exp_001_base_vanilla โ€” whisper.cpp GGML builds

whisper.cpp conversions of ihanif/exp_001_base_vanilla, a Whisper-base fine-tune for Pashto ASR (reported WER 21.22 / CER 8.52 on its eval split).

File Format Size Use
ggml-exp_001_base_vanilla-q8_0.bin GGML q8_0 ~78 MB on-device (whisper.cpp / whisper.rn)
ggml-exp_001_base_vanilla-f16.bin GGML f16 ~141 MB requantization source / desktop

Converted with whisper.cpp/models/convert-h5-to-ggml.py and whisper-quantize (whisper.cpp 1.9.1, ggml 0.15.3). Tokenizer files taken from openai/whisper-base (unchanged by the fine-tune).

Usage:

whisper-cli -m ggml-exp_001_base_vanilla-q8_0.bin -l ps audio.wav

Used by the PashtoPal app as its default speech-to-text model.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for ihanif/exp_001_base_vanilla-gguf

Finetuned
(1)
this model