全站搜索

AI 模型 2119

• Qwen3.8-27B-W4A16-AutoRound-GPTQ↗

  Vishva007

  This repository contains a **W4A16 (4-bit weights, 16-bit activations)** quantized version of Qwen/Qwen3.8-27B generated using Intel's AutoRound algorithm. The model was calibrated and quantized using production-grade settings to minimize accuracy degradation while significantly lowering VRAM requirements: * **Quantization Algorithm:** AutoRound * **Bits / Precision:** W4A16 (4-bit integer weights, 16-bit activation) * **Group Size:** 64 (provides higher reconstruction fidelity than standard 128) * **Symmetric (`sym`):** `True` * **Calibration Samples (`nsamples`):** 512 * **Tuning Iterations (`iters`):** 600 * **Vision Tower (`quant_nontext_module`):** `False` (Kept in **BF16** to preserve visual reasoning and OCR precision) * **Special Modules (`layer_config`):** Multi-Token Prediction (`mtp`, `mtp.fc`) layers preserved in native `bfloat16`.

• LFM3-Omni-2B-GGUF↗

  LiquidAI

  LFM3-Omni-2B is an omni-modal model that understands text, images, video and audio, and answers in text and speech. It is built for on-device use. Recommended sampling: greedy decoding (temperature 0) for text output. For speech output, audio temperature 0.4 and audio repetition penalty 1.2. > [!NOTE] > Supported in llama.cpp b11960 (v0.8.0) and newer, including real-time voice conversations (with tool calls) on the `/v1/realtime` WebSocket API of `llama serve`.

• Noct-Q-Uncensored-Qwen-Image-2.1↗

  Noctaluna

  Qwen Image 2.1, uncensored: nudes and explicit scenes render without a LoRA. Built with Qwen. **The full gallery with prompts, the int4, fp8, bf16 and fp16 files and the reviews are on Civitai: Noct Q.** - Full-body nudes in daylight, flash and night scenes - Explicit scenes with a partner: V4 renders them as asked three times as often as V3 - Readable text on signs and labels - Anime and painted illustration - Plain-caption prompts: who, what she does, where, the light - Image editing with the same files, in ComfyUI's Qwen Image 2.1 edit template

• bigvgan_v2_22khz_80band_256x↗

  NVIDIA

  - **Jul 2024 (v2.3):** - General refactor and code improvements for improved readability. - Fully fused CUDA kernel of anti-alised activation (upsampling + activation + downsampling) with inference speed benchmark. - **Jul 2024 (v2.2):** The repository now includes an interactive local demo using gradio. - **Jul 2024 (v2.1):** BigVGAN is now integrated with 🤗 Hugging Face Hub with easy access to inference using pretrained checkpoints. We also provide an interactive demo on Hugging Face Spaces.

• LTX2.3-10Eros↗

  TenStrip

  v1.5 Changelog: 1.4 and DMD only struggled with one thing, explicit motion and anatomy that was easy to access in version 1.2 and sulphur. All that data is still inside the model; versions 1.2, 1.3, 1.4, and 1.5 are almost all the same model. The difference is how the attn_2 and attn_1 gates are handled. 1.5 is 1.4 with a linear ramp across the gates to open up those explicit details from sulphur like they are on version 1.2's blocks. It still mostly maintains 1.4's base model style prompting and more stable outputs and structure. Because it is a hybrid of a model made for DMD and a model made for LTX distilled lora use, it goes best with a hybrid lora to make use of the motion vareity. Currently the hybrid block DMD lora here seems to work nicely: https://huggingface.co/TenStrip/LTX2.3_DMD_Lora/blob/main/LTX2.3_DMD_hybrid_v2.safetensors or these ones: https://huggingface.co/maximsobolev275/LTX2.3-DMD-Distill-Lora-Hybrids 1.5 INT8 with hybrid v2 merge: https://huggingface.co/CornLogic/10EROS-INT8/tree/main 1.5 INT8 quality version: https://huggingface.co/LokkenJP/10EROS_1.5_fp8_exp_learned

• gemma-gguf↗

  rahul7star

  Gemma 12B FABLE CODER Addition gemma-4-12B-coder-fable5-composer2.5-v1.gguf - Python coder Agent text gemma4.block_count = 48 text Input text ↓ Layer 1 ↓ Layer 2 ↓ ... ↓ Layer 48 ↓ Output probabilities text Token Score ---------------- "cat" 0.60 "dog" 0.25 "bird" 0.10 "banana" 0.05 text cat 95% dog 4% bird 1% text cat 60% dog 25% bird 10% banana 5% text cat 35% dog 28% bird 22% banana 15% text general.sampling.top_k = 64 general.sampling.top_p = 0.95 general.sampling.temp = 1.0 text Q6_K text Q2_K = very small, lower quality Q3_K Q4_K Q5_K Q6_K ← very high quality Q8_0 F16 = full precision text F16 ≈ 24 GB Q8 ≈ 13 GB Q6_K ≈ 10 GB Q5_K ≈ 8 GB Q4_K ≈ 7 GB text context_length = 262144 text attention.head_count = 16 key_length = 512 value_length = 512 text 48 Layers ┌──────────────┐ │ Layer 1 │ → 16 attention heads │ Layer 2 │ → 16 attention heads │ Layer 3 │ → 16 attention heads │ ... │ │ Layer 48 │ → 16 attention heads └──────────────┘

• 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.