What Is This?

This is an HTML+RDFa representation of metadata describing this Web-addressable resource.

Why Is This Important?

The property (attributes and values pairs) links on this page unveil a different kind of link, one which enables the following on HTTP networks such as the Web:

Unambiguous identity for entities (aka. strong identifiers)
Implicit binding of an entity and its metadata via strong identifiers
Multiple metadata representations that enable a variety of presentations
High precision Search and Find queries that simply use the metadata documents (by referencing entity URIs) as the query's Data Source Name

How Do I Discover Alternative Metadata Representations?

This document exposes metadata in the following formats: (X)HTML+RDFa, Turtle, N3, RDF/JSON, or RDF/XML. In the most basic form, you can simply view the (X)HTML source markup of this page, and go directly to the <head/> section which contains a <link/> tag with relationship and type properties for each format.

In addition, you can also explicitly request a desired metadata representation for a given resource via HTTP GET requests that use the entity's strong identifier as the call target.

How Can I Expose My Web Resources In This Manner?

Simply include the following in the <head/> section of your (static or dynamically generated) (X)HTML page:

<link rel="alternate" title="My Data in RDF Linked Data form"
type="application/rdf+xml"
href="http://linkeddata.uriburner.com/about/id/<this-page-URL>/>"

How Is This Related To The Linked Data Meme?

As stated above, the links in this page expose strong identifiers for its primary topic, secondary topics, attributes, and some values. These links, via implicit association, act as conduits to their metadata-bearing documents, in a variety formats.

content	<p>Your browser does not support iframes.</p>
Description	What this PR does / why we need it? This PR fixed 2 bugs that disable Data Parallel on V1. [BUG 1] All-to-all communication is only used in MC2 mode, which is still not fully supported. Before the backend supports the communication, we closed all-to-all function while not using MC2. [BUG 2] When DP ranks have different max_tokens params, the rank finishes forwarding first must wait for others before termination due to all-reduce. Otherwise there will show communication error. This should be supported by execute_dummy_batch() to run a dummy forward. However, the dummy MoE will return immediately before all-reduce. We need to let dummy run the all_reduce func. Does this PR introduce any user-facing change? Yes, as a bug-fix PR, it introduces a new method and several branches. How was this patch tested? This patch can be tested with: # torch version: 2.5.1+cpu # torch_npu version: 2.5.1 export VLLM_ENABLE_MC2=0 export VLLM_USE_V1=1 export TASK_QUEUE_ENABLE=2 export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True export VLLM_ENABLE_GRAPH_MODE=0 export USING_LCCL_COM=0 export VLLM_WORKER_MULTIPROC_METHOD=spawn # recommended test model: DeepSeek-V2-Lite python examples/offline_inference/data_parallel.py --model= --dp-size=2 --tp-size=8 Co-authored-by: HanlinDu ftdl21@pku.edu.cn Co-authored-by: ZhengWG zwg0606@gmail.com Co-authored-by: realliujiaxu realliujiaxu@163.com
Title	V1 parallel fix: bug fix to enable DP in V1 by HanlinDu · Pull Request #710 · vllm-project/vllm-ascend · GitHub
container of	http://dev.restore.ovi.cnr.it:8890/abo...scend/pull/710/files#id0x7fd569ab3758 http://dev.restore.ovi.cnr.it:8890/abo...scend/pull/710/files#id0x7fd568785018 http://dev.restore.ovi.cnr.it:8890/abo...scend/pull/710/files#id0x7f05b43408d8 http://dev.restore.ovi.cnr.it:8890/abo...scend/pull/710/files#id0x7f05b48c1528
links to	https://github.com/resources/whitepapers https://github.com/security/advanced-security https://desktop.github.com https://github.com/login?return_to=%2Fvllm-project%2Fvllm-ascend https://github.com/vllm-project/vllm-ascend/actions https://github.com/vllm-project/vllm-ascend/projects https://github.com/vllm-project/vllm-ascend/security https://github.com/vllm-project/vllm-ascend/pulse https://github.com/ https://github.com/features/copilot https://github.com/features/actions https://github.com/features/codespaces https://github.com/features/issues https://github.com/features/code-review https://github.com/features/discussions https://github.com/features/code-search https://github.com/features https://docs.github.com https://skills.github.com https://github.blog https://github.com/enterprise https://github.com/team https://github.com/enterprise/startups https://github.com/solutions/industry/nonprofits https://github.com/solutions/use-case/devsecops https://github.com/solutions/use-case/devops https://github.com/solutions/use-case/ci-cd https://github.com/solutions/use-case https://github.com/solutions/industry/healthcare https://github.com/solutions/industry/financial-services https://github.com/solutions/industry/manufacturing https://github.com/solutions/industry/government https://github.com/solutions/industry https://github.com/solutions https://github.com/resources/articles/ai https://github.com/resources/articles/devops https://github.com/resources/articles/security https://github.com/resources/articles/software-development https://github.com/resources/articles https://resources.github.com/learn/pathways https://resources.github.com https://github.com/customer-stories https://partner.github.com https://github.com/solutions/executive-insights https://github.com/sponsors https://github.com/readme https://github.com/topics https://github.com/trending https://github.com/collections https://github.com/premium-support https://github.com/pricing https://docs.github.com/search-github/...derstanding-github-code-search-syntax https://github.com https://docs.github.com/site-policy/github-terms/github-terms-of-service https://docs.github.com/site-policy/privacy-policies/github-privacy-statement https://github.com/security https://www.githubstatus.com/ https://docs.github.com/ https://support.github.com?tags=dotcom-footer https://github.com/why-github https://docs.github.com/terms https://docs.github.com/privacy https://github.co/hiddenchars https://github.com/features/copilot/copilot-business https://github.com/features/models https://github.com/vllm-project/vllm-ascend/pull/710 https://github.com/HanlinDu https://github.com/HanlinDu/vllm-ascend/tree/v1-parallel-fix https://github.com/vllm-project/vllm-ascend/pull/710/commits https://github.com/vllm-project/vllm-ascend/pull/710/checks https://github.com/vllm-project/vllm-ascend/pull/710/files https://github.com/vllm-project/vllm-a...174e86b202425f9c304ea53d87a0f98f3ef8a https://github.com/vllm-project/vllm-a...ac1123a72ca75986d51e95ce0d9f88906a41c https://github.com/vllm-project/vllm-a...18e58443595981c44d3debf067e3f946cbbc4 https://github.com/vllm-project/vllm-a...f18d49b6f22078fc6344355ca73ea0c1e7c76 https://github.com/vllm-project/vllm-a...4ae7c8fb74b26bac37c15f3aa0468a4385034 https://github.com/vllm-project/vllm-a...e81040397ca86c1c49dd5933aada0a44f9cf5 https://github.com/vllm-project/vllm-ascend https://github.com/vllm-project/vllm-ascend/issues https://github.com/vllm-project/vllm-ascend/pulls https://github.com/vllm-project https://github.com/signup?return_to=%2...2Fvllm-ascend%2Fissues%2Fnew%2Fchoose https://github.com/login?return_to=%2F...2Fvllm-ascend%2Fissues%2Fnew%2Fchoose https://github.com/vllm-project/vllm-ascend/tree/main https://github.com/vllm-project/vllm-ascend/pull/710/files#start-of-content https://github.com/login?return_to=htt...ct%2Fvllm-ascend%2Fpull%2F710%2Ffiles https://github.com/vllm-project/vllm-ascend/pull/710/files#top https://github.com/vllm-project/vllm-a...f969fea8cb1587557fa1950959103d55de056 https://github.com/vllm-project/vllm-a...10fe6f255330d2aa9359f485a638cedd8cbb7 https://github.com/HanlinDu/vllm-ascen...5933aada0a44f9cf5/vllm_ascend/envs.py https://github.com/vllm-project/vllm-ascend/pull/710/{{ revealButtonHref }} https://github.com/HanlinDu/vllm-ascen...cf5/vllm_ascend/models/deepseek_v2.py https://github.com/vllm-project/vllm-a...ns/14974291737/job/42062543703?pr=710 https://github.com/vllm-project/vllm-a...ns/14974291780/job/42062544036?pr=710 https://github.com/vllm-project/vllm-a...ns/14974291780/job/42062544052?pr=710 https://github.com/HanlinDu/vllm-ascen...a44f9cf5/vllm_ascend/ops/fused_moe.py https://github.com/HanlinDu/vllm-ascen...f9cf5/vllm_ascend/worker/worker_v1.py https://github.com/signup?ref_cta=Sign...ource_repo=vllm-project%2Fvllm-ascend
opl:providedBy	http://www.stackoverflow.com/#this
type	Document
xhv:alternate	https://github.com/vllm-project/vllm-ascend/pull/710.diff https://github.com/vllm-project/vllm-ascend/pull/710.patch
described by	https://github.com/vllm-project/vllm-ascend/pull/710/files

Alternative Linked Data Views: Facets | iSPARQL | ODE Raw Linked Data formats: CXML | CSV | RDF ( N-Triples N3/Turtle JSON XML ) | OData ( Atom JSON ) | Microdata ( JSON HTML) | JSON-LD

This material is Open Knowledge

W3C Semantic Web Technology

This material is Open Knowledge

Creative Commons License

Valid XHTML + RDFa

This work is licensed under a Creative Commons Attribution-Share Alike 3.0 Unported License.
OpenLink Virtuoso version 07.20.3231, on Linux (x86_64-generic_glibc25-linux-gnu), Single Edition