shivnathtathe

3 ms·

1.
UnlimitedNIM – Proxy that makes Nvidia Nim's 40 RPM feel unlimited(github.com)
1 points by shivnathtathe 2mo ago|0 comments
2.
by shivnathtathe 2mo ago
yes! opensmith export dumps to JSON out of the box. Parquet isn't there yet but it's a good shout — would just need a pandas dependency. if you open an issue I'll prioritize it.
3.
Show HN: OpenSmith – Trace LLM pipelines locally with trace, no cloud(github.com)
1 points by shivnathtathe 2mo ago|2 comments
4.
by shivnathtathe 5mo ago
Hey HN. I built this because LangSmith needs a cloud account to see my own traces. The full stack is: → @trace decorator on any Python function → autopatch() for OpenAI, Anthropic, LiteLLM, Qdrant, ChromaDB, Pinecone → SQLite at ~/.o
5.
Show HN: I built an open-source, local-first alternative to LangSmith(pypi.org)
6 points by shivnathtathe 5mo ago|1 comments
6.
by shivnathtathe 5mo ago
Interesting approach! I've been working on something complementary — local-first LLM observability called opensmith. Different angle but same philosophy of keeping everything local.
7.
by shivnathtathe 5mo ago
If you have any org email, you can get free chatgpt + subscription.
8.
by shivnathtathe 5mo ago
Observability is the missing piece here — built opensmith for exactly this reason, tracing agent control flow locally
9.
by shivnathtathe 5mo ago
Been working on local-first LLM observability for exactly this use case — tracing local model pipelines without sending data to cloud. Happy to share if anyone's interested.
10.
True 4-Bit Quantized CNN Training on CPU – 92.34% on Cifar-10(arxiv.org)
3 points by shivnathtathe 7mo ago|3 comments