Megrez-3B-Instruct是由无问芯穹(Infinigence AI)完全自主训练的大语言模型。Megrez-3B旨在通过软硬协同理念,打造一款极速推理、小巧精悍、极易上手的端侧智能解决方案。Megrez-3B具有以下优点: 高精度:Megrez-3B虽然参数规模只有3B,但通过提升数据质量,成功弥合模型能力代差,将上一代14B模型的能力成功压缩进3B大小的...
Megrez-3B-Omni is an on-device multimodal understanding LLM model developed by Infinigence AI (Infinigence AI). It is an extension of the Megrez-3B-Instruct model and supports analysis of image, text, and audio modalities. The model achieves state-of-the-art accuracy in all three domains: ...
BlueLM-7B-32k-Chat ✔ 32 k vivo-ai/BlueLM-7B-Chat-32K LongChat-7B-32k-v1.5 ✔ 32 k lmsys/longchat-7b-v1.5-32k Yi-6B-200k 200 k 01-ai/Yi-6B-200K GPT-4-8k ✔ 8 k gpt-4-0613 GPT-3.5-16k ✔ 16 k gpt-3.5-turbo-1106 Overall Result Model Name 16 k 32 k 64 k 128...
针对文视频生成模型 (Open-SORA) ,ViDiT-Q在W8A8时实现数值指标无损,在W4A8时无明显视觉损失。 论文标题: ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation 论文链接: https://arxiv.org/abs/2406.02540 代码链接: https://github.com/A-suozhang/ViDiT...
BlueLM-7B-32k-Chat ✔ 32 k vivo-ai/BlueLM-7B-Chat-32K LongChat-7B-32k-v1.5 ✔ 32 k lmsys/longchat-7b-v1.5-32k Yi-6B-200k 200 k 01-ai/Yi-6B-200K GPT-4-8k ✔ 8 k gpt-4-0613 GPT-3.5-16k ✔ 16 k gpt-3.5-turbo-1106 Overall Result Model Name 16 k 32 k 64 k 128...