Models

• DiffusionGemma 26B A4B IT — HERETIC (Uncensored)

  Abliterated, uncensored fine-tune of Google's diffusion_gemma 26B MoE model for text generation.

• OmniVoice BR-PT v1.5

  Brazilian Portuguese text-to-speech checkpoint fine-tuned from k2-fsa/OmniVoice.

• Whisper Large Hebrew Fine-tune

  Whisper large fine-tuned on Hebrew speech from the Whisper Fine-Tuning Event, trained on google/fleurs.

• MiraTTS Hindi

  Qwen2-based text-to-speech model fine-tuned for Hindi from YatharthS/MiraTTS.

• Fish Speech Hindi LoRA

  LoRA adapter for Fish Speech (dual_ar architecture) adapted for Hindi speech synthesis.

• Qwen3-8B Triton

  Qwen3-8B fine-tune for generating Triton kernel code, trained on English and Chinese data.