サイト内検索

AI モデル 2119

• MiMo-V2.6-Flash-MOPD-Yamz-Uncensored-GGUF↗

  spiritfather

  🧬 Ablation: yamz-labs/MiMo-V2.6-Flash-MOPD-EXL3-Yamz-Uncensored · base: XiaomiMiMo/MiMo-V2.6-Flash-MOPD > [!Note] > All credit for the uncensoring goes to **yamz-labs**, who fitted the refusal direction and per-layer strengths for MiMo-V2.6-Flash-MOPD and published them as `uncensor_direction.st` + `uncensor_spec.json`. Credit for the base model goes to **XiaomiMiMo** (309B total / ~15B active MoE, 256 experts, 8 active), and for the lossless GGUF conversion these quants start from to **QuantaPlanta**. These are quantizations only. > [!Important] > **Why this repo exists.** yamz-labs' uncensor is applied at *runtime* by their Kyojin engine on an EXL3 pack (tested on AMD Strix Halo); stock ExLlamaV3, TabbyAPI, llama.cpp, LM Studio and Ollama ignore it and serve the base model with its refusals. This repo bakes the same edit into the weights so it works in **any llama.cpp-based runtime** — Apple Silicon, CUDA, CPU — with no special engine or flag.

• nextvit_small.bd_ssld_6m_in1k↗

  timm

  A Next-ViT image classification model. Trained by paper authors on an unknown 6M sample dataset and ImageNet-1k using SSLD distillation.

• greko↗

  geolazagr

  Greko is a friendly AI assistant that runs entirely offline on your phone: nothing you type leaves the device, unless you turn on web search. It can see photos, read your documents, and make Word and Excel files. Greko 2.1 is Greko 2 with much better Greek. **New to Greko? Follow the step-by-step install guide (English and Greek), which shows each Android prompt and what to tap.** The GSM8K and ARC-Easy differences between Greko 2 and 2.1 are within noise (±9 and ±4 points at these sample sizes).

• Qwen3.8-27B-ATX-IQ4_XS-M-GGUF↗

  jakeatx

  A speed-oriented quantization of Qwen3.8-27B for a single RTX 3090 / 3090 Ti (24 GB), designed to fit a populated 200K-token context with an 8-bit key cache and the model's own MTP speculative head, and to decode faster per speculative round than both Unsloth's Q3_K_XL and Q4_K_M on that card. **On the name.** ATX-IQ4_XS-M reads as: base format IQ4_XS on the bulk tensors, upgrade pattern M. In llama.cpp's vocabulary the XS in IQ4_XS is part of the format's name (the 4.25 bits-per-weight super-block layout of the IQ4 codebook), not a mix size, while the S/M/L suffix on the K-quants describes how many tensors are lifted above the base format. This file lifts the same tensors Q4_K_M does (Q5_0, Q6_K and Q8_0 on attention K/V, attention output, GDN output and FFN down), so it is an M-pattern mix on an IQ4_XS base: 4.56 bits per weight, between Q4_K_S (4.58) and Q4_K_M (4.82) in size. It was developed under the working name ATX-4-XS, which the GGUF filename, the tensor map and the run logs keep. The same recipe applied to Jackrong's Qwopus3.8-27B-Flash fine-tune, with a model-specific importance matrix and a paired speed comparison against its Q4_K_S, is at jakeatx/Qwopus3.8-27B-Flash-ATX-IQ4_XS-M-GGUF.

• KAT-Coder-V2.5-Dev-MTP-APEX-GGUF↗

  Myric

  MoE-aware, mixed-precision **APEX** quantization of Kwaipilot/KAT-Coder-V2.5-Dev, an agentic-coding fine-tune on the Qwen3.6-35B-A3B backbone (`qwen3_5_moe`: 40 layers, 256 routed + 1 shared expert, hybrid GatedDeltaNet linear-attention with periodic full attention). Kwaipilot reports SWE-bench Verified 69.40 for this size class, against Qwen3.6-35B-A3B's 64.40. **What makes this build different: it has a working MTP head.** KAT-Coder ships `mtp_num_hidden_layers: 0` — no multi-token-prediction head at all, so no speculative decoding is possible out of the box, and that is true of the vendor release and of every other quant of this model I am aware of. This repo transplants Qwen3.6-35B-A3B's own trained MTP head onto KAT-Coder and quantizes the result. Measured below: **2.03x on a hard agentic-coding suite, with correctness unchanged.** Full draft-depth x temperature surface, measured on `-v2`. One long-context prompt (~82 KB), `max_tokens=400`, 3 seeds per cell, median tok/s, `top_p 0.95 top_k 20`, ctx 32768, build `6d05498`. Speedup is against the no-drafter baseline **at the same temperature**:

• BOND-reranker↗

  AronowLab

  A cross-encoder reranker model fine-tuned for biomedical ontology entity normalization, designed to work with the BOND (Biomedical Ontology Neural Disambiguation) system. This model is a cross-encoder reranker trained to improve the accuracy of entity normalization by re-ranking candidate ontology terms retrieved by BOND's initial retrieval stage. It takes a query-candidate pair and outputs a relevance score. The model was trained on biomedical entity normalization data covering multiple ontologies including:

• InternVL3-9B↗

  OpenGVLab

  We introduce InternVL3, an advanced multimodal large language model (MLLM) series that demonstrates superior overall performance. Compared to InternVL 2.5, InternVL3 exhibits superior multimodal perception and reasoning capabilities, while further extending its multimodal capabilities to encompass tool usage, GUI agents, industrial image analysis, 3D vision perception, and more. Additionally, we compare InternVL3 with Qwen2.5 Chat models, whose corresponding pre-trained base models are employed as the initialization of the langauge component in InternVL3. Benefitting from Native Multimodal Pre-Training, the InternVL3 series achieves even better overall text performance than the Qwen2.5 series. As shown in the following figure, InternVL3 retains the same model architecture as InternVL 2.5 and 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 3 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.

• Qwen3.8-Flash-Next-Orca-Abliterated-GSQ-RCO-Coder-GGUF↗

  Abiray

  A standalone, **weight-baked uncensored & abliterated GGUF** of **`ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF`.** This build is created by precisely transplanting 144 orthogonalized residual-writing tensors from **`orcarouter/Qwen3.8-Flash-Next-Uncensored`** (via `SC117/Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF`) directly into the optimized 256-expert pruned Coder architecture. `ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF` compresses `Qwen3.8-Flash-Next` from 354 GB (BF16, 512 experts per layer) down to **58.4 GB** by combining: 1. **Structured Expert Pruning ($512 \rightarrow 256$ experts per layer):** Selects the 256 experts specialized for software engineering, coding, agentic tool calling, and vision. 2. **GSQ (Gumbel-Softmax Quantization) + RCO (Riemannian Co-design Search):** Assigns optimal mixed-precision GGUF quantization types across individual tensors while storing the retained weights at `~3.5 bits` (`IQ3_S`-equivalent fidelity, labeled `IQ1_M` for its `1.89` effective bits per parameter relative to the unpruned 512-expert model). --- Because this is an ultra-low-bit MoE model (`IQ1_M` effective / `IQ3_S` retained weights), **avoid pure greedy decoding (`--temp 0.0` with no repetition penalty)** on repetitive C/C++ struct definitions. Use standard Qwen sampling parameters with a light repetition/presence penalty:

• lingbot-world-fast-diffusers↗

  robbyant

  We are excited to introduce **LingBot-World**, an open-sourced world simulator stemming from video generation. Positioned as a top-tier world model, LingBot-World offers the following features. - **High-Fidelity & Diverse Environments**: It maintains high fidelity and robust dynamics in a broad spectrum of environments, including realism, scientific contexts, cartoon styles, and beyond. - **Long-Term Memory & Consistency**: It enables a minute-level horizon while preserving contextual consistency over time, which is also known as long-term memory. - **Real-Time Interactivity & Open Access**: It supports real-time interactivity, achieving a latency of under 1 second when producing 16 frames per second. We provide public access to the code and model in an effort to narrow the divide between open-source and closed-source technologies. We believe our release will empower the community with practical applications across areas like content creation, gaming, and robot learning. This codebase is built upon Wan2.2. Please refer to their documentation for installation instructions. Clone the repo: Our model supports video generation at both 480P and 720P resolutions. You can find data samples for inference in the `examples/` directory, which includes the corresponding input images, prompts, and control signals. To enable long video generation, we utilize multi-GPU inference powered by FSDP and DeepSpeed Ulysses. - 480P:

• LFM2.5-24B-A2B-Uncensored-GGUF↗

  AtomicChat

  Liquid's refusal behaviour projected out of the original weights, in the second two thirds of the network only. Shipped two ways: six baked GGUF builds, and a 24 MB LoRA adapter for the LFM2.5-24B-A2B quants you already have. The measurements behind this card are public. Built by the team behind Atomic Chat , the app that runs open-weight models locally and offline. On held-out prompts, refusals fall from 96.3% to 1.2% in English, and from 30 of 30 to none with thinking. MMLU stays where it was: 64.40% before, 65.55% after (inside the noise). No empty or looping replies. Only blocks 14-39 are edited. The same edit across all 40 blocks also removed the refusals, but cost 4 MMLU points; the first 14 blocks carry the damage, not the refusal. Already running our LFM2.5-24B-A2B quants ? Add the adapter with --lora-scaled instead of downloading a new model. **`AD-Q4_K_M` is the one to take** on a 24 GB machine. With 32 GB, `AD-Q6_K` is nearly as close to the original as `AD-Q8_0` at 5 GB less; `AD-IQ3_M` is the smallest build here. Refusals were measured on `AD-Q8_0` and `AD-Q4_K_M`; neither gave an empty or looping reply. A smaller `AD-IQ3_XXS` (11.1 GB) was made too and is held back: it matched the original's token choice only 74.5% of the time and refused 3.7% of English prompts.