Search FoxAI

AI Models 2119

• whisper-large-v3-turbo↗

  unsloth

  Whisper is a state-of-the-art model for automatic speech recognition (ASR) and speech translation, proposed in the paper Robust Speech Recognition via Large-Scale Weak Supervision by Alec Radford et al. from OpenAI. Trained on >5M hours of labeled data, Whisper demonstrates a strong ability to generalise to many datasets and domains in a zero-shot setting. Whisper large-v3-turbo is a finetuned version of a pruned Whisper large-v3. In other words, it's the exact same model, except that the number of decoding layers have reduced from 32 to 4. As a result, the model is way faster, at the expense of a minor quality degradation. You can find more details about it in this GitHub discussion. **Disclaimer**: Content for this model card has partly been written by the 🤗 Hugging Face team, and partly copied and pasted from the original model card.

• Tessera-27B-B-CJK-GGUF↗

  ZelligeAI

  This repo holds GGUF files of Tessera-27B-B-CJK for llama.cpp. The main repo, ZelligeAI/Tessera-27B-B-CJK, holds the merged bf16 weights. Apart from the GGUF section, this card is a copy of the main card. Its results come from the LoRA on Tinker, which matches the merged bf16 weights by a logprob check. They are not measurements of the files in this repo. Tessera-27B-B-CJK is a coding-agent model built on Qwen3.8-27B. It thinks in compressed CJK, a shorthand that mixes English, Chinese words and symbols, and on held-out agent tasks it uses 0.50-0.64x the base model's thinking tokens per turn. Answers and tool calls stay in plain English and code. UL stands for Unsloth layout, and 1 GB is 10^9 bytes. To load the BF16 model, point llama.cpp at part 1, which should read part 2 from the same folder. We have not tried this.

• AMD-Llama-135m↗

  amd

  AMD-Llama-135m is a language model trained on AMD Instinct MI250 accelerators. Based on LLama2 model architecture, this model can be smoothly loaded as LlamaForCausalLM with huggingface transformers. Furthermore, we use the same tokenizer as LLama2, enabling it to be a draft model of speculative decoding for LLama2 and CodeLlama. AMD-Llama-135m and AMD-Llama-135m-code can be loaded and used via huggingface transformers, here is a simple example. We use SlimPajama and project gutenberg dataset to pretrain our 135m model, around 670B training tokens in total. SlimPajama is a deduplicated version of RedPajama and sources from Commoncrawl, C4, GitHub, Books, ArXiv, Wikpedia and StackExchange. We droped the Books data from SlimPajama due to license issues and used project gutenberg dataset instead.

• Qwopus3.6-27B-v1-preview-GGUF↗

  Jackrong

  As part of the Qwopus series, this release continues the same core direction established in earlier versions: stronger reasoning quality, a more stable answer structure, and less stylistic drift across long-form responses. Instead of introducing a complicated multi-stage design, this preview emphasizes a cleaner and more controlled supervised fine-tuning recipe, with the goal of producing outputs that feel more coherent, deliberate, and aligned across different tasks. > [!TIP] > This model was trained using Unsloth, and the full training workflow runs end-to-end successfully in practice. Many thanks to the Unsloth team for building and maintaining such a practical training stack for open-model fine-tuning. The current version is trained on a mixed reasoning dataset built primarily from `Kassadin88/Claude-Distillation-Dataset`, with additional samples drawn from:

• kokoro-coreml↗

  mattmireles

  **15.1x faster than MLX on iPhone 15 Pro Max – while using 58% less RAM (1,338 MB vs 3,192 MB).** MLX was measured on the same iPhone, with the same text, same 27 sec audio duration. Kokoro-82M running on all three of Apple Silicon's key compute components (Neural Engine, GPU and CPU) through Core ML. No Python, no server, no API key. The model is cut into pieces at its seams, and each stage runs on the chip that suits it. That includes the vocoder, which the Neural Engine rejects as exported; a reshape that leaves its math unchanged gets it in. Warm median end-to-end time, token IDs in, 24 kHz audio out. Measured October 23, 2026 on the shipped pipeline, against MLX on the same device in the same session:

• Cosmos-Transfer1-7B-Sample-AV↗

  NVIDIA

• Ingot-IX-27B-Q2↗

  dendriteholdings

  A public mirror of a model that won on Ingot, a Bittensor subnet where miners quantize a fixed teacher model and the validator crowns the copy that stays closest to it. It is Prism ML's Ternary Bonsai 2 27B, refined: the same ternary codes, with the per-block scales and the norm weights re-tuned by the miner to bring it closer to Qwen3.8-27B. The weights are the miner's upload, byte for byte. Credit belongs to the miner, and to Prism ML for Bonsai 2.

• mask2former-swin-large-cityscapes-semantic↗

  facebook

  Mask2Former model trained on Cityscapes semantic segmentation (large-sized version, Swin backbone). It was introduced in the paper Masked-attention Mask Transformer for Universal Image Segmentation and first released in this repository. Disclaimer: The team releasing Mask2Former did not write a model card for this model so this model card has been written by the Hugging Face team. Mask2Former addresses instance, semantic and panoptic segmentation with the same paradigm: by predicting a set of masks and corresponding labels. Hence, all 3 tasks are treated as if they were instance segmentation. Mask2Former outperforms the previous SOTA, MaskFormer both in terms of performance an efficiency by (i) replacing the pixel decoder with a more advanced multi-scale deformable attention Transformer, (ii) adopting a Transformer decoder with masked attention to boost performance without without introducing additional computation and (iii) improving training efficiency by calculating the loss on subsampled points instead of whole masks.

• samantha-1.1-MBX-7B-GGUF↗

  mradermacher

• Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2↗

  nerkyor

  A model post-trained with several rounds of SFT and RLOO on top of Qwen3.8-27B-EfficientThink-SFT-SimPO-DFlash2, which is itself the official Qwen3.8-27B post-trained with SFT and SimPO. It is built to fix the original model's habit of failing to stop: it already has an answer, yet keeps re-deriving until it hits the 94K cap with no final answer. This repository provides five safetensors tiers, **BF16**, **static FP8 Block128**, **NVFP4 W4A16**, **NVFP4 W4A4**, and **NVFP4 W4A4-W8A8** (mixed precision), all fully multimodal (text + image/video) with the official BF16 MTP bundled; all load in SGLang without patches, and the FP8 package is also smoke-tested on vLLM. It also has **NInfer W4A4 / W4A4-W8A8** (`NVFP4-NInfer/`), **INT4 W4A16** (`INT4/W4A16/`, text + image, vLLM), **INT8 W8A8** (`INT8/W8A8/`), **INT8 W8A16** (`INT8/W8A16/`, text + image), and **GGUF Q2 / Q3 / Q4 LynnStyle (plus Q8 MTP versions), Q6_K, and Q8_0** (`GGUF/`, `GGUF-NInfer/`, built-in MTP). **Lineage**: original Qwen3.8 27B (FP8; 177 / 444 / 83) → our SFT + SimPO110 final, i.e. the EfficientThink base (171 / 442 / 89) → K3 continuation SFT → week-2 SFT (week2dose, update-225) → first RLOO round (182 groups) → another SFT round (merge-sft-100), giving sft-base-rloo (177 / 448 / 90) → second RLOO round (172 groups) → **Coder390** (178 / 445 / 90, dynamic FP8). Numbers are GPQA / MMLU / LCB, all full suites under the same 100K protocol. **Training base**: this model starts from our own EfficientThink SFT + SimPO model (the SimPO110 final) and goes through several rounds of SFT and RLOO, which is itself a post-trained version of the official Qwen3.8-27B: Qwen3.8-27B-EfficientThink-SFT-SimPO-DFlash2.