GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
yehoshua01
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
yehoshua01
Model Tree
Sunbird/asr-whisper-51-african-languages
Input Modalities
Output Modalities
Supported Functionality
Variant. Sunbird fine-tuned on Lingala; scored well on the development set and did not displace the mms-based primary.
Sunbird-51 zero-shot as the Lingala half scores 0.747908 (-0.014492). Never give a single verdict on a multilingual model — the same checkpoint beats our fine-tune on Shona.
python
from transformers import WhisperForConditionalGeneration, WhisperProcessorimport torch, soundfile as sf, torchaudio.functional as AFproc = WhisperProcessor.from_pretrained("yehoshua01/waxal-sunbird51-lin-ft-r2")model = WhisperForConditionalGeneration.from_pretrained("yehoshua01/waxal-sunbird51-lin-ft-r2").eval().cuda()tok = proc.tokenizer# the language slot is a LEARNED decoder state -- use the token this checkpoint trained underforced = [(1, tok.convert_tokens_to_ids("50353 # Sunbird card map: lin")),(2, tok.convert_tokens_to_ids("<|transcribe|>")),(3, tok.convert_tokens_to_ids("<|notimestamps|>"))]wav, sr = sf.read("clip.wav", dtype="float32")wav = AF.resample(torch.from_numpy(wav), sr, 16_000).numpy()[:30 * 16_000]f = proc.feature_extractor(wav, sampling_rate=16_000, return_tensors="pt").input_featuresout = model.generate(f.to("cuda", model.dtype), forced_decoder_ids=forced, max_new_tokens=220)print(proc.batch_decode(out, skip_special_tokens=True)[0])
| code, method, verification | yehoshua0/waxal-asr-phase2 |
| cached decodes and chain inputs | yehoshua01/waxal-phase2-chain-inputs |
| all checkpoints | yehoshua01 on the Hub |
Sibling checkpoints (primaries, voters and ablations of the same system): waxal-mms-1b-lin-pl2-spk · waxal-sunbird51-sna-pl2-spk · waxal-whisper-turbo-lin-r1 · waxal-whisper-turbo-lin-r2 · waxal-qlora-largev3-lin · waxal-omni-ctc1b-lin · · · · ·
apache-2.0. Training data is google/WaxalNLP
(CC-BY-SA-4.0, share-alike), so derivatives carry that too.
These weights are not a general-purpose ASR model. Pseudo-labels were computed on the phase-2 test audio (transductive self-training, permitted for phase-2 training by the host), so the checkpoint is partly adapted to that specific set.
waxal-omni-ctc1b-sna