Repository Issue Activity (beta)

#3001 Broadcast store writes only one chunk under looped reduction

#2972 Add failure tracking for benchmark

#2705 Same autotuned "static_ranges" config fails across different specialized shapes

#2745 [RFC] Helion Remote Autotuning Service

#2842 [Pallas] adjust_block_size_constraints max-propagates per-tensor TPU alignment across all tile-indexed tensors sharing a block_id

#2643 [Bug][CodeGen] CodeGen compilation error for reduction along non-tiled dimension

#2304 [Pallas] test_atomic_add_w_tile_attr broken with recent torch_tpu

#97 Support TensorDescriptor + extra_mask

#2531 Compile-phase subprocess isolation to survive Triton C++ aborts

#2469 Support subprocess benchmarking for non-picklable epilogue arguments

#2468 Optimize fixed launch overheads

#2417 fast nvfp4 gemv

#1346 [RFC] Helion kernel + Inductor ops prologue / epilogue fusion

#934 Setting to disable 0/1 specialization

#1539 Control recompilations/specialization tighter from helion, including triton recompilations

#2266 [RFC] Structured compilation pipeline with `KernelCompiler`

#1803 [RFC] ConfigPipeline/AutotuneBackend: Separating compilation/benchmarking from search algorithms

#2044 Kernel perf regression using latest main branch with configs autotuned from v1.0.0

#2148 Better support for Larger Grouped Operations in Helion in hl.grid

#2147 Better support for hl.rand compared to tl.rand (custom offsets, tl.rand4x, etc.)

#2184 large world size causes a crash

#2176 [bug] missing clamp in attention autotuning config generation

#1064 Discover a working default config automatically

#1113 Support an autotune policy between "none" and "quick"

#1114 Optimize kernel dispatch at runtime.