div papers
Searches arXiv and prints one paper a line, or reads papers whole by id. Reading before running, with no key.
div papers search mixture of experts inference
div papers get 2410.17954Before a run spends anything, it is worth knowing what has been tried and by whom. div papers asks arXiv directly, not the control plane, so it needs no key and costs nothing. It is the one command that talks to something other than Divergent.
Usage
div papers search <query> [--author <name>] [--category <cat>] [--sort <by>] [--limit <n>]
div papers get <id...>| Flag | What it does |
|---|---|
--author <name> |
only papers with this author |
--category <cat> |
only papers in this category, like cs.LG or stat.ML |
--sort <by> |
relevance (the default), or newest first by submitted or updated |
--limit <n> |
how many, from 1 to 100. Default 10 |
--json |
print what arXiv said, as JSON |
Searching
$ div papers search mixture of experts inference --limit 3
2410.17954v2 2024-10-23 cs.AI ExpertFlow: Efficient Mixture-of-Experts Inference via Predictive Expert Caching and Token Scheduling
2402.14800v2 2024-02-22 cs.CL Not All Experts are Equal: Efficient Expert Pruning and Skipping for Mixture-of-Experts Large Language Models
2601.15021v1 2026-01-21 cs.LG Mixture-of-Experts Models in Vision: Routing, Optimization, and Generalization
3 of 1137 papers that match. div papers get <id> prints one wholeOne paper a line on stdout: its id, the day its first version went up, its primary category, and its title last. The count goes to stderr, so | head -1 is the top paper and | cut -d' ' -f1 is the ids.
Every word has to appear somewhere in the paper. arXiv’s index leaves out stopwords like “of” and “the”, and asking for one would match nothing, so div does not send them. A phrase in double quotes has to appear as written:
div papers search '"speculative decoding"' --category cs.CL --sort submittedA query in arXiv’s own syntax goes to arXiv as it is, with --author and --category added to it. Its syntax is field prefixes (ti:, au:, abs:, cat:) or AND, OR and ANDNOT:
div papers search 'ti:lora ANDNOT abs:diffusion'When arXiv cannot read the query, div prints arXiv’s reason and exits 1:
error arXiv refused the query: Invalid query string: 'ti:lora AND'Reading one
$ div papers get 2410.17954
2410.17954v2 ExpertFlow: Efficient Mixture-of-Experts Inference via Predictive Expert Caching and Token Scheduling
Xin He, Shunkang Zhang, Kaijie Tang, Shaohuai Shi, Yuxin Wang, Zihao Zeng, Zhenheng Tang, Xiaowen Chu, Haiyan Yin, Ivor W. Tsang, Yew Soon Ong
cs.AI cs.CL · published 2024-10-23 · updated 2026-04-02
https://arxiv.org/abs/2410.17954v2 https://arxiv.org/pdf/2410.17954v2 https://doi.org/10.1145/3770743.3804292
Accepted in DAC'26, Mixture-of-Experts, Inference, Offloading
Sparse Mixture-of-Experts (MoE) models can outperform dense large language models at similar computation…The authors, the categories, the dates, the links, what the authors wrote in arXiv’s comment field, and then the abstract with its line wrapping undone. get takes several ids at once, in any form they were pasted in: 2410.17954, arXiv:2410.17954, hep-th/9901001, or the abs or pdf link. An id with no version gets the latest version, and 2410.17954v1 gets that version only.
An id that is not an arXiv id is a usage error and exits 2. An id that is well formed but that arXiv does not have is named on stderr, and div exits 1 after printing the papers it did find:
arXiv has no paper 2401.99999Asking too often
arXiv asks its callers to send one request every three seconds, and it answers a burst with a 429 or a 503 rather than queueing it. div says so and exits 1, and trying again a few seconds later works. A script that loops over searches should sleep between them.
An agent with an MCP client has the same two operations as search_papers and get_papers; see the MCP server.