breadislove

5 ms·

61.
by breadislove 1y ago
I uploaded a file and its been processing for over an hour now. No failure or anything. Maybe you should look into that.
62.
by breadislove 2y ago
we are using claudie(.io) together with ray. but we put a lot time into the infrastructure but we managed to cut our cost for a factor of 20x by doing this.
63.
Baked-in Brilliance: Reranking Meets RL with mxbai-rerank-v2(mixedbread.com)
2 points by breadislove 2y ago|0 comments
64.
Perplexity launches $50M seed and pre-seed VC fund(techcrunch.com)
1 points by breadislove 2y ago|1 comments
65.
Unlock DeepSeek-R1 Inference Performance on AMD Instinct MI300X GPU(rocm.blogs.amd.com)
2 points by breadislove 2y ago|0 comments
66.
ML Valentines(ml-valentines.github.io)
1 points by breadislove 2y ago|0 comments
67.
by breadislove 2y ago
The systems they tested against the LLMs are mostly used as a part of a larger system. A more fair comparison would be to use something like MinerU [1] and proper benchmark like the OHR Bench [2] and Reductos table bench [3]. This paper is
68.
V0.2: Networking in Python(github.com)
2 points by breadislove 2y ago|0 comments
69.
by breadislove 2y ago
no this wont work. embedding models at the moment are pretty bad with negations
70.
An Analysis of Chinese LLM Censorship and Bias with Qwen 2 Instruct(huggingface.co)
3 points by breadislove 2y ago|0 comments
71.
by breadislove 2y ago
the way forward should be making llms understand tables better. those documentations should be written for humans, as an alternative you could have an llm friendly version
72.
by breadislove 2y ago
There is this really interesting blog post about making rope (by the main author of the paper) multimodal as used by qwen2 vl. it's in chinese but google translate does a pretty good job: https://spaces.ac.cn/archives&#
73.
YOLO Unraveled: A Clear Guide(opencv.ai)
2 points by breadislove 2y ago|0 comments
74.
BMX: A Freshly Baked Take on BM25(mixedbread.ai)
91 points by breadislove 2y ago|11 comments
75.
Most AI startups are service businesses, pretending to be product companies(nextword.substack.com)
7 points by breadislove 2y ago|0 comments
76.
64 Bytes per Embedding, Yee-Haw(mixedbread.ai)
1 points by breadislove 2y ago|0 comments
77.
SWE-Agent(swe-agent.com)
6 points by breadislove 3y ago|0 comments