons/gpu-sampling-equivalence-checks
28s
CUDA: Add top-k implementation Check Pre-Tokenizer Hashes #6: Commit f23b306 pushed by ORippler
33s gpu-sampling
gpu-sampling
33s
Check Pre-Tokenizer Hashes Check Pre-Tokenizer Hashes #5: Commit 6ed6d2b pushed by ORippler
48s osimons/prefetch_gate_bias_in_fused_mmvq
osimons/prefetch_gate_bias_in_fused_mmvq
48s
Add fastdiv and fastmodulo to k_bin_bcast kernel Check Pre-Tokenizer Hashes #4: Commit 956a1d0 pushed by ORippler
33s osimons/add_fastdiv_to_k_bin_bcast
osimons/add_fastdiv_to_k_bin_bcast
33s