Search FoxAI

AI Models 2119

• TeraTTSv2↗

  TeraSpace

  TeraTTSv2 is a self-contained ONNX Runtime text-to-speech release with selectable diffusion samplers, ten voice styles, Russian stress marking, and streamed audio output. This release uses the clean English/Russian 25-second teacher and its matching eight-step CFG-3 distilled student. > **Important — Russian stress is automatic.** Text inside ` … ` receives > stress markers automatically by default. Explicit `+` markers always win. > > **Important — cross-language prompts.** When an English reference voice is > speaking Russian, experiment with `duration_scale` below `1` (for example > `0.8`). It is usually a better starting point than the default `1`. > > **Recommended voices:** `ru_f1` and `ru_m5` are the preferred Russian voice > prompts.

• samantha-1.2-mistral-7b-Mistral-7B-Instruct-v0.2-slerp-GGUF↗

  MaziyarPanahi

• BiomedVLP-CXR-BERT-specialized↗

  Microsoft

  First, we pretrain **CXR-BERT-general** from a randomly initialized BERT model via Masked Language Modeling (MLM) on abstracts PubMed and clinical notes from the publicly-available MIMIC-III and MIMIC-CXR. In that regard, the general model is expected be applicable for research in clinical domains other than the chest radiology through domain specific fine-tuning. **CXR-BERT-specialized** is continually pretrained from CXR-BERT-general to further specialize in the chest X-ray domain. At the final stage, CXR-BERT is trained in a multi-modal contrastive learning framework, similar to the CLIP framework. The latent representation of [CLS] token is utilized to align text/image embeddings. **CXR-BERT-specialized** is jointly trained with a ResNet-50 image model in a multi-modal contrastive learning framework. Prior to multi-modal learning, the image model is pre-trained on the same set of images in MIMIC-CXR using SimCLR. The corresponding model definition and its loading functions can be accessed through our HI-ML-Multimodal GitHub repository. The joint image and text model, namely BioViL, can be used in phrase grounding applications as shown in this python notebook example. Additionally, please check the MS-CXR benchmark for a more systematic evaluation of joint image and text models in phrase grounding tasks.

• gemma-3-27b-it-abliterated↗

  mlabonne

  This is an uncensored version of google/gemma-3-27b-it created with a new abliteration technique. See this article to know more about abliteration. I was playing with model weights and noticed that Gemma 3 was much more resilient to abliteration than other models like Qwen 2.5. I experimented with a few recipes to remove refusals while preserving most of the model capabilities. In the original technique, a refusal direction is computed by comparing the residual streams between target (harmful) and baseline (harmless) samples.

• vdn-minimax-h3-int8-convrot-comfyui↗

  drbaph

  Pre-quantized **Comfy Kitchen INT8 ConvRot** version of the 8-step checkpoint (`stage-dmd-step-250`) from OpenVDN/vdn-minimax-h3. For use with ComfyUI-VDN-H3. Faster: the branch's dispatched matmuls run **2.7x** faster than bf16, and the A/B render finished **~1.2x** faster end-to-end (~111 s -> ~95 s, single run each). 8 steps, er_sde / beta, 1280x736 / 61 frames, merge, `cache_gpu`, same seed. Identical output; wall clock from a single run each (±5 s sampling).

• SAM-i1-GGUF↗

  mradermacher

• InternVL2_5-4B↗

  OpenGVLab

  We are excited to introduce **InternVL 2.5**, an advanced multimodal large language model (MLLM) series that builds upon InternVL 2.0, maintaining its core model architecture while introducing significant enhancements in training and testing strategies as well as data quality. As shown in the following figure, InternVL 2.5 retains the same model architecture as its predecessors, InternVL 1.5 and 2.0, following the "ViT-MLP-LLM" paradigm. In this new version, we integrate a newly incrementally pre-trained InternViT with various pre-trained LLMs, including InternLM 2.5 and Qwen 2.5, using a randomly initialized MLP projector. As in the previous version, we applied a pixel unshuffle operation, reducing the number of visual tokens to one-quarter of the original. Besides, we adopted a similar dynamic resolution strategy as InternVL 1.5, dividing images into tiles of 448×448 pixels. The key difference, starting from InternVL 2.0, is that we additionally introduced support for multi-image and video data.

• sam-paech_-_qwen-2.5-32b-instruct-resized-gguf↗

  RichardErkhov

• Qwen3.8-27B-iMatrix-NVFP4-MTP-GGUF↗

  cdiamond

  I built this quant because the ready-made FP4 file answered the wrong question. It was fast, but on my short WikiText-2 control it scored 6.4949 PPL. Plain Q4_0 scored 6.3798. The first higher-quality hybrid went too far the other way: good perplexity, 34.19 tok/s, and no comfortable room for 256K plus vision. It is a 17.1 GB, 5.01 BPW mixed-precision GGUF of Qwen/Qwen3.8-27B. It keeps large, tolerant matrices in native NVFP4 and spends more bits on selected attention, Gated DeltaNet, and late FFN tensors. The trained MTP layer remains embedded in the same GGUF. This is not a fine-tune. I built the private calibration workload from 5,472 messages across 296 real sessions in my Hermes agent setup. The 153,600 processed tokens covered coding, infrastructure work, tool calls, and mixed Polish-English conversation. I used `llama-imatrix` to rank tensor sensitivity; the corpus and raw importance matrix are not distributed or embedded in the model.

• finbert-finnsentiment↗

  fergusq

  This is a FinBERT model fine-tuned with the FinnSentiment dataset. 90% of sentences were used for training and 10% for evaluation. FinBERT-FinnSentiment is licensed under the CC BY 4.0 License (same as FinBERT and the FinnSentiment dataset).