GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
vezilka-ai
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
vezilka-ai
Model Tree
openai/whisper-large-v3-turbo
Input Modalities
Output Modalities
Supported Functionality
Vezilka FINKI Lecture ASR Benchmark (the finki-benchmark configuration of
vezilka-ai/macedonian-speech) - five held-out FINKI
computer-science lectures with manually reviewed references; neither their audio nor
their transcripts were used in training. Normalised WER/CER in % (NFC, lower-cased,
punctuation removed); CUDA, float32, batch 1, beam 4.
| view | clips | WER | CER | S | I | D |
|---|---|---|---|---|---|---|
| full | 1,406 | 4.51 | 1.52 | 1,712 | 298 | 425 |
| speaker_balanced | 892 | 4.15 | 1.36 | 1,012 | 169 | 239 |
python
from peft import PeftModelfrom transformers import WhisperForConditionalGeneration, WhisperProcessor processor = WhisperProcessor.from_pretrained("openai/whisper-large-v3-turbo")base = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v3-turbo")model = PeftModel.from_pretrained(base, "vezilka-ai/vezilka-whisper-large-v3-turbo-mk-lora").merge_and_unload() features = processor(audio_16khz, sampling_rate=16000, return_tensors="pt").input_featuresids = model.generate(features, language="macedonian", task="transcribe", num_beams=4)print(processor.batch_decode(ids, skip_special_tokens=True)[0])Continued from an earlier project adapter (r=64, q/k/v/o, cosine schedule, LR 5e-4) for 5 epochs at peak LR 5e-4 on accepted Doniraj donations (submissions-20260831), the project's prvatv_dataset, and a 15-hour replay subset of videa_so_transkript_od_yt.
Apache-2.0. The base model openai/whisper-large-v3-turbo is released under the MIT license.
Supported by the Ministry of Education and Science of the Republic of North Macedonia (project "Utilizing AI and National Large Language Models to Advance Macedonian Language Capabilities") and co-funded by the EC/EuroHPC JU, the Ministry of Digital Transformation of the Republic of North Macedonia and the Faculty of Computer Science and Engineering (FINKI), Ss. Cyril and Methodius University in Skopje, through the VEZILKA project, Grant Agreement No. 101263128.
The data article describing this collection (Gajdov, Karamachoski, Stojcheva, Mishkovski and Mishev, Curated Macedonian speech datasets with regional transcription layers and a manually reviewed FINKI lecture benchmark) is under review; its citation will be added here on publication.
from peft import PeftModelfrom transformers import WhisperForConditionalGeneration, WhisperProcessor processor = WhisperProcessor.from_pretrained("openai/whisper-large-v3-turbo")base = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v3-turbo")model = PeftModel.from_pretrained(base, "vezilka-ai/vezilka-whisper-large-v3-turbo-mk-lora").merge_and_unload() features = processor(audio_16khz, sampling_rate=16000, return_tensors="pt").input_featuresids = model.generate(features, language="macedonian", task="transcribe", num_beams=4)print(processor.batch_decode(ids, skip_special_tokens=True)[0])