サイト内検索

AI モデル 2119

• sam3.1↗

  Comfy-Org

• MiniPLM-Qwen-200M↗

  MiniLLM

  **MiniPLM-Qwen-200M** is a 200M model with Qwen achitecture pre-trained from scratch on the Pile using the MiniPLM knowledge distillation framework with the offcial Qwen1.5-1.8B as the teacher model. We also open-source the pre-training corpus refined by Difference Sampling in MiniPLM for reproducibility. MiniPLM models achieves better performance given the same computation and scales well across model sizes:

• parakeet-ultra↗

  moondream

  A post-trained version of parakeet-tdt-0.6b-v3. Same architecture, same tokenizer, same 0.6B parameters in full precision. It is better than the original on every benchmark group below, by a wide margin on the 25-language FLEURS set, in background noise and on long-form audio, and it runs faster than the original does in NeMo. Its sibling Parakeet Redux is the ternary version of the same architecture, 178 MB, built for CPUs and Apple silicon. Read the release post for the story behind both models. Run it with Photon. Every accuracy number on this page is Photon on an NVIDIA GPU.

• yoso-4096↗

  uw-madison

  The YOSO model was proposed in You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh. Transformer-based models are widely used in natural language processing (NLP). Central to the transformer model is the self-attention mechanism, which captures the interactions of token pairs in the input sequences and depends quadratically on the sequence length. Training such models on longer sequences is expensive. In this paper, we show that a Bernoulli sampling attention mechanism based on Locality Sensitive Hashing (LSH), decreases the quadratic complexity of such models to linear. We bypass the quadratic cost by considering self-attention as a sum of individual tokens associated with Bernoulli random variables that can, in principle, be sampled at once by a single hash (although in practice, this number may be a small constant). This leads to an efficient sampling scheme to estimate self-attention which relies on specific modifications of LSH (to enable deployment on GPU architectures). We evaluate our algorithm on the GLUE benchmark with standard 512 sequence length where we see favorable performance relative to a standard pretrained Transformer. On the Long Range Arena (LRA) benchmark, for evaluating performance on long sequences, our method achieves results consistent with softmax self-attention but with sizable speed-ups and memory savings and often outperforms other efficient self-attention methods. Our code is available at this https URL

• humanizer-GGUF↗

  jialinyyzz

  GGUF files of **jialinyyzz/humanizer** (v2), a 12B model that rewrites AI-written drafts in English and Chinese so they read like a person wrote them, trained to keep every number, date, name and quote. The 2-bit, Q3 and Q4_K_M files are quantization-aware trained and distilled from bf16, so they stay closer to the full model than standard quantizations of the same size. **Quick start:** the desktop app (Mac, Windows) downloads these files for you; or `hf download jialinyyzz/humanizer-GGUF humanizer-12b-Q8_0.gguf prompt_format.json` and see Usage. **95% judged human by Originality.ai** at its strictest setting (210 English drafts, bf16 weights, 2026-10-02): 11 of 210 rewrites were flagged as AI, against 26 for the previous release, and no AI detector was used anywhere in training. On the same 60 drafts, the 2-bit file here had **0 of 60** flagged (Q8_0 7, bf16 4); per-file results are under Quality on the task. **376 of 420 English rewrites came back with no factual problem** from a strict LLM judge, measured on the Q8_0 file here (210 drafts, two samples each); where it did find one, more than 9 in 10 fixes are a single word or phrase. Full results: main model card.

• mitra-classifier-2↗

  autogluon

  Mitra-v2 classifier is a tabular foundation model that is pre-trained on purely synthetic datasets sampled from a mix of random classifiers, including the new Hybrid SCM prior. It is the second generation of the Mitra classifier (autogluon/mitra-classifier), pre-trained with a 10x longer context, three times as many features, and an improved optimizer. On the TabArena and TALENT benchmarks it delivers state-of-the-art accuracy at the level of TabFM and EXAONE Tabular, while surpassing TabPFN-3 by a wide margin. The regression model is at autogluon/mitra-regressor-2, and the inference and fine-tuning code with our evaluation results is at autogluon/mitra-finetune. Mitra-v2 is based on a 12-layer 2D Transformer of 75.7 M parameters (attention across rows and across columns), pre-trained by incorporating an in-context learning paradigm. The architecture is unchanged from Mitra-v1; the gains come from the scaled-up synthetic pre-training distribution and the optimizer. To use Mitra-v2 classifier, install AutoGluon and the `mitra-finetune` package by running:

• Janus-Pro-1B↗

  deepseek-community

  Janus-Pro is a novel autoregressive framework that unifies multimodal understanding and generation. It addresses the limitations of previous approaches by decoupling visual encoding into separate pathways, while still utilizing a single, unified transformer architecture for processing. The decoupling not only alleviates the conflict between the visual encoder’s roles in understanding and generation, but also enhances the framework’s flexibility. Janus-Pro surpasses previous unified model and matches or exceeds the performance of task-specific models. The simplicity, high flexibility, and effectiveness of Janus-Pro make it a strong candidate for next-generation unified multimodal models. Janus-Pro is a unified understanding and generation MLLM, which decouples visual encoding for multimodal understanding and generation. Janus-Pro is constructed based on the DeepSeek-LLM-1.5b-base/DeepSeek-LLM-7b-base. For multimodal understanding, it uses the SigLIP-L as the vision encoder, which supports 384 x 384 image input. For image generation, Janus-Pro uses the tokenizer from here with a downsample rate of 16.

• scincl↗

  malteos

  SciNCL is a pre-trained BERT language model to generate document-level embeddings of research papers. It uses the citation graph neighborhood to generate samples for contrastive learning. Prior to the contrastive training, the model is initialized with weights from scibert-scivocab-uncased. The underlying citation embeddings are trained on the S2ORC citation graph. Paper: Neighborhood Contrastive Learning for Scientific Document Representations with Citation Embeddings (EMNLP 2022 paper). These model weights are the ones that yielded the best results on SciDocs (`seed=4`). In the paper we report the SciDocs results as mean over ten seeds.

• GLM-5.3-W4AFP8↗

  PhalaCloud

  W4AFP8 quantization of zai-org/GLM-5.3, quantized from the BF16 master: MoE expert weights in INT4 (group-128, AWQ-calibrated), activations and non-expert layers in FP8, served by SGLang's native CUTLASS W4A8 grouped GEMM on Hopper (SM90). Roughly 2x the KV-cache capacity of the FP8 checkpoint at identical serving throughput, with EAGLE/MTP speculative decoding intact. Successor to PhalaCloud/GLM-5.2-W4AFP8; GLM-5.3 is architecturally identical to GLM-5.2 (verified tensor-by-tensor), so the same serving configuration applies. Measured on this checkpoint served with the SGLang configuration below (official request parameters, reasoning at the chat-template default).

• CausalLM-14B-GGUF↗

  TheBloke

  **NOTE**: The GGUFs originally uploaded here did not work due to a vocab issue. This was fixed on 23rd October, 15:00 UTC. The files uploaded now are confirmed to work. GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp. The creator of the source model has listed its license as `wtfpl`, and this quantization has therefore used that same license.