Issues

• import thunder fails on torch >= 2.14.0: Function.apply has no __func__ (pytorch #190694 moved apply into C++)

  Lightning-AI/lightning-thunder#2833 · Filter by author BwL1289 opened on Sep 19, 2026

• Consider using pre-built Flash Attention kernels via kernels

  Lightning-AI/lightning-thunder#2821 · Filter by author sayakpaul opened on May 28, 2026

• Refactor Triton cross-entropy executor: cleanup | Reduce activation memory in cross-entropy backward

  Lightning-AI/lightning-thunder#2818 · Filter by author oonyshch opened on Mar 23, 2026
  1

• Remove xfails once PyTorch autograd stabilizes

  autogradtesting
  Lightning-AI/lightning-thunder#2807 · Filter by author mattteochen opened on Dec 16, 2025

• args_tensor_mask has been removed from torch.ops.higher_order.autograd_function_apply

  autograd
  Lightning-AI/lightning-thunder#2803 · Filter by author mattteochen opened on Dec 15, 2025

• In-place ops on cloned tensor propagates to the original tensor

  Lightning-AI/lightning-thunder#2793 · Filter by author shino16 opened on Dec 10, 2025

• Investigate torch.Tensor.__setitem__ as unsupported in thunderfx

  operators
  Lightning-AI/lightning-thunder#2790 · Filter by author mattteochen opened on Dec 3, 2025
  3

  

• [symbolic values] Trace bakes in the initial value of input used for trace construction

  symbolic values
  Lightning-AI/lightning-thunder#2789 · Filter by author kshitij12345 opened on Dec 3, 2025
  2

  

• AssertionError for mutation on intermediates being reordered after return

  Lightning-AI/lightning-thunder#2776 · Filter by author shino16 opened on Nov 27, 2025
  1

  

• [inference] High thunder.jit/thunderfx overhead for small module

  Lightning-AI/lightning-thunder#2775 · Filter by author kshitij12345 opened on Nov 27, 2025
  4

  

• Too many update_aliases after in-place op

  aliasingin-place
  Lightning-AI/lightning-thunder#2768 · Filter by author shino16 opened on Nov 25, 2025
  1

• Errors on in-place ops on tensor aliases unresolved by proxy substitution

  aliasingin-place
  Lightning-AI/lightning-thunder#2766 · Filter by author shino16 opened on Nov 24, 2025
  13

• Replace usage of nvFuser's broadcast_in_dim in executor with broadcast+expand ops

  enhancementexecutorsnvfuser
  Lightning-AI/lightning-thunder#2765 · Filter by author IvanYashchuk opened on Nov 23, 2025

• thunderfx: Applying thunderfx should return a nn.Module like object.

  thunderfxux
  Lightning-AI/lightning-thunder#2757 · Filter by author kshitij12345 opened on Nov 19, 2025

• Integration Test: KV-Cache management with symbolic values

  enhancementsymbolic values
  Lightning-AI/lightning-thunder#2742 · Filter by author IvanYashchuk opened on Nov 13, 2025
  4

• Support symbolic tensor shapes and dynamic dimensions

  enhancementsymbolic values
  Lightning-AI/lightning-thunder#2741 · Filter by author IvanYashchuk opened on Nov 13, 2025

• Support symbolic values in torch.arange and other tensor operations

  enhancementsymbolic values
  Lightning-AI/lightning-thunder#2738 · Filter by author IvanYashchuk opened on Nov 13, 2025
  2

• Support symbolic integer scalars

  enhancementsymbolic values
  Lightning-AI/lightning-thunder#2735 · Filter by author IvanYashchuk opened on Nov 13, 2025
  

• [Pattern matching] Replace computation patterns on the FX Graph level

  pattern matching
  Task

  Lightning-AI/lightning-thunder#2731 · Filter by author riccardofelluga opened on Nov 12, 2025
  

• Make prims.shape fusible with nvFuser executor

  enhancementfusion logicnvfuserprimitives
  Lightning-AI/lightning-thunder#2730 · Filter by author IvanYashchuk opened on Nov 12, 2025
  

• Using symbolic values leads to unfused program.

  symbolic values
  Lightning-AI/lightning-thunder#2729 · Filter by author kshitij12345 opened on Nov 12, 2025
  2

• Tests incompatible with recent transformers

  Lightning-AI/lightning-thunder#2726 · Filter by author t-vi opened on Nov 11, 2025
  1

• [DTensor] Refactor to re-use prims

  DTensor
  Lightning-AI/lightning-thunder#2722 · Filter by author kshitij12345 opened on Nov 10, 2025
  1

• [benchmark_inference] Investigate why the first and last Dynamo subgraph are not wrapped with cuda graph regions

  cudagraphs
  Lightning-AI/lightning-thunder#2717 · Filter by author mattteochen opened on Nov 4, 2025

• DTensor: nvFP4 grouped_mm and linear support for TensorParallel

  DTensorenhancement
  Lightning-AI/lightning-thunder#2712 · Filter by author kshitij12345 opened on Nov 3, 2025