Downloads · 30 days
0
sookie90/whisper-hinglish-ggml
whisper-hinglish-ggml is a machine learning model from sookie90. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
ggml-trelis-hinglish-q50.bin — a ggml/whisper.cpp conversion of Trelis/whisper-hinglish-preview, quantized to q50, for use with whisper.cpp and whisper-rs.
Downloads · 30 days
0
Access
Public
Updated Sep 17, 2026
Repo size
1.2 GB
Likes
0
Public
Click a slice to open those files.
.bin1.2 GB · 100%
From the Hugging Face model README
ggml-trelis-hinglish-q5_0.bin — a ggml/whisper.cpp conversion of
Trelis/whisper-hinglish-preview,
quantized to q5_0, for use with whisper.cpp and
whisper-rs.
Converted with whisper.cpp's own models/convert-h5-to-ggml.py from the source model's
2026-09-02 HF snapshot, then quantized with whisper-quantize to q5_0 (~1.03 GB).
This checkpoint's vocabulary implies 101 apparent languages — one past the 100-entry
whisper_lang_str() table hardcoded in whisper.cpp's src/whisper.cpp. Stock whisper.cpp (and
any whisper-rs release up to and including 0.16.0 / whisper-rs-sys 0.15.0, the newest published
as of 2026-09-03) will crash on load (SIGSEGV) with this file, because
whisper_lang_str() correctly returns nullptr for the out-of-range id, but its only caller
(whisper.cpp:1665) does std::string(nullptr) unconditionally — undefined behavior.
If you hit a crash loading this model: patch whisper.cpp:1665 to null-check the result of
whisper_lang_str() before constructing a std::string from it (fall back to any placeholder
label on nullptr). This is a one-line fix; no other part of the file needs to change. Tracked
upstream at tazz4843/whisper-rs — check there for
whether a release has since folded the fix in.
whisper-cli -m ggml-trelis-hinglish-q5_0.bin -l hi -f audio.wav