Update Operations Documentation · Workflow runs · gaugarg-nv/llama.cpp · GitHub

Actions

All workflows
Workflows
- Check Pre-Tokenizer Hashes Check Pre-Tokenizer Hashes
- Copilot Setup Steps Copilot Setup Steps
- Python check requirements.txt Python check requirements.txt
- Python Type-Check Python Type-Check
- Update Operations Documentation Update Operations Documentation
Management
- Caches

Update Operations Documentation

Actions

Loading...
Loading

13 workflow runs

13 workflow runs

cache subgraph splits when cgraph is unchanged Update Operations Documentation #13: Commit 4bfe714 pushed by gaugarg-nv

30s tp_cpu_overhead

tp_cpu_overhead

30s

Change CMake to use NCCL on Windows Update Operations Documentation #12: Commit 68fb0ba pushed by gaugarg-nv

45s nccl_windows

nccl_windows

45s

Change CMake to use NCCL on Windows Update Operations Documentation #11: Commit a92f245 pushed by gaugarg-nv

41s nccl_windows

nccl_windows

41s

Use the new kernel only for nblocks_stream_k_raw > 4 * ntiles_dst to … Update Operations Documentation #10: Commit 25ef2df pushed by gaugarg-nv

40s fa_opt

fa_opt

40s

Change CMake to use NCCL on Windows Update Operations Documentation #9: Commit 23e1600 pushed by gaugarg-nv

48s nccl_windows

nccl_windows

48s

Optimize MOE GEMV kernel for BS > 1. Update Operations Documentation #8: Commit 7cfb67e pushed by gaugarg-nv

59s moe_opt

moe_opt

59s

Increase per-thread work if the K-dimension is small Update Operations Documentation #7: Commit cfbbfb2 pushed by gaugarg-nv

31s small_k_optimization

small_k_optimization

31s

Use a single warp per element instead of a single block per element i… Update Operations Documentation #6: Commit 2afd6a3 pushed by gaugarg-nv

46s small_k_optimization

small_k_optimization

46s

Change CMake to use NCCL on Windows Update Operations Documentation #5: Commit 6c90394 pushed by gaugarg-nv

40s nccl_windows

nccl_windows

40s

Improve CUDA graph capture Update Operations Documentation #4: Commit a874a2c pushed by gaugarg-nv

34s cuda_graph_fix

cuda_graph_fix

34s

Fix the seg fault without NCCL Update Operations Documentation #3: Commit 10385e8 pushed by gaugarg-nv

41s minor_fixes

minor_fixes

41s

Fix Issue !19219 Update Operations Documentation #2: Commit 027ef82 pushed by gaugarg-nv

32s revert_scale_queue_env

revert_scale_queue_env

32s

[CUDA] Reduce CPU-side stalls due to the CUDA command buffer being full Update Operations Documentation #1: Commit d3298dc pushed by gaugarg-nv

11s pp_perf_improve

pp_perf_improve

11s