Issues

• [runtime] Float16<->BFloat16 host conversion turns some NaNs into Inf/±0 and flushes f16-subnormal values to zero

  bugcommunity
  Bug

  tenstorrent/tt-mlir#9388 · Filter by author CloneOfAlex opened on Oct 12, 2026
  1

• Golden PCC check returns 1.0 (pass) when only one tensor is constant, and for any same-sign scalar output

  bugcommunity
  Bug

  tenstorrent/tt-mlir#9387 · Filter by author CloneOfAlex opened on Oct 12, 2026

• [tt-mlir][d2m] TTMetal backend computes an f32-declared op at a 10-bit mantissa, silently

  community
  tenstorrent/tt-mlir#9369 · Filter by author amirnd51 opened on Sep 24, 2026

• [tt-mlir][d2m] --ttir-to-ttmetal-pipeline emits invalid MLIR for any matmul, and exits 0

  community
  tenstorrent/tt-mlir#9368 · Filter by author amirnd51 opened on Sep 24, 2026
  3

• [tt-crank] .to("tt") is slow on multichip

  tt-crank
  Bug

  tenstorrent/tt-mlir#9359 · Filter by author umalesTT opened on Sep 21, 2026
  

• [runtime] ttnn.isfinite returns FLOAT32 for a bf16 input; compiler types it BFLOAT16 (debug runtime asserts, #7930 still open in practice)

  tenstorrent/tt-mlir#9344 · Filter by author agobeljicTT opened on Sep 18, 2026

• Change the indexer_score_dsa pipeline to expect weights in [B, 1, S, H] shape

  tenstorrent/tt-mlir#9334 · Filter by author amalkovTT opened on Sep 16, 2026

• Change ttnn.gelu_bw variant attribute from string to enum

  tenstorrent/tt-mlir#9333 · Filter by author amalkovTT opened on Sep 16, 2026

• [tt-crank] compiled graphs: keep input mutations on device instead of aot's host copy-back

  tt-crank
  tenstorrent/tt-mlir#9328 · Filter by author agobeljicTT opened on Sep 16, 2026

• Enable/check does tt-swiss works as expected

  tt-crank
  tenstorrent/tt-mlir#9322 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Perf degradation after tt-mlir uplift

  tt-crank
  tenstorrent/tt-mlir#9321 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Tracy on Galaxy is slow

  tt-crank
  tenstorrent/tt-mlir#9320 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Add multi-host support

  tt-crank
  tenstorrent/tt-mlir#9319 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Recompute Support Exploration

  tt-crank
  tenstorrent/tt-mlir#9318 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Missing DTensor Sharding Strategies

  tt-crank
  tenstorrent/tt-mlir#9317 · Filter by author pilkicTT opened on Sep 15, 2026
  1

  

• Fused SDPA training via ttml sdpa_fw/sdpa_bw

  tt-crank
  tenstorrent/tt-mlir#9316 · Filter by author pilkicTT opened on Sep 15, 2026
  1

  

• train_llama.py does not train in eager mode

  tt-crank
  tenstorrent/tt-mlir#9315 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• embedding_backward lowering allocates a full weight-sized tensor that is never read

  tt-crank
  tenstorrent/tt-mlir#9314 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Embedding weight gradient comes back rank-4 and cannot be bound as a program input

  tt-crank
  tenstorrent/tt-mlir#9313 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Feature proposal: flag for dumping pre and post decomposition fx graph

  tt-crank
  tenstorrent/tt-mlir#9309 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Feature proposal: Counter for how many times the compile pipeline has been called

  tt-crank
  tenstorrent/tt-mlir#9308 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Feature proposal: Flag for logging every time tensors are being pulled between hosts and device

  tt-crank
  tenstorrent/tt-mlir#9307 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Eager aten view pull tensors to host

  tt-crank
  tenstorrent/tt-mlir#9306 · Filter by author pilkicTT opened on Sep 15, 2026
  2

• Support propagating user defined tt-lang kernels

  tt-crank
  tenstorrent/tt-mlir#9305 · Filter by author pilkicTT opened on Sep 15, 2026
  1

• Enable chisel in tt-kurbla

  tt-crank
  tenstorrent/tt-mlir#9304 · Filter by author pilkicTT opened on Sep 15, 2026
  1