Skip to content

feat[asr-ggml]: benchmark Indic Conformer accuracy - #3798

Merged
Zbig9000 merged 5 commits into
mainfrom
feat/asr-indic-benchmark-results
Aug 13, 2026
Merged

feat[asr-ggml]: benchmark Indic Conformer accuracy#3798
Zbig9000 merged 5 commits into
mainfrom
feat/asr-indic-benchmark-results

Conversation

@Zbig9000

@Zbig9000 Zbig9000 commented Aug 12, 2026

Copy link
Copy Markdown
Contributor

Summary

  • integrate Indic Conformer into the existing Parakeet accuracy matrix
  • benchmark the complete 22-lane matrix across TDT, CTC, EOU, Sortformer, and Indic Conformer
  • merge every matrix artifact into one workflow summary and tracked results table
  • forward and validate Indic model language identifiers through the benchmark client and server

Full matrix results

Model Type Language GPU Mode WER CER Dataset
indic-conformer-ctc.q8_0.gguf indic-conformer Gujarati CPU batch 10.81 6.07 FLEURS
indic-conformer-ctc.q8_0.gguf indic-conformer Gujarati GPU batch 10.81 6.07 FLEURS
indic-conformer-ctc.q8_0.gguf indic-conformer Hindi CPU batch 5.37 2.97 FLEURS
indic-conformer-ctc.q8_0.gguf indic-conformer Hindi GPU batch 5.43 2.97 FLEURS
indic-conformer-ctc.q8_0.gguf indic-conformer Kannada CPU batch 8.65 3.67 FLEURS
indic-conformer-ctc.q8_0.gguf indic-conformer Kannada GPU batch 8.65 3.67 FLEURS
indic-conformer-ctc.q8_0.gguf indic-conformer Tamil CPU batch 17.96 12.04 FLEURS
indic-conformer-ctc.q8_0.gguf indic-conformer Tamil GPU batch 17.96 12.04 FLEURS
parakeet-ctc-0.6b.f16.gguf ctc English CPU batch 1.51 0.65 LibriSpeech
parakeet-ctc-0.6b.f16.gguf ctc English CPU streaming 1.51 0.65 LibriSpeech
parakeet-ctc-0.6b.f16.gguf ctc English GPU batch 1.51 0.65 LibriSpeech
parakeet-eou-120m-v1.f16.gguf eou English CPU batch 3.01 1.46 LibriSpeech
parakeet-eou-120m-v1.f16.gguf eou English CPU streaming 3.01 1.46 LibriSpeech
parakeet-eou-120m-v1.f16.gguf eou English GPU streaming 3.01 1.46 LibriSpeech
parakeet-tdt-0.6b-v3.f16.gguf tdt French CPU batch 5.20 2.19 FLEURS
parakeet-tdt-0.6b-v3.f16.gguf tdt German CPU batch 4.61 1.40 FLEURS
parakeet-tdt-0.6b-v3.f16.gguf tdt Spanish CPU batch 3.18 1.19 FLEURS
parakeet-tdt-0.6b-v3.f16.gguf tdt English CPU batch 2.26 0.77 LibriSpeech
parakeet-tdt-0.6b-v3.f16.gguf tdt English CPU streaming 2.26 0.77 LibriSpeech
parakeet-tdt-0.6b-v3.f16.gguf tdt English GPU batch 2.26 0.77 LibriSpeech
sortformer-4spk-v1.f16.gguf sortformer English CPU batch n/a n/a LibriSpeech
sortformer-4spk-v1.f16.gguf sortformer English GPU batch n/a n/a LibriSpeech

Test plan

  • Run the Parakeet Python benchmark tests (25 passed)
  • Run benchmark server validation tests (2 passed)
  • Validate server JavaScript formatting and lint
  • Validate the complete 22-lane matrix generation
  • Run the complete matrix and generate the consolidated summary artifact

Add multilingual FLEURS lanes and language-aware benchmark plumbing so Indic Conformer quality can be measured and tracked consistently.
@Zbig9000
Zbig9000 requested review from a team as code owners August 12, 2026 13:33
@github-actions

github-actions Bot commented Aug 12, 2026

Copy link
Copy Markdown
Contributor

Review Status

Current Status: ✅ APPROVED
Approvals so far: Team Lead: 1, Member: 1

@github-actions

github-actions Bot commented Aug 12, 2026

Copy link
Copy Markdown
Contributor

License compliance — findings detected (warn-only)

Critical: 0 · High: 2 · Medium: 0

Dependency License Scope Severity Outcome
actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c (none detected) runtime High blocks
actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f (none detected) runtime High blocks

How to resolve a blocking finding:

  • Remove or replace the disallowed dependency, or
  • If the license is genuinely acceptable, run the compliance SKILL and record the decision in .github/license-allowlist.yml (CODEOWNERS-reviewed), or
  • For a one-off, a maintainer can apply the license-override label (High findings only; Critical cannot be overridden).

Warn-only (shadow) mode — this check does not block merges yet.

Updated automatically by the canonical license compliance workflow.

NOTICE presence (advisory)

Missing NOTICE (advisory, does not block):

  • ./.github/actions/release-merge-guard
  • ./packages/vla-ggml/sim/server
  • ./packages/asr-ggml/benchmarks/server
  • ./packages/embed-llamacpp/benchmarks/performance
  • ./packages/embed-llamacpp/benchmarks/server
  • ./packages/llm-llamacpp/benchmarks/performance
  • ./packages/llm-llamacpp/benchmarks/server
  • ./packages/inference-addon-cpp/mobile
  • ./packages/ggml-coload-smoke
  • ./packages/sdk/e2e
  • ./packages/fabric/test/integration
  • ./docs/website

@github-actions

github-actions Bot commented Aug 12, 2026

Copy link
Copy Markdown
Contributor

Workflow security (shadow mode)

zizmor found 1110 finding(s) in .github/ (highest severity: high). This check is warn-only and does not block the merge.

Findings are annotated inline on the changed files and listed in the job summary.

Reproduce locally:

pipx run zizmor==1.27.0 --offline .github/

Pin the accuracy workflow to Python 3.12 so the dataset stack can load FLEURS without dill compatibility failures.
Replace the isolated model rows with the consolidated all-model table generated by the successful benchmark workflow.
@Zbig9000

Copy link
Copy Markdown
Contributor Author

License-bot finding reviewed:

This is a scanner none detected false positive in warn-only mode, not a disallowed dependency. A durable allowlist entry requires a CODEOWNERS/Tier-1 compliance approval and a truthful approved-by value, so it should be added only if an authorized reviewer chooses to record that decision. The documented license-override label is not currently defined in this repository.

@Zbig9000
Zbig9000 requested a review from freddy311082 August 13, 2026 10:40
@Zbig9000
Zbig9000 merged commit c677706 into main Aug 13, 2026
49 checks passed
@Zbig9000
Zbig9000 deleted the feat/asr-indic-benchmark-results branch August 13, 2026 11:02
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants