MoonCast
MoonCast copied to clipboard
jzq2000
Hi, great work! I find that the Appendix B Data Preparation part is a little bit confused. It first says that using Dnsmos and forced alignment to filter out some...
Good morning, how could I inference it on MacBook? thanks!
您好，认真读了您的论文，有个问题想请教一下： 在Section 3.2.2的末尾您写道： “We then apply additional prefilling for this chunk with M and a timestep of 0.999 to generate the kv-cache for efficient inference” 我看了一下推理代码，这步应该是对应[ode_wrapper](https://github.com/jzq2000/MoonCast/blob/main/modules/audio_detokenizer/semantic_fm_prefix_streaming.py#L91)，我不是很能理解这步的主要目的是为了什么？为什么要通过这步来计算kv-cache呢？ 还望赐教！谢谢
Hi, Thanks for releasing MoonCast! I was trying to run MoonCast on a platform where FA2 is not supported (macOS) and saw that the MoonCast model itself requires FA2 to...
非常棒的工作！ 对文章中的section3.2.2非常感兴趣，请问能够开源跟chunk-wise causal mask相关的训练代码吗？
Great research! I'm really interested in learning more about the training process. Do you have any plans to open-source the training code for the audio tokenizer and transformer? I'd love...
1. 播报的剧本 ```json [ { "role": "0", "text": "欢迎收听数学奇妙夜，今天咱们聊点课本里没有的数学宝藏。最近有位教育专家整理了十五个初中教材外的神奇公式，说能帮孩子解几何题像开挂一样。" }, { "role": "1", "text": "对对对，就像工具箱里的隐藏款螺丝刀，平时看不到，关键时刻能解决大麻烦。比如说梅涅劳斯定理，听着像外国名字，用起来可接地气了。" }, { "role": "0", "text": "梅涅劳斯？这名字让我想起篮球明星，不过定理说的是啥呢？" }, { "role": "1", "text": "想象你画个三角形，突然有条直线像把刀一样把三边都切了，这条线就叫截线。定理说的就是切出来的六段小线段有个神秘的比例关系。" }, { "role":...
