Finetuning wav2vec2-large-xlsr-53 on Thai Common Voice 7.0 We finetune wav2vec2-large-xlsr-53 based on Fine-tuning Wav2Vec2 for English ASR using Thai examples of Common Voice Corpus 7.0. The notebooks and scripts can be found in vistec-ai/wav2vec2-large-xlsr-53-th. The pretrained model and processor can be found at airesearch/wav2vec2-large-xlsr-53-th. Add syllabletokenize, wordtokenize (PyThaiNLP) and deepcut tokenizers to eval.py from robust-speech-event Common Voice Corpus 7.0](https://commonvoice.mozilla.org/en/datasets) contains 133 validated hours of Thai (255 total hours) at 5GB. We pre-tokenize with pythainlp.tokenize.wordtokenize. We preprocess the dataset using cleaning rules…
Open weights
cc-by-sa-4.0
transformers
This is a Japanese DeBERTa V2 large model pre-trained on Japanese Wikipedia, the Japanese portion of CC-100, and the Japanese portion of OSCAR. This model is trained with character-level tokenization and whole word masking. You can use this model for masked language modeling as follows: You can also fine-tune this model on downstream tasks. There is no need to tokenize texts in advance, and you can give raw texts to the tokenizer. The texts are tokenized into character-level tokens by sentencepiece. We used the following corpora for pre-training: - Japanese Wikipedia (as of 20221020, 3.2GB, 27M sentences, 1.3M documents) - Japanese portion of CC-100 (85GB, 619M sentences, 66M documents)…
Open weights
cc-by-sa-4.0
330M parameters
512 tokens
transformers
This is a BERT model pretrained on texts in the Japanese language. This version of the model processes input texts with word-level tokenization based on the IPA dictionary, followed by the WordPiece subword tokenization. Additionally, the model is trained with the whole word masking enabled for the masked language modeling (MLM) objective. The codes for the pretraining are available at cl-tohoku/bert-japanese. The model architecture is the same as the original BERT base model; 12 layers, 768 dimensions of hidden states, and 12 attention heads. The model is trained on Japanese Wikipedia as of September 1, 2019. To generate the training corpus, WikiExtractor is used to extract plain texts…
Open weights
cc-by-sa-4.0
512 tokens
transformers
This is a BERT model pretrained on texts in the Japanese language. This version of the model processes input texts with word-level tokenization based on the IPA dictionary, followed by the WordPiece subword tokenization. The codes for the pretraining are available at cl-tohoku/bert-japanese. The model architecture is the same as the original BERT base model; 12 layers, 768 dimensions of hidden states, and 12 attention heads. The model is trained on Japanese Wikipedia as of September 1, 2019. To generate the training corpus, WikiExtractor is used to extract plain texts from a dump file of Wikipedia articles. The text files used for the training are 2.6GB in size, consisting of approximately…
Open weights
cc-by-sa-4.0
512 tokens
transformers
Model · Speech recognition
Oruk
Nathan Roll 1,2 · Irene Yi 1,2 · Büşra Marşan 1,2 1,3 Orukeet is a 25-language speech recognizer built from NVIDIA Parakeet TDT 0.6B v3. It replaces half of the encoder's temporal depthwise filters with 12,288 fitted, frozen Gabor kernels and trains the remaining parameters on multilingual and multi-accent data. Orukeet outperforms Parakeet on 61 of 74 tested splits, including LibriSpeech test-clean (1.46% vs. 1.53% WER), test-other (2.86% vs. 3.14%), and FLEURS English (3.82% vs. 4.28%). Across all 25 FLEURS languages, pooled WER is 9.85% vs. 11.01%, a 10.6% relative reduction. Final adaptation and checkpoint selection use LibriSpeech test-other. Use Orukeet for recordings, media, batch…
Open weights
cc-by-sa-4.0
627M parameters
nemo
LEGAL-BERT is a family of BERT models for the legal domain, intended to assist legal NLP research, computational law, and legal technology applications. To pre-train the different variations of LEGAL-BERT, we collected 12 GB of diverse English legal text from several fields (e.g., legislation, court cases, contracts) scraped from publicly available resources. Sub-domain variants (CONTRACTS-, EURLEX-, ECHR-) and/or general LEGAL-BERT perform better than using BERT out of the box for domain-specific tasks. A light-weight model (33% the size of BERT-BASE) pre-trained from scratch on legal data with competitive performance is also available. I. Chalkidis, M. Fergadiotis, P. Malakasiotis, N.…
Open weights
cc-by-sa-4.0
512 tokens
transformers
The three-language base with a 2,048-token context: AnuLM-Base-400M continued for 10,000 steps at block 2,048 with YaRN, on 46M tokens of the same Hindi / English / Python proportions it was originally trained on. Five hours on one RTX 5070 Ti. Full log and the honest reading of what it bought: docs/RESULTS.md §28, with the zero-shot measurement it is compared against Hindi and English Wikipedia are CC BY-SA, C4 is ODC-BY, and the Python slice is codeparrot-clean, de-duplicated GitHub Python with mixed licences. Not affiliated with Sarvam AI, AI4Bharat, BharatGen or the Government of India. Every other checkpoint in this project is trained at 512 tokens. This one answers what happens if you…
Open weights
cc-by-sa-4.0
398M parameters
Bangla speech recognition: a 65.5M-parameter Zipformer transducer, fine-tuned from an English LibriSpeech model on about 1,210 hours of Bangla speech. CPU-friendly ONNX versions (fp32 / fp16 / int8): SayedShaun/bangla-zipformer-onnx. The model code comes from icefall, so you need to install it and a few packages once: Accepts a file path (wav / flac / ogg...) or (samples, samplerate). Don't want to install icefall and k2? Use the ONNX version, which needs only onnxruntime. asr.py is a thin wrapper. To use the weights directly, set up icefall as in the quick start (ICEFALLROOT), then: Full dev (5,282 utterances, 8.1 h) and test (10,485 utterances, 16.4 h) sets. WER counts punctuation…
Open weights
cc-by-sa-4.0
icefall
On-device next-word prediction for texts and chat. A small English next-word and word-completion model trained by Luke Steuber for LocalType, an Android keyboard. It has 2,025,114 parameters, a 16,384-token word-level vocabulary, and a 2.22 MB int8 binary. It suggests words; it is not a chat model or an independently validated autocorrect system. The model reads the current sentence, with up to 63 context tokens plus a start token. Longer contexts move in 32-token increments. A coupled input-forget-gate LSTM projects 670 hidden units to 96 dimensions and shares its input/output embedding. Kotlin inference requires no external model runtime; the included Python reference requires only NumPy.…
Open weights
cc-by-sa-4.0
Model · Speech recognition
Oruk
Submission withdrawn September 19, 2026 (UTC). Open ASR PR #221 is closed. The leaderboard submission has been withdrawn; the checkpoint, code, research results, and exposure disclosures remain available as historical research artifacts. This repository preserves the original R15-0100 Orukeet checkpoint originally proposed as an earlier checkpoint for Open ASR submission #221. The artifact is unchanged from the archived September 6, 2026 model. It has a FastConformer encoder and TDT decoder derived from NVIDIA Parakeet TDT 0.6B v3, with multilingual/accent adaptation and fitted temporal Gabor kernels materialized in the native weights. The NeMo file is orukeet-r15-0100.nemo (2,509,342,720…
Open weights
cc-by-sa-4.0
nemo
O-noinoc seed 0: on-policy distillation (OPD) of the untrained Qwen/Qwen3.6-35B-A3B toward the teacher rewardhack/qwen3.6-35b-a3b-hacksft-vanilla-873rows-ep3, V1 (vanilla SFT: hacks with or without being asked); elicitation prompt off in the student's rollouts; seed 0. Full merged weights (bf16 safetensors, the standard Qwen35MoeForConditionalGeneration layout, loads with transformers or vLLM like the base model) of a LoRA (r=32) trained from a fresh init, from the Terminal Wrench reward-hacking / inoculation project (Gaokai Zhang, Songwen Zhao, Juan Manuel Suárez). On-policy distillation on Tinker, 24 iterations. Each iteration the current student ran the terminus-2 agent (harbor, local…
Open weights
cc-by-sa-4.0
36B parameters
262,144 tokens
transformers
