autolabel
autolabel copied to clipboard
refuel-ai
Label, clean and enrich text datasets with LLMs.
**Describe the bug** Currently, we see an error during `agent.run` if a user has specified `compute_confidence = True`. While this error is useful, we can throw this error at `agent.plan`...
该工具非常好, 虽然是初期, 但我一直会关注. 现在这样的工具网上很多, 且基本类似, 但实际需求中遇到的问题比这个复杂, 比如: 对话数据中对话错位的问题(用户连续发问, 客服依次回答问题),这样对话一对一就被破坏了, 需要进行数据标注进行一对一的问答,这个需要处理逻辑上下不一致的问题
**Describe the bug** Today, running `get_data` prints out the number of bytes being downloaded, but we should explicitly say it's bytes being downloaded and not kilobytes or number of rows....
Currently, autolabel will retry requests to refuel-llm (generation and confidence) as long as the response isn't 200. Ideally, we should only retry for error codes where the likelihood of succeeding...
**Is your feature request related to a problem? Please describe.** Support for a new task: Translation. I'd like to be able to translate data from any language to another. Modern...
**Is your feature request related to a problem? Please describe.** Currently we pass in configs as dictionary or path to json objects. We should instead be passing in config objects...
**Is your feature request related to a problem? Please describe.** Currently, we create a task for every labeling run and store metrics like accuracy and the labels in the sqlite...
- Write the cache to a consistent location to make it easier to find - Allow clearing the cache - Make the use of cache = False consistent. @rishabh-bhargava any...