GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
p-quincy
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
p-quincy
Model Tree
openai/whisper-small
Input Modalities
Output Modalities
Supported Functionality
A fine-tune of openai/whisper-small on the
Indonesian subset of Common Voice 27 (5,381 training clips and 400 evaluation clips,
approximately 6 hours of audio), trained on a single RTX 2050 4 GB GPU.
Re-measured with a corrected decoding configuration, this checkpoint reaches WER 17.98% / wer_norm 14.61% on a random 30-clip sample of CV27-id, with zero decode failures. The training-time figure of 59.82% WER was a measurement artifact: the evaluation decodes ran with free language detection (see Measurement note).
| Item | Value |
|---|---|
| Base model | openai/whisper-small |
| Dataset | Common Voice 27, subset id (validated), speaker-disjoint split |
| Train / eval clips | 5,381 / 400 |
| Hardware | 1x RTX 2050 4 GB (CUDA), Python 3.12, transformers 5.17 |
| Batch size | 2 (gradient accumulation 4, effective batch 8) |
| Learning rate | 1e-5, linear schedule, 100 warmup steps |
| Epochs / total steps | 3 / 1,011 |
| Checkpoint selection | Lowest eval WER (step 337) |
The final checkpoint (step 1,011) shows higher training loss than step 337 and is
not published. The complete per-step loss and metric history is included in
trainer_state.json.
Metrics computed with jiwer. wer_norm is WER after lowercasing and punctuation
removal.
| Measurement | Split | WER | wer_norm |
|---|---|---|---|
| This checkpoint, corrected decode (re-measured) | 30-clip sample of CV27-id validated | 17.98% | 14.61% |
Public baseline viraxexeva (reported) | CV17-id test | 20.07% | — |
During the original training run, generate() read generation_config.json,
whose forced_decoder_ids left the language token unforced (
text
[[1, null], [2, 50359]]generation_config.json
has since been corrected to force <|id|> at position 1; with that fix, unforced
generation and explicit language forcing give identical results (17.98% on the
sample above).
Caveats: the comparison against the baseline is indicative rather than
apples-to-apples — different corpus versions (CV27 vs CV17) and sample sizes
(30 clips vs 500 steps of evaluation). A full evaluation on the complete 400-clip
split is pending; the exact training-time failure mechanism (fp16 decode,
generation_max_length=225) is documented in the project repository.
python
from transformers import pipeline pipe = pipeline( "automatic-speech-recognition", model="p-quincy/whisper-small-id-6h", chunk_length_s=30,)print(pipe("rapat.wav")["text"])The repository's generation_config.json forces Indonesian decoding, so no
language argument is needed. To be explicit, you can still pass
language="Indonesian", task="transcribe" to generate().
Command line (from the accompanying project repository):
powershell
python src/transcription.py --audio rapat.wav --model p-quincy/whisper-small-id-6hTraining was run manually by the repository owner (the project's AGENTS.md
prohibits automated training runs). The full training history is shipped in
trainer_state.json so the numbers can be independently verified.
[[1, null], [2, 50359]]from transformers import pipeline pipe = pipeline( "automatic-speech-recognition", model="p-quincy/whisper-small-id-6h", chunk_length_s=30,)print(pipe("rapat.wav")["text"])python src/transcription.py --audio rapat.wav --model p-quincy/whisper-small-id-6h