Issues

• Rules.md 4.6.2 "more than 4 accelerators per host": working-group disposition needed — 7 v3.0 orgs (and 4 v2.0 rows) ran exactly 4 per host

  mlpstorageRules
  mlcommons/storage#866 · Filter by author FileSystemGuy opened on Sep 20, 2026

• Running KVcache after VectorDB causes a hang in cluster_collector.py

  mlcommons/storage#843 · Filter by author willo7734 opened on Aug 11, 2026
  1

• Finite replay cycles drop queued requests

  FutureKVCache TF
  mlcommons/storage#839 · Filter by author jay-tau opened on Aug 2, 2026
  

• Rules.md does not describe public_ids.json, which reportgen now writes into submission packages

  FuturemlpstorageRules
  mlcommons/storage#838 · Filter by author FileSystemGuy opened on Jul 30, 2026

• KV cache: seeded runs produce different logical file-I/O workloads

  FutureKVCache TF
  mlcommons/storage#822 · Filter by author jay-tau opened on Jul 20, 2026
  7

• [Bug] MLPStorage Validation Fails with "Expected 6 run files, but found 12" for RetinaNet on Both B200 and MI355

  mlcommons/storage#679 · Filter by author xutao00090 opened on Jul 5, 2026
  5

  

• DLIO summary.json: host_memory_GB / host_cpu_count arrays are per-nnodes-slot, not per-host — positional access is broken

  DLIO or mlpstorageFuture
  mlcommons/storage#676 · Filter by author FileSystemGuy opened on Jul 5, 2026
  

• Clarification on client host memory requirement for 1T model checkpointing – changed between v2.0 and v3.0?

  documentationRules
  mlcommons/storage#518 · Filter by author litianqi00315 opened on Jun 25, 2026
  3

• ^C doesn't abort all the worker processes

  Future
  Bug

  mlcommons/storage#515 · Filter by author FileSystemGuy opened on Jun 25, 2026
  

• Enforce --data-dir post-YAML for all training commands; eliminate inherited DLIO YAML default

  Future
  Bug

  mlcommons/storage#508 · Filter by author FileSystemGuy opened on Jun 24, 2026
  1

• Intermittent errors between epochs

  bugDLIO or mlpstorageFuture
  mlcommons/storage#499 · Filter by author rasar07 opened on Jun 23, 2026
  2

• DISKANN 10M×1536-dim recall requires search_list=4000 to reach 95%, causing unacceptable latency and throughput

  VectorDB TF
  mlcommons/storage#425 · Filter by author Raysmond opened on Jun 10, 2026
  6

  

• KVCache: eviction-driven tier I/O is excluded from latency P95 SLA checks (Storage/CPU write & read latencies never recorded in _demote_entry)

  FutureKVCache TF
  mlcommons/storage#421 · Filter by author jay-tau opened on Jun 9, 2026
  2

• [Training 3.0] Add interlock file to data directories

  enhancementTraining
  mlcommons/storage#390 · Filter by author dslik opened on May 26, 2026

• [Training v3.0 consolidation] Ensure the Parquet data generator delivers reasonable generation throughput

  enhancementFuturePASS Score BlockerTraining
  mlcommons/storage#358 · Filter by author wolfgang-desalvador opened on Apr 27, 2026

• [Training v3.0 consolidation] Evaluate Parquet reader performance in FLUX to validate it can meet target throughput

  enhancementFuturePASS Score BlockerTraining
  mlcommons/storage#357 · Filter by author wolfgang-desalvador opened on Apr 27, 2026

• [Training v3.0 consolidation] Keep a single Parquet file handle open in the FLUX reader to avoid row group cache eviction

  enhancementFuturePASS Score BlockerTraining
  mlcommons/storage#356 · Filter by author wolfgang-desalvador opened on Apr 27, 2026

• [Training v3.0 consolidation] Evaluate Arrow IPC as alternative format for DLRM

  enhancementFuturePASS Score BlockerTraining
  mlcommons/storage#355 · Filter by author wolfgang-desalvador opened on Apr 27, 2026

• [Training v3.0 consolidation] DLRM optimization: evaluate Parquet reader and generation parameters to improve throughput

  enhancementFuturePASS Score BlockerTraining
  mlcommons/storage#354 · Filter by author wolfgang-desalvador opened on Apr 27, 2026
  5

• [Training v3.0 consolidation] Confirm computation time step for DLRM

  bugFuturePASS Score BlockerTraining
  mlcommons/storage#353 · Filter by author wolfgang-desalvador opened on Apr 27, 2026

• ValueError: 'parquet' is not a valid FormatType when using mlpstorage training datagen

  Future
  mlcommons/storage#336 · Filter by author litianqi00315 opened on Apr 15, 2026

• KVCache Verify client CPU choice has minimal influence on results

  FutureKVCache TF
  mlcommons/storage#334 · Filter by author dslik opened on Apr 14, 2026
  1

• [Proposal] Evaluate Arrow IPC format for reader

  enhancementFuture
  mlcommons/storage#333 · Filter by author wolfgang-desalvador opened on Apr 14, 2026
  3

• mlpstorage error with proxy, needs httpx[socks]

  enhancementFuture
  mlcommons/storage#332 · Filter by author dslik opened on Apr 14, 2026
  3

• mlpstorage training run --model=flux has considerably lower I/O Throughput causing train_au_meet_expectation to fail

  bugFuture
  mlcommons/storage#330 · Filter by author ddn-kums opened on Apr 12, 2026
  6