Issues

• Running the test_run_model test results in a segmentation fault error

  AI-Hypercomputer/torchprime#377 · Filter by author erichuang-cienet opened on Sep 2, 2025

• Some tests will hang when running tests with pytest on a TPU VM

  AI-Hypercomputer/torchprime#375 · Filter by author erichuang-cienet opened on Aug 28, 2025

• Jobs hang with fsdp only (1d) sharding with >=64 tpu chips

  AI-Hypercomputer/torchprime#363 · Filter by author jialei777 opened on Aug 13, 2025

• different SyncTensorsGraph for different training steps

  AI-Hypercomputer/torchprime#360 · Filter by author jialei777 opened on Aug 12, 2025
  3

• Bug in torch_xla._XLAC._xla_tensors_from_aten

  AI-Hypercomputer/torchprime#353 · Filter by author liurupeng opened on Jul 29, 2025

• Add backwards UnitTests to existing models

  AI-Hypercomputer/torchprime#352 · Filter by author vlasenkoalexey opened on Jul 25, 2025
  1

  

• Update e2e_testing/update_step_time.py tool to preserve existing values

  AI-Hypercomputer/torchprime#348 · Filter by author vlasenkoalexey opened on Jul 23, 2025

• Llama 8.0 8B SFT and assume_pure errors

  bug
  Bug

  AI-Hypercomputer/torchprime#343 · Filter by author pgmoka opened on Jul 16, 2025

• replace hardcoded string with enum

  AI-Hypercomputer/torchprime#333 · Filter by author liurupeng opened on Jul 8, 2025

• Allow passing entire config to attention layer to dedup parameters spec in model config and sharding config

  AI-Hypercomputer/torchprime#330 · Filter by author liurupeng opened on Jun 30, 2025

• Refactor test directory structure

  AI-Hypercomputer/torchprime#325 · Filter by author yaoshiang opened on Jun 24, 2025

• Inquiry on Torch XLA Scalability and Performance Comparison with JAX on TPU

  AI-Hypercomputer/torchprime#313 · Filter by author Rayjryang opened on Jun 15, 2025

• Import modules instead of individual functions

  AI-Hypercomputer/torchprime#310 · Filter by author jialei777 opened on Jun 13, 2025

• MegaScale (DCN) performance regression as of libtpu May 15

  AI-Hypercomputer/torchprime#300 · Filter by author tengyifei opened on Jun 9, 2025

• Add script to run SFT with GPU, and compare GPU metric and TPU metrics

  AI-Hypercomputer/torchprime#299 · Filter by author jialei777 opened on Jun 9, 2025
  

• Consolidate unit test with the same dummy LLAMA model

  AI-Hypercomputer/torchprime#298 · Filter by author jialei777 opened on Jun 9, 2025
  

• Replace dynamic importlib imports with lookup table

  AI-Hypercomputer/torchprime#283 · Filter by author yaoshiang opened on Jun 5, 2025

• Standardize step time calculation

  AI-Hypercomputer/torchprime#276 · Filter by author tengyifei opened on Jun 2, 2025

• Integrate GCS cache loading into torchprime

  AI-Hypercomputer/torchprime#262 · Filter by author jack8558 opened on May 29, 2025
  

• Experiment with cacheing data to GCS and loading data from GCS

  AI-Hypercomputer/torchprime#261 · Filter by author jack8558 opened on May 29, 2025
  1

  

• Unwanted recompilation near the end of training

  AI-Hypercomputer/torchprime#260 · Filter by author tengyifei opened on May 29, 2025
  2

• Support uploading results to BigQuery in tp docker-run

  AI-Hypercomputer/torchprime#254 · Filter by author tengyifei opened on May 28, 2025
  

• Support for scheduling torchprime from XLML using xpk

  AI-Hypercomputer/torchprime#253 · Filter by author tengyifei opened on May 28, 2025

• Add Qwen3 in torch_xla_models

  unscheduled
  AI-Hypercomputer/torchprime#251 · Filter by author weirayao opened on May 26, 2025
  1

• Let trainer to default use number of devices avaialble in tp run

  AI-Hypercomputer/torchprime#229 · Filter by author jack8558 opened on May 8, 2025