Add nemotron-3.5-asr-streaming-0.6b (multilingual prompt-conditioned streaming)#10

Merged

mudler merged 16 commits into

masterfrom

feat/nemotron-3.5-asr-streaming

Jun 6, 2026

Commits on Jun 6, 2026

feat(convert): emit prompt-conditioning KVs + use_bias; handle nested att_context presets

mudler
and
claude
committed
feat(loader): read prompt-conditioning config + encoder.use_bias

mudler
and
claude
committed
test: make nemotron loader assertions reachable without PARAKEET_TEST_GGUF

mudler
and
claude
committed
feat(baseline): dump prompt_kernel_out + per-language RNNT reference for nemotron

mudler
and
claude
committed
feat: PromptKernel post-encoder conditioning unit + isolated NeMo parity test

mudler
and
claude
committed
feat(model): apply PromptKernel + resolve target_lang in offline decode

mudler
and
claude
committed
test: offline nemotron end-to-end NeMo parity (multi-language)

mudler
and
claude
committed
feat(streaming): apply PromptKernel per chunk; target_lang on session

mudler
and
claude
committed
test: streaming nemotron end-to-end NeMo parity

mudler
and
claude
committed
feat(capi): target_lang variants for transcribe + stream (ABI bump)

mudler
and
claude
committed
feat(cli): --lang flag for multilingual prompt models

mudler
and
claude
committed
test: e2e NeMo-vs-parakeet.cpp comparison harness (per-language, offline+stream)

mudler
and
claude
committed
docs: document nemotron multilingual streaming support + prompt KVs

mudler
and
claude
committed
feat(publish): add nemotron-3.5-asr-streaming-0.6b (5 variants, OpenMDW-1.1 card)

mudler
and
claude
committed
bench: nemotron-3.5-asr CPU benchmark vs NeMo (WER 0, 2.4x f32 / 2.5x q8_0)

mudler
and
claude
committed
fix(streaming): reject unknown target_lang for prompt models (match offline + capi contract)

mudler
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add nemotron-3.5-asr-streaming-0.6b (multilingual prompt-conditioned streaming)#10

Add nemotron-3.5-asr-streaming-0.6b (multilingual prompt-conditioned streaming)#10
mudler merged 16 commits into
masterfrom
feat/nemotron-3.5-asr-streaming

Commits on Jun 6, 2026

feat(convert): emit prompt-conditioning KVs + use_bias; handle nested att_context presets

feat(loader): read prompt-conditioning config + encoder.use_bias

test: make nemotron loader assertions reachable without PARAKEET_TEST_GGUF

feat(baseline): dump prompt_kernel_out + per-language RNNT reference for nemotron

feat: PromptKernel post-encoder conditioning unit + isolated NeMo parity test

feat(model): apply PromptKernel + resolve target_lang in offline decode

test: offline nemotron end-to-end NeMo parity (multi-language)

feat(streaming): apply PromptKernel per chunk; target_lang on session

test: streaming nemotron end-to-end NeMo parity

feat(capi): target_lang variants for transcribe + stream (ABI bump)

feat(cli): --lang flag for multilingual prompt models

test: e2e NeMo-vs-parakeet.cpp comparison harness (per-language, offline+stream)

docs: document nemotron multilingual streaming support + prompt KVs

feat(publish): add nemotron-3.5-asr-streaming-0.6b (5 variants, OpenMDW-1.1 card)

bench: nemotron-3.5-asr CPU benchmark vs NeMo (WER 0, 2.4x f32 / 2.5x q8_0)

fix(streaming): reject unknown target_lang for prompt models (match offline + capi contract)