rawsh
Zyphra ZAYA1-base: First large-scale model trained on AMD
Debugging divergence between engine and transformers logprobs for RL
Batched reward model inference and Best-of-N sampling
Re:Grandmaster-Level Chess Without Search
on Thursday October 17, 2024 @09:42PM (#41875628)
You can actually get solid performance with pretrained chat models: https://raw.sh/posts/chess_puzzles
On lichess puzzles gpt4o with the compiled prompt is around 70%, I think the 270M transformer is around 95%
Teaching LLMs to solve chess puzzles with DSPy and Finetuning
Re:Show HN: Relari – Auto Prompt Optimizer as Lightweight Alternative ...
on Wednesday August 28, 2024 @02:50PM (#41382865)
Bit confused what the value add is over a framework like DSPy. This still requires you to create an eval dataset with ground truth, basically the only hard part of using DSPy. Easily getting the optimized prompt and having some metrics out of the box is not worth nearly $1k/mo IMO
Side note: I’ve had a lot of luck combining automatic prompt optimization with finetuning. There is definitely some synergy https://raw.sh/posts/chess_puzzles
Teaching chat models to solve chess puzzles
Re:Systematically Improving Your RAG
on Wednesday May 22, 2024 @06:58PM (#40447827)
I built a web version with WASM at https://pdfgrep.com a few years ago in case it’s helpful to anyone
Re:Systematically Improving Your RAG
on Wednesday May 22, 2024 @06:56PM (#40447807)
https://github.com/VikParuchuri/marker is solid, but slow and needs gpu(s) to be practical
Re:Pg_bm25: Elastic-Quality Full Text Search Inside Postgres
on Sunday October 08, 2023 @02:56PM (#37813354)
Is it possible to use this for hybrid search in combination with pg_embedding? My understanding is that hybrid search currently requires syncing with Postgres
Ask HN: Why aren't there Open source embedding models with context length > 512?
Re:Show HN: DankGPT – Chat with Your Documents
on Thursday July 27, 2023 @12:07PM (#36895706)
Documents actually never get uploaded! PDF text extraction happens on the client using a web worker and MuPDF compiled to WASM.
1. PDF parsed and chunked on the client
2. Sparse vectors are regenerated for the entire document corpus and the existing vectors are updated
3. Dense vectors are generated for the new text and upserted along with the new sparse values
The original documents stay on your device.
Re:Show HN: DankGPT – Chat with Your Documents
on Wednesday July 26, 2023 @02:01PM (#36882094)
Nope, it’s a serious project; I mostly made it for personal use during my last semester of college. I rewrote it a few times and packaged it up because I think it’s genuinely useful. Langchain gets you 80% of the way there but you run into issues with it very quickly.
Show HN: DankGPT – Chat with Your Documents
Uses hybrid semantic search (combination of dense embeddings and sparse vectors) to retrieve high quality answers across your documents.Features- Significantly faster than competition (Process a 200 page PDF in <5s)- Much better answer quality- Fast summarization tool- Beta API for end to end extractive document QA ([email protected])Try it out (no login)- Llama 2 paper...
Show HN: Search PDFs in the browser using PDFgrep compiled to WebAssembly
pdfgrep.org is a CLI tool I use a lot to reference documents from classes. I’ve been playing around with Emscripten and I wanted to make a UI version of the tool. All documents stay local so there is no upload time, and search time is similar to the native binary. Click on a search result to open a preview of the PDF. Any feedback would be much appreciated. Dockerfile source code:...
Re:Ask HN: What's your favorite GPT powered tool?
on Monday May 15, 2023 @04:35PM (#35954129)
DankGPT is able to draw context from a library of documents (textbook, papers, class slides) to explain any topic and answer complicated reasoning problems.
It’s very similar to ChatPDF, but you can include multiple documents and it has much better context selection. This leads to better answers in practice (less “the source does not contain information on…” and hallucinations)
Show HN: Search PDFs using WASM in the browser
pdfgrep.org is a CLI tool I use a lot to reference documents from classes. I’ve been playing around with Emscripten and I wanted to make a UI version of the tool. All documents stay local so there is no upload time, and search time is similar to the native binary. Click on a search result to open a preview of the PDF. Any feedback would be much appreciated.Dockerfile source code (warning: very...