Bokai Xu comments

Results 29 comments of


                                            Bokai Xu

我用官方提供的代码部署会报错RuntimeError: probability tensor contains either `inf`, `nan` or element < 0

不要用fp16，用bf16就没有问题了 @lukeewin

may i have a look at the training loss curve

@Oxi84 Thanks! That is cool, mine is still 0.8 after 1k steps, my batch size=128, lora-r=8, base_model=llama-7b, i used int8 training. :D

add chinese character and english words generation count evaluation (17% accuracy, 40 samples)

@1c7 Exactly!

where is the train file?

also interested in training data composition, as it was not mentioned in technical report :)

想要某个主题的语料，只能下载下来后通过关键词匹配？有其他办法嘛？

Hi, you can use `http://39.101.77.220/query/search/?start=10&size=10&keyword=hello` to fetch desired data.

Inquiry About Latency of streaming generate with audio in MiniCPMo Web Demo

Hi! Thank you for using MiniCPM-o 2.6, I checked the code on huggingface, if I understand your question correctly, streaming_generate=True means to generate audio, and streaming_generate=False will not generate audio....

Inquiry About Latency of streaming generate with audio in MiniCPMo Web Demo

Hello biraj! Thank you for your feedback, which is very important. The audio generation's latency has two concept: - initial latency, the delay between end of user question and the...

💡 [REQUEST] - Simultaneous multimodal inputs

Yes absolutely you can, but you should consider using `omni` mode. You can refer to this code: https://github.com/OpenBMB/MiniCPM-o?tab=readme-ov-file#multimodal-live-streaming ```python import math import numpy as np from PIL import Image from...

💡 [REQUEST] - Simultaneous multimodal inputs

> hey [@bokesyo](https://github.com/bokesyo), i've a question wrt to your [previous comment](https://github.com/OpenBMB/MiniCPM-o/issues/848#issuecomment-2671552404). > > _note: we're not using vLLM. we're using transformers._ > > 1. do we have to use ``...

关于语音克隆的评估

您好，请问这里的 seedtts test-zh SIM-o 42的推理代码是如何的？感谢！