Crisp ASR Cohere: add Arabic and Japanese models#12383
Merged
Merged
Conversation
The Cohere backend only listed the general cohere-transcribe-03-2026 GGUFs. Two more model families exist for the same backend and are worth offering: * cohere-transcribe-arabic-07-2026 - Arabic/English specialist tuned for Arabic dialects and code switching, clearly ahead of the general model on Arabic (~25.9 vs ~30.7 WER on Cohere's Open Universal Arabic ASR board). Same uploader (cstr) as the general GGUFs we already ship. * efwkjn/cohere-asr-ja - Japanese fine-tune of the same base, general/anime domains, well ahead of the stock model on Japanese CER. Suggested by muaz978 in #12290. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds two more model families to the existing Crisp ASR Cohere backend. No new engine, no new dependency, no Python — just extra entries next to the current
cohere-transcribe-03-2026quants.Arabic —
cstr/cohere-transcribe-arabic-07-2026-GGUFThe Arabic/English specialist, tuned for Arabic dialects and Arabic/English code switching. On Cohere's Open Universal Arabic ASR leaderboard it is a clear step ahead of the general model (~25.9 vs ~30.7 WER). Uploaded by
cstr, the same author whose general Cohere GGUFs we already ship.cohere-transcribe-arabic-q4_k.ggufcohere-transcribe-arabic-q8_0.ggufcohere-transcribe-arabic-f16.ggufJapanese —
CKHO/cohere-asr-ja-GGUFGGUF quant of
efwkjn/cohere-asr-ja, a Japanese fine-tune of the samecohere-transcribe-03-2026base (general + anime domains). The author's own eval has it well ahead of the stock Cohere model on Japanese CER — 12.9 vs 22.0 on TEDx, 17.5 vs 29.8 on jsut-book. Converted with CrispASR's ownconvert-cohere-asr-to-gguf.pyand tagged for thecoherebackend.cohere-asr-ja-q4_k.ggufcohere-asr-ja-q6_k.ggufcohere-asr-ja-q8_0.ggufcohere-asr-ja-f16.ggufNote the Japanese quants are hosted by a third party rather than by
cstr— happy to drop them or swap inTransWithAI/cohere-transcribe-ja-v0.1-GGUF(older v0.1 checkpoint, measurably worse per the author's table) if we would rather stay oncstr-hosted repos only.Suggested by @muaz978 in #12290.
🤖 Generated with Claude Code