Issues

• [BUG] Replacing or deleting a named TensorDictSequential entry raises TypeError

  pytorch/tensordict#2904 · Filter by author mottopanikeiku opened on Oct 17, 2026

• [BUG] select_subsequence fails on a named TensorDictSequential containing a nested sequence

  pytorch/tensordict#2903 · Filter by author mottopanikeiku opened on Oct 17, 2026

• [BUG] ProbabilisticTensorDictModule ignores custom log-probability keys for a CompositeDistribution

  pytorch/tensordict#2902 · Filter by author mottopanikeiku opened on Oct 17, 2026

• [BUG] Under torch.compile(dynamic=True), slicing a lazy stack along a member dim recompiles for every bound

  bugCompiletriaged
  pytorch/tensordict#2896 · Filter by author peterdsharpe opened on Oct 17, 2026
  1

• [Performance] pack(layout=...) traces twice as many FX nodes as a plain torch.cat under torch.compile

  CompilePerformancetriaged
  pytorch/tensordict#2891 · Filter by author peterdsharpe opened on Oct 16, 2026
  1

• [BUG] split_keys on a lazy stack can't select a key that only some members hold

  bugtriaged
  pytorch/tensordict#2827 · Filter by author peterdsharpe opened on Oct 16, 2026
  2

• [Feature Request] Call a module with a stack of parameter sets, with a loop fallback for modules vmap rejects

  enhancementFeaturetriaged
  pytorch/tensordict#2813 · Filter by author peterdsharpe opened on Oct 15, 2026
  1

• [Feature Request] Typed exceptions for stacking and lock failures, so callers can stop matching error text

  enhancementFeaturetriaged
  pytorch/tensordict#2812 · Filter by author peterdsharpe opened on Oct 15, 2026
  1

• [BUG] is_empty() returns True for a lazy stack whose members hold different leaves

  bugtriaged
  pytorch/tensordict#2807 · Filter by author peterdsharpe opened on Oct 15, 2026
  3

• Compiled code that calls state_dict() fails on torch 2.14 and later

  triaged
  pytorch/tensordict#2793 · Filter by author peterdsharpe opened on Oct 15, 2026
  3

• Compiled code that calls detach_() on a tensordict fails with fullgraph=True

  triaged
  pytorch/tensordict#2792 · Filter by author peterdsharpe opened on Oct 15, 2026
  2

• Compiled code fails when it writes to, clones, casts or updates a TensorDictParams

  triaged
  pytorch/tensordict#2791 · Filter by author peterdsharpe opened on Oct 15, 2026
  2

• Compiled code returns UnbatchedTensor values with an empty batch size

  triaged
  pytorch/tensordict#2790 · Filter by author peterdsharpe opened on Oct 15, 2026
  1

• update() and TensorDictParams let a tensordict contain itself, which ends in RecursionError

  triaged
  pytorch/tensordict#2741 · Filter by author peterdsharpe opened on Oct 15, 2026
  1

• TensorDict.empty returns names that don't fit the batch size when given a names list of another length

  triaged
  pytorch/tensordict#2711 · Filter by author peterdsharpe opened on Oct 14, 2026
  3

• [BUG] A with-block around flatten(..., inplace=True) leaves the tensordict flattened, and one around unflatten(..., inplace=True) raises on exit

  bugtriaged
  pytorch/tensordict#2700 · Filter by author peterdsharpe opened on Oct 14, 2026
  3

• [BUG] Lazy stacks reject a valid renaming of their dims and accept duplicate dim names

  bugDeprecationtriaged
  pytorch/tensordict#2690 · Filter by author peterdsharpe opened on Oct 14, 2026
  2

• [Feature Request] Make repeated to(device, non_blocking_pin=True) calls faster by reusing their threads

  enhancementFeaturetriaged
  pytorch/tensordict#2688 · Filter by author peterdsharpe opened on Oct 14, 2026
  2

• [BUG] A tuple of floats stored in a TensorDict becomes a float64 tensor, while the same list becomes float32

  triaged
  pytorch/tensordict#2679 · Filter by author peterdsharpe opened on Oct 14, 2026
  1

• [CI] test_shared_memmap_mult fails at random on macOS because the child process's write never shows

  CItriaged
  pytorch/tensordict#2669 · Filter by author peterdsharpe opened on Oct 14, 2026
  2

• TypedTensorDict.from_dict returns instances that lack required fields or hold undeclared ones

  bugtriaged
  pytorch/tensordict#2661 · Filter by author peterdsharpe opened on Oct 14, 2026
  3

• [BUG] Adding a 0-dim CUDA tensor to a tensordict waits for the GPU

  triaged
  pytorch/tensordict#2616 · Filter by author peterdsharpe opened on Oct 14, 2026
  1

• [BUG] A names list one entry too long is accepted when it holds exactly batch_dims None entries, and its names are dropped

  bugDeprecationtriaged
  pytorch/tensordict#2600 · Filter by author peterdsharpe opened on Oct 14, 2026
  2

• [BUG] LazyStackedTensorDict(*tds, names=[...]) ignores the names it is given

  bugtriaged
  pytorch/tensordict#2559 · Filter by author peterdsharpe opened on Oct 13, 2026
  1

• [BUG] Sampling rows of image data with an index tensor is about 5x slower than it needs to be on CPU

  triagedwontfix
  pytorch/tensordict#2553 · Filter by author peterdsharpe opened on Oct 13, 2026
  1