hanxiao Follow

Han Xiao hanxiao

Founder @jina-ai | Creator of Fashion-MNIST & bert-as-service

2.7k followers · 23 following · 270

Highlights

Arctic Code Vault Contributor

Organizations

Pinned

jina-ai/jina

An easier way to build neural search in the cloud

Python 1.4k 249
bert-as-service

Mapping a variable-length sentence to a fixed-length vector using BERT model

Python 8.4k 1.7k
zalandoresearch/fashion-mnist

A MNIST-like fashion product database. Benchmark 👉

Python 8.4k 2k
hanxiao.github.io

Blog

HTML 21 4

5,158 contributions in the last year

Activity overview

Contributed to jina-ai/jina, jina-ai/jina-hub, jina-ai/examples and 5 other repositories

Contribution activity

October 2020

hanxiao/mindspore C++ Oct 28
hanxiao/utterances TypeScript Oct 27
hanxiao/copyright-action Python Oct 5

Created a pull request in jina-ai/jina that received 5 comments

Oct 16

feat(flow): inspect pods are collected before gateway

New Feature .gather_inspect() can be used to manually collect inspect pods --inspect is added to Flow CLI/API, allowing HANG, REMOVE, COLLECT thre…

+106 −38 • 5 comments

refactor(encoder): change get_model to get_cell mindspore idiom Oct 28
fix(hub): hub repo name should given by user Oct 27
refactor(hubio): remove duplicate logic on hub push Oct 27
refactor(driver): eval drivers focus on data type of field Oct 26
refactor(cli): remove the use of packaging Oct 26
test: add docker hello-world with no devel install Oct 22
fix: raise better exception for invalid doc id Oct 22
feat(cli): show local version side by side Oct 22
feat(hub): show local manifest Oct 21
refactor(hubio): remove the dependency of requests from list Oct 21
feat(cli): check new jina version on start Oct 21
feat(flow): add plot for uses before after Oct 20
refactor: flatten evaluator inheritance Oct 20
feat(pod): allow uses-before/after on single pea Oct 20
style: fix line space Oct 16
ci: just for testing a PR Oct 16
ci: just for testing a PR Oct 16
ci: replace match-py-ver to devel Oct 16
feat(driver): add collect evaluation driver Oct 16
feat(flow): inspect without introducing side-effect on socket Oct 15
refactor(eval): remove evaluate flow Oct 15
feat(flow): add eval() interface generates hanging pod Oct 14
refactor(evaluator): rename, style and format Oct 14
Revert "fix: add devel tag to Pillow dependency" Oct 13
refactor(indexer): refactor cache index add shortcuts Oct 13
Some pull requests not shown.

fix: build history no need to be list Oct 27
fix: limit in find_many Oct 22
style: reformat the code Oct 22

fix(docker): fix env path in dockerfile Oct 28

feat: allow zipped-like batching Oct 28
refactor: ranking evaluator inherit from field evaluator Oct 28
fix: disabled pip cache in docker Oct 28
fix: failure not combinable Oct 27
feat: add multimodal driver Oct 27
style: added failure Oct 26
chore: add packaging in extra requirements Oct 26
style: black on merge Oct 26
feat: design pattern doc for incremental index Oct 26
ci: new workflow for testing examples Oct 23
docs: add windows support communication to readme Oct 23
fix: removed top_k default Oct 22
feat: add file based evaluator Oct 22
test: take out jinatestcase from protobuf test Oct 22
feat: add evaluation encoder and crafter executors and drivers Oct 19
feat: potentially add argument of ordered indexes from batch Oct 15
Fix some machine translated bugs in the Russian read me. Oct 14
ci: add test job for examples in cd Oct 14
fix: fix the driver queryset Oct 13
fix: check existence of model path Oct 13
fix: add devel tag to Pillow dependency Oct 13
feat: add pretrained model exception and accept it while test_uses Oct 13
test: refactor load yaml test Oct 12
fix: avoid the bounding RouteDriver to the ControlRequest Oct 12
feat: check whether a document has been already indexed Oct 11
Some pull request reviews not shown.

fix: find_many invocation Oct 22

Created an issue in jina-ai/jina that received 4 comments

Oct 2

Chunk2DocRankDriver does not preserve parent_id when building the new matches

Describe your problem In the Chunk2DocRankDriver, non parent_id info is filled in. This makes bilevel traversal_paths=['cc', 'c'] impossible. if m…

4 comments

Why does the `AnnoyIndexer` appears several times in the list? Can we order the list alphabetically? Especially, if double entries are OK, it would make it much better readable. Oct 22

77 contributions in private repositories Oct 1 – Oct 27