fangzhangmnm

Results 14 comments of fangzhangmnm

After the ludum dare, I realized that it is very useful to add a online highscore service for tic-80.

> I meant doing the optimization under the consumption that the constant tensors can be contracted beforehand offline. The above functionality only cache the intermediates, but the path planning might...

一般的处理方法就是把聊天记录存下来用text2vec编一下索引做一个小号google。然后聊天时用faiss查最近的vec对应的聊天记录一起喂到模型里。可以参考glm langchain这个repo。 不过这只是记笔记,模型没有真正学到东西哦

did you unzipped the .mcworld file?

Can I post my old Unity code here? I used a sphere collider for the head. ```csharp Vector3 ResolveCollision(bool resolveHead = true) { Vector3 totalMoved = Vector3.zero; //Head Collision Resolving(?)...

你试过训练吗。我自己写了一个mla结果loss收敛的很差 ![image](https://github.com/user-attachments/assets/9bf9cd88-ec3f-4135-b9d3-9dbe63a51923) Disclaimer: 这个是我用自己写的代码在自己的数据集上训练的. 不是这个repo里的代码和数据 Edit: 后来发现是一个低级错误,导致这个attention层被短路了. 修复之后loss曲线和没有MLA时差不多,虽然巨慢(2x time) (因为没有MLA版的fast attention)

> > 你试过训练吗。我自己写了一个mla结果loss收敛的很差 ![image](https://private-user-images.githubusercontent.com/6160173/420710981-9bf9cd88-ec3f-4135-b9d3-9dbe63a51923.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.aoqmSJ4tsp1oOxPWJCNtCq25o4xNV1bu7N18vIKecJk) > > 请问这个是在仓库里公开的那个pretrain_hq.jsonl 上训练的loss吗? 这个是我用自己写的代码在自己的数据集上训练的. 不是这个repo里的代码和数据

A simpler fix is to disable the collision of the table with the player lol

顶,需要ir模式。wysiwyg模式用的没安全感

我用165W的4090m一晚上就能说人话了。