# dynamic\_shapes

**URL:** https://dev-discuss.pytorch.org/c/dev-logs/dynamic-shapes/33.md

[Latest](https://dev-discuss.pytorch.org/latest.md) · [Categories](https://dev-discuss.pytorch.org/categories.md)

---

## [About the dynamic\_shapes category](https://dev-discuss.pytorch.org/t/about-the-dynamic-shapes-category/3340)

<div class="topic-metadata">

**Author:** [@albanD](https://dev-discuss.pytorch.org/u/albanD)\
**Replies:** 0

</div>

---

## [Guard-Free Dynamic Shapes](https://dev-discuss.pytorch.org/t/guard-free-dynamic-shapes/3335)

<div class="topic-metadata">

**Author:** [@laithsakka](https://dev-discuss.pytorch.org/u/laithsakka)\
**Replies:** 0\
**Last updated:** [March 31, 2026, 10:35pm UTC](https://dev-discuss.pytorch.org/t/guard-free-dynamic-shapes/3335 "2026-03-31T22:35:08Z")

</div>

TL;DR – Data-dependent errors (DDEs) are the dominant barrier to exporting models with dynamic shapes. There is widespread consensus that DDEs are a significant issue for export — among the various errors observed, data…

---

## [Reducing Compile-Time Overhead in Unbacked-Symbol-Heavy torch.export Traces](https://dev-discuss.pytorch.org/t/reducing-compile-time-overhead-in-unbacked-symbol-heavy-torch-export-traces/3334)

<div class="topic-metadata">

**Author:** [@laithsakka](https://dev-discuss.pytorch.org/u/laithsakka)\
**Replies:** 0\
**Last updated:** [March 31, 2026, 10:30pm UTC](https://dev-discuss.pytorch.org/t/reducing-compile-time-overhead-in-unbacked-symbol-heavy-torch-export-traces/3334 "2026-03-31T22:30:45Z")

</div>

TL;DR – A regression report revealed that exporting a model with many unbacked (data-dependent) symbols took 264s. Profiling showed the latency was dominated by repeated symbolic reasoning in the shape system. A serie…

---

## [Backed to Unbacked: From Guardable to Guardless Shapes in PyTorch](https://dev-discuss.pytorch.org/t/backed-to-unbacked-from-guardable-to-guardless-shapes-in-pytorch/3333)

<div class="topic-metadata">

**Author:** [@laithsakka](https://dev-discuss.pytorch.org/u/laithsakka)\
**Replies:** 0\
**Last updated:** [March 31, 2026, 10:29pm UTC](https://dev-discuss.pytorch.org/t/backed-to-unbacked-from-guardable-to-guardless-shapes-in-pytorch/3333 "2026-03-31T22:29:51Z")

</div>

TL;DR – We expect unbacked dynamic shapes to become the dominant shape mechanism for Frontier-style workloads due to their better predictability and controllability. However, some blockers remain for their ideal usage…

---

## [Slaying Framework Data-Dependent Errors Dragon 🐉](https://dev-discuss.pytorch.org/t/slaying-framework-data-dependent-errors-dragon/3332)

<div class="topic-metadata">

**Author:** [@laithsakka](https://dev-discuss.pytorch.org/u/laithsakka)\
**Replies:** 0\
**Last updated:** [March 31, 2026, 10:27pm UTC](https://dev-discuss.pytorch.org/t/slaying-framework-data-dependent-errors-dragon/3332 "2026-03-31T22:27:01Z")

</div>

TL;DR – Framework DDE dragon has been slain! :dragon: We’ve eliminated the vast majority of framework data-dependent errors — reducing user issues by over 85% — and unlocked specialization-free full graph capture that j…

---

## [Unbacked Dynamic Shapes Shouldn't Be Slower — Now They Aren't](https://dev-discuss.pytorch.org/t/unbacked-dynamic-shapes-shouldnt-be-slower-now-they-arent/3331)

<div class="topic-metadata">

**Author:** [@laithsakka](https://dev-discuss.pytorch.org/u/laithsakka)\
**Replies:** 0\
**Last updated:** [March 31, 2026, 10:24pm UTC](https://dev-discuss.pytorch.org/t/unbacked-dynamic-shapes-shouldnt-be-slower-now-they-arent/3331 "2026-03-31T22:24:27Z")

</div>

TL;DR – Unbacked dynamic shapes had 2x–20% slowdowns on TorchBench and ~30% regressions on vLLM. We fixed the root causes — now unbacked matches backed across all tested models and configurations. Motivation Unb…