RustingSword

> 建立一個負向詞語清單應該可以解決這個問題? 更好一点的办法可能是用 nlp 技术，计算一下语意褒贬。

This is strange. According to Tensorboard code [[1]](https://github.com/tensorflow/tensorboard/blob/348af4cf9564b8217750e8d84e0b2b89555be62c/tensorboard/data/server/logdir.rs#L60)[[2]](https://github.com/tensorflow/tensorboard/blob/29cfcec9d15c47d1bf54b15df24e65bc38c188e6/tensorboard/data/server/gcs/logdir.rs#L119), any file containing `tfevents` in the file name will be treated as event file. `events.out.tfevents` is a hardcoded prefix in the [EventFileWriter](https://github.com/tensorflow/tensorboard/blob/master/tensorboard/summary/writer/event_file_writer.py#L76),...

Ah, I see the problem. No it does not, the file name is passed as a parameter to the constructor for flexibility. Maybe it is better to use an auto...

> Do you knwo how to change this value I mean, Scale it down by 1000 like 0.003 to 0..007 something like that I have no idea, but I think...

I think bucket generation still have some issues, histogram looks kind of strange now.  I also tested pr curve using data generated by [pr curve demo code](https://github.com/tensorflow/tensorboard/blob/master/tensorboard/plugins/pr_curve/pr_curve_demo.py#L218) in TensorBoard...

Sorry for the delay. The pr curve looks correct now, however the histogram still has some issue. I will try to find a fix.

精确的 token 数量可以用 [tiktoken](https://github.com/openai/tiktoken) 来获取，`gpt-3.5-turbo`用的 encoder 是`tiktoken.encoding_for_model("gpt-3.5-turbo")` (`cl100k_base`) 这个[官方 notebook](https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb) 最后有现成的函数可以用。

No specific version of protobuf is required, and I have used many versions since the beginning of this project without such issue. I have no Windows development environment currently, and...