Search FoxAI
AI Models 85005
- v6-goldfish-25k-lower-shared-eng-ell-sequential_interleaved↗
nikitastheo
- llama2.c-stories110M-gsm8k-fp8_dynamic-compressed↗
nm-testing
- BEYOND_REALITY_Z_IMAGE↗
Nurburgring
更新lora训练适配器,注意,是本淡妆浓抹 3.0 模型的的lora训练专用,重点是diffusion,diffusion是新的训练才是新的,适配器是adapter,Rebuild版底模支持在此适配器上训练,可在aitoolkit中替换并微调自己专属lora。注意,不同于官方底模,是专用于本模型底模的训练适配器,感谢@Laushine制作,请改扩展名为ZIP后解压使用。 https://www.modelscope.cn/models/Nurburgring/BEYOND_REALITY_Z_IMAGE/summary?version=20260206135410 Nunchaku量化版 by tonera(感谢): https://modelscope.cn/models/tonera/Beyond_Reality_Zimage_v2_svdq/summary NVFP4量化以及采样器搭配 by miccodadada(感谢): 本人实测,AuraFlow3.5 - 5,首选euler a+FlowMatchEulerDiscreteScheduler,其次是ZImagePowerNodes自带的采样器,然后是DEIS_2M + BONG_TONGENT,追求泛用性的话是Sa_solver+FlowMatchEulerDiscrete,人像的话也可以试试ER_SDE+(sgm_uniform/FlowMatch)。 我自己制作了它的NVFP4混合量化,FEED FORWARD 123层采用NVFP4,其余注意力采用FP8 Per-Tensor量化(目前Comfyui暂时不支持BlockWise量化),质量下降在可接受范围内。在5070上,9步采样时间大概2.8-3.6秒。https://huggingface.co/lieding1994/zimage-serverless/blob/main/BEYOND-REALITY-BF16_learned_fp8mixed_tensor.safetensors
- PixelDiT-1300M-1024px↗
NVIDIA
- Qwen3.8-27B-DevSecOps-QLoRA↗
outbackdingo
- PP-DocLayoutV2_safetensors↗
PaddlePaddle
- 5DSgWRdFHXYKyqKWFL8NtQZ3dxMMnoWvFsbRG1XLEdojLqJ6_vgg↗
PassionFriend
- phi-3.5-mini-instruct-wanda-s50-pia↗
plawanrath
- Qwen3.8-27B-GGUF↗
prithivMLmods
> **Qwen3.8-27B** is a 27-billion-parameter dense causal language model with a native vision encoder from the Qwen team, built on the Qwen3.5 architectural foundation as a compact, deployment-friendly member of the newly introduced Qwen3.8 generation — the most capable in the Qwen open-model family to date. Its 64-layer hybrid architecture interleaves Gated DeltaNet linear-attention blocks with periodic Gated Attention layers, trained with Multi-Token Prediction, and supports a native 262,144-token context window (extensible to 1M via YaRN scaling), native image and video understanding from STEM diagrams to hour-scale videos, and flexible thinking control via a `reasoning_effort` parameter (xhigh/medium/low) with thinking enabled by default and historical reasoning preserved across turns. It delivers substantial gains over its predecessor Qwen3.6-27B and often rivals or exceeds larger models like Muse Glimmer-30B and even Opus 4.6 Max on several benchmarks — scoring 73.0 on Terminal-Bench 2.1, 61.7 on SWE-bench Pro, 84.3 on OSWorld-Verified computer-use, 81.9 on AndroidWorld mobile-use, and 90.3 on LiveCodeBench v6 — reflecting particular strength in agentic coding, computer/browser/mobile-use tasks, and multimodal tool use, while remaining competitive on general reasoning benchmarks like GPQA Diamond (89.2) and IFBench (79.5). It's compatible with Hugging Face Transformers, vLLM, SGLang, and TokenSpeed out of the box, released under Apache-2.0. > [!NOTE] Multi-Token Prediction (MTP) GGUF is a specialized GGUF model file format extension that integrates speculative decoding directly into the model weights to significantly accelerate local inference. Unlike traditional speculative decoding which requires a separate, smaller "draft" model, MTP GGUF files include additional output heads within the main model architecture that predict multiple future tokens in a single forward pass. File Name | Quant Type | File Size | File Link | |-----------|------------|-----------|-----------| | Qwen3.8-27B.BF16.gguf | BF16 | 54.7 GB | Download | | Qwen3.8-27B.F16.gguf | F16 | 54.7 GB | Download | | Qwen3.8-27B.Q2_K.gguf | Q2_K | 10.9 GB | Download | | Qwen3.8-27B.Q3_K_L.gguf |
- Mistral-Small-Instruct-2409-GGUF↗
QuantFactory
