All issues

Search results

BoostingTreeModelConfig.use_triton defaults to True, so GPU-PB via key_phrases_* crashes on CUDA when Triton is missing

community-request

waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respond

Status: Open.

#16122 In NVIDIA-NeMo/Speech;

· rips4w opened on Aug 20, 2026

Nemotron Labs VoiceChat image for streaming interaction

community-request

waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respond

Status: Open.

#16118 In NVIDIA-NeMo/Speech;

· yangyyt opened on Aug 20, 2026

UnboundLocalError: storage_options when an EMA callback is present and save_last_n_optim_states >= 0

community-request

Status: Open.

#16111 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

exp_manager S3 resume check runs on every rank because the guard ANDs two mutually exclusive URL schemes

community-request

Status: Open.

#16110 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

ENCharParser(abbreviation_version='expanded') returns None for every input

community-request

Status: Open.

#16108 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

FeatureLabel.uniq_labels collects the characters of each label, breaking FeatureToLabelDataset

community-request

Status: Open.

#16106 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

random_segment augmentation raises TypeError: pad_width must be of integral type whenever it has to pad

community-request

Status: Open.

#16104 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

save_adapters/load_adapters do not round-trip: subset saves fail to load, and names are matched by substring

community-request

Status: Open.

#16102 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

Lhotse RIR augmentation is seeded from seed instead of shard_seed, so all ranks draw identical reverberation

community-request

Status: Open.

#16100 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

Frame offset is 10x too large when merging hour-long chunk hypotheses

community-request

waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respond

Status: Open.

#16099 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

GreedyTDTInfer skips an extra encoder frame when the symbol budget is exhausted

community-request

waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respond

Status: Open.

#16095 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026

Corpus average_cpWER still uses the concatenation algorithm that PR #15573 replaced

community-request

Status: Open.

#16093 In NVIDIA-NeMo/Speech;

· udsy19 opened on Aug 18, 2026