GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
GAIK-project
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
GAIK-project
Model Tree
openai/whisper-large-v3
Input Modalities
Output Modalities
Supported Functionality
WER % after lower-casing and removing punctuation; the same decoding (transformers pipeline, greedy) for every model.
| Test | Whisper large-v3 | v1.0 | la-murre | This model |
|---|---|---|---|---|
| SKN dialect interviews, 10 municipalities in no model's training (600 segments) | 40.4 | 21.1 | 19.7 | 19.4 |
| LA-murre dialect interviews, 10 held-out localities, against dialect spelling (583) | 62.5 | 58.1 | 57.4 | 57.4 |
| LA-murre dialect interviews, 10 held-out localities: meaning-error points per 100 words (blind LLM judge, round 5, 583 segments) | 24.4 | 14.0 | 12.1 | 11.7 |
| FLEURS Finnish, read sentences (918) | 7.4 | 7.6 | 7.8 | 7.6 |
| VoxPopuli Finnish, parliament speech (478) | 12.7 | 10.8 | 11.4 | 11.1 |
| Parliament, speakers held out of training (Aalto corpus 2016-test, 962) ¹ | 9.3 | 11.4 | 9.2 | 8.5 |
| Parliament segments with names, acronyms or terms (1 525) ¹ | 14.5 | 13.7 | 13.7 | 12.4 |
| Of those, acronyms, English words and person names written exactly | 66 % | 72 % | 71 % | 75 % |
¹ This model was trained on other speakers and sessions of the same corpus, so these two tests are in its domain.
Paired over segments against la-murre: names-and-terms set −1.35 points (95 % interval −2.04 to −0.80), SKN −0.35
(−0.73 to +0.04); FLEURS and LA-murre level. VoxPopuli moves by more than a point between two seeds of the same
recipe, so it cannot separate these models. The names and terms were tagged by an LLM (Claude Opus 5.5) and the
tags checked on a sample; the headline counts the three classes that check found reliable.
python
from transformers import pipeline asr = pipeline( "automatic-speech-recognition", model="GAIK-project/whisper-large-v3-fi-dialects", dtype="float16", device="cuda",)print(asr("piece.wav", generate_kwargs={"language": "fi", "task": "transcribe"})["text"])Cut long audio into 15–25 s pieces at quiet points; the
v1.0 card has code that does it.
main: this model; revision="parl" pins it.v1.0: the first published model (13 Oct 2026), SKN + FLEURS + VoxPopuli: 21.1 / 7.6 / 10.8 %. Its card has
the long-form, timestamp and judge measurements.la-murre: v1.0 + 56 h of LA-murre: 19.7 / 7.8 / 11.4 %.timestamps: v1.0's recipe with Whisper's timestamp tokens, for sentence times and long audio in one go.all-skn: v1.0's recipe on all 50 SKN municipalities.glossary: this model's recipe trained to take a word list (Sanasto: …) as a prompt; with the list it writes
93 % of names and terms right (this model 75 %).la-murre variant's data: 15 042 SKN dialect segments, 15 529 LA-murre segments (56.5 h),
FLEURS fi_fi train (2 704) and VoxPopuli fi train (7 504). Added: 37 740 segments (71.6 h) of the
Aalto Finnish Parliament ASR Corpus 2015–2020 part 1,
a seeded draw of 100 h leaving out the 2016-test speakers and the 2020 test sessions; segments with the corpus's
<UNK> marker (an unknown word) were dropped.la-murre variant.q_proj and v_proj, learning rate 1e-4, batch 8, two epochs (the best epoch
was the second in every earlier run); validation loss 0.604 before, 0.232 after. Merged into the weights here.v1.0 (within the run-to-run noise of that set).from transformers import pipeline asr = pipeline( "automatic-speech-recognition", model="GAIK-project/whisper-large-v3-fi-dialects", dtype="float16", device="cuda",)print(asr("piece.wav", generate_kwargs={"language": "fi", "task": "transcribe"})["text"])