Website: separate Fun-ASR-Nano and MLT-Nano language claims

[LauraGPT]

Problem

As of 2026-07-15, the live FunASR website attributes the 31-language capability to Fun-ASR-Nano-2512, but that capability belongs to the separate Fun-ASR-MLT-Nano-2512 checkpoint.

Affected live pages:

• Chinese homepage: https://www.funasr.com/ shows Fun-ASR-Nano · LLM-ASR · 31 语种
• English homepage: https://www.funasr.com/en/ shows Fun-ASR-Nano · LLM-ASR · 31 languages
• Chinese model guide: https://www.funasr.com/blog/which-funasr-model.html describes Nano as a 31-language model and uses FunAudioLLM/Fun-ASR-Nano-2512
• English model guide: https://www.funasr.com/en/blog/which-funasr-model.html repeats the same claim and model ID
• Both homepages still show 16K+ GitHub Stars; modelscope/FunASR currently has 19,236 stars. If this is intended as an ecosystem total, the four maintained repos currently total 35,404, so the label also needs to state what is being counted.

The canonical checkpoint split is already documented in the current Fun-ASR README:

• Fun-ASR-Nano-2512: Chinese, English, Japanese, plus 7 Chinese dialect groups and 26 regional accents
• Fun-ASR-MLT-Nano-2512: 31 languages

The Transformers-compatible Nano model card now reflects the same scope: FunAudioLLM/Fun-ASR-Nano-2512-hf discussion #1.

This is a high-traffic accuracy issue: the current 14-day GitHub referrer snapshot attributes 2,343 visits across FunASR, Fun-ASR, SenseVoice, and FunClip to funasr.com.

Requested changes

1. Separate Nano and MLT-Nano in the Chinese and English homepage product copy.
2. Change the Nano copy to Chinese / English / Japanese plus Chinese dialects and accents.
3. Present MLT-Nano as the separate 31-language checkpoint, with its own Hugging Face and ModelScope links.
4. Correct both which-funasr-model.html guides so the 31-language row points to FunAudioLLM/Fun-ASR-MLT-Nano-2512, while the Nano row keeps the Nano model ID and its actual language scope.
5. Replace the hard-coded 16K+ value with a dynamic count, or remove it. If an ecosystem aggregate is used, label it explicitly.
6. Keep Chinese and English page copy, links, and metadata in sync.

Source / deployment ownership

The live pages are static HTML (quickstart.html, models.html, ecosystem.html, and /blog/). The old modelscope/FunASR/web-pages/ Vue source does not match the current live site, and I could not locate the current page source on ind-gpu8.

Please identify the authoritative source repository and deployment owner in this issue so the fix is made in the maintained source rather than patched only on the server.

Acceptance criteria

• curl checks on both homepages and both model-guide pages no longer attribute 31 languages to Nano.
• Nano and MLT-Nano have distinct model links and accurate capability copy.
• The displayed star count is current or removed, with an unambiguous label.
• Chinese and English pages agree.
• Desktop and mobile visual checks show no clipped or overlapping hero/model text.

[LauraGPT]

Fixed and deployed to https://www.funasr.com on 2026-07-15.

What changed:

• Separated the flagship Fun-ASR-Nano scope (Chinese, English, Japanese, 7 Chinese dialect groups, and 26 regional accents) from the separate Fun-ASR-MLT-Nano 31-language checkpoint.
• Corrected both homepages and both model-selection guides, including titles, descriptions, Open Graph metadata, JSON-LD, decision tables, direct Hugging Face/ModelScope links, and CTAs.
• Audited the full sitemap and fixed 20 additional Chinese/English pages that repeated the old claim: quickstarts, model directories, Whisper comparisons, blog indexes, Nano guides, OpenAI-compatible API guides, Python STT guides, llama.cpp pages, cloud-alternative guides, and Chinese ASR guides.
• Removed stale hard-coded popularity counts.
• Fixed the affected mobile layouts (navigation overlap, page-wide table overflow, and long inline-code overflow).

Deployment safety:

• Backups were created next to every changed production file (*.bak_20260715_nano_mlt and *.bak_20260715_nano_mlt_audit).
• /tmp/gen_which_model.py was updated and test-generated output matched the deployed selection guides exactly.
• 12 older one-off /tmp generators/patchers containing the obsolete claim were backed up and replaced with explicit refusal wrappers, preventing accidental rollback.

Verification:

• All 24 public HTTPS pages were downloaded after deployment and matched the validated local bytes exactly.
• HTML/JSON-LD and targeted content checks passed.
• Full sitemap audit: audited=60 findings=0 errors=0.
• Browser QA passed at 1440x900 and 390x844; the 20-page mobile batch had no page-width overflow or navigation/title overlap.
• All four Nano/MLT model links return HTTP 200.