vvipgupta

5 ms·

1. 1.▲
   Show HN: UpTrain – A Practical Approach to Finetuning LLMs for Custom Use-Cases(github.com)

   10 points by vvipgupta 4y ago|1 comments
2. 2.▲
   by vvipgupta 4y ago

   Thanks for letting us know. We did some docs restructuring before the launch, and missed fixing this link. It is now available here: https://docs.uptrain.ai/docs/uptrain-examples/quickstart-tut...
3. 3.▲
   by vvipgupta 4y ago

   Generally, MLOps helps in reducing engineering headaches. During our user interviews and customer calls, we realized very early that customization is key for ML model monitoring since all models are different. Thus, we have built the framew
4. 4.▲
   by vvipgupta 4y ago

   Thanks! Also, wondering how did you hear about Arize? Have you dealt with the pain of ML model monitoring in the past?
5. 5.▲
   by vvipgupta 4y ago

   Just curious, what kind of use cases do you have in mind?
6. 6.▲
   by vvipgupta 4y ago

   Thanks for the very relevant comment :) We provide users the option to attach their training data from csv/json (working to support loading from cloud storage provider or data lakes). We have illustrated this in some of our examples, s
7. 7.▲
   by vvipgupta 4y ago

   Additionally, refinement is a key focus of ours. Figuring out the best data points to retrain the model upon has twin benefits: 1) It provides automated issue resolution and saves data scientists' effort to debug and fix their models.
8. 8.▲
   by vvipgupta 4y ago

   Thanks for the suggestion and links. Completely agree, ML production data management can be painful and to support model refinement for users that operate at scale, an abstraction at the data layer would be a useful feature.
9. 9.▲
   Launch HN: UpTrain (YC W23) – Open-source performance monitoring for ML models

   138 points by vvipgupta 4y ago|31 comments
10. 10.▲
    by vvipgupta 4y ago

    The same is true for data (aka gradients) consistency while training large ML models. Asynchronous SGD is as good (and maybe even faster) than synchronous SGD: https://papers.nips.cc/paper/2011/file/218a0aefd1
11. 11.▲
    by vvipgupta 4y ago

    You might also want to check out https://github.com/lucidrains/PaLM-rlhf-pytorch
12. 12.▲
    by vvipgupta 4y ago

    When training over multiple GPUs, it's hard not to think about Ray ( https://docs.ray.io/en/latest/train/train.html ). Ray, as an open-source project, has exploded over the last few years and helps with th
13. 13.▲
    by vvipgupta 4y ago

    This would be a good application of chatGPT, testing a test for whether it tests for rote learning or fundamentals.