--- title: From Upstream Changes to Downstream Confidence: Inside... description: PyTorch's Cross-Repository CI Relay (CRCR) lets out-of-tree accelerator backends plug into upstream CI, but leaves each backend to decide which of PyTorch's... canonical: https://daily.dev/posts/from-upstream-changes-to-downstream-confidence-inside-torch-spyre-s-integration-with-pytorch-crcr--1inp5j2xe twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: From Upstream Changes to Downstream Confidence: Inside Torch Spyre’s Integration with PyTorch CRCR – PyTorch | daily.dev og:description: PyTorch's Cross-Repository CI Relay (CRCR) lets out-of-tree accelerator backends plug into upstream CI, but leaves each backend to decide which of PyTorch's... og:url: https://daily.dev/posts/from-upstream-changes-to-downstream-confidence-inside-torch-spyre-s-integration-with-pytorch-crcr--1inp5j2xe og:image: https://api.daily.dev/og/posts/1INp5j2xe.png og:image:alt: From Upstream Changes to Downstream Confidence: Inside Torch Spyre’s Integration with PyTorch CRCR – PyTorch og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # From Upstream Changes to Downstream Confidence: Inside Torch Spyre’s Integration with PyTorch CRCR – PyTorch **[PyTorch](https://daily.dev/sources/pytorch)** · 16 min read · 0 upvotes · 0 comments ## Summary PyTorch's Cross-Repository CI Relay (CRCR) lets out-of-tree accelerator backends plug into upstream CI, but leaves each backend to decide which of PyTorch's tens of thousands of tests matter for its hardware. Torch Spyre, the IBM Spyre Accelerator's PyTorch backend, built an agentic four-stage pipeline (high-level selection, repository memory indexing, low-level selection, execution-based refinement) to automatically select and bucket relevant tests, plus a declarative YAML framework that adapts upstream tests (dtype exclusions, xfail/skip buckets, capability-driven inclusion) without patching PyTorch source. The post also covers workflow patterns for consuming CRCR dispatches, nightly and release testing, test splitting/parallelization, build-once-test-many artifact reuse, retry and failure-classification patterns for reliability, handling matrix job callbacks, and capturing metadata for reproducibility. These mechanisms took Torch Spyre to CRCR's L2 integration, and the team plans to upstream reusable parts alongside PyTorch OpenReg. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does Torch Spyre decide which PyTorch tests to run for the IBM Spyre accelerator backend? An agentic four-stage pipeline handles it: a first agent narrows the test tree to relevant folders based on which PyTorch extension points the backend hooks into, a repository memory generator builds a queryable index of symbols and embeddings, a second agent buckets tests into mandatory_success or skip using backend code and metadata, and a final pass refines buckets using real hardware execution logs to catch runtime failures static analysis misses. _daily.dev surfaces engineering deep dives like this for teams building custom hardware CI pipelines._ ### How can I adapt PyTorch's CUDA-oriented test suite for a custom accelerator backend without patching upstream test files? Torch Spyre uses a declarative YAML framework that patches upstream's @ops, @modules, and @dtypes decorators at collection time rather than editing test source. It provides parameter-level knobs to exclude specific dtypes or shapes, four outcome buckets (mandatory_success, xfail, xfail_strict, skip), and capability declarations so enabling a new op automatically admits every test waiting on it, keeping the upstream test tree pristine. _developers weighing test-reuse strategies for hardware backends can track patterns like this on daily.dev._ ### How much test-selection rework is needed when upgrading to a new PyTorch version under an agentic test-selection pipeline? Version upgrades become config diffs rather than full reprocessing: the low-level selection agent runs only on the delta of newly added and modified tests. For example, upgrading from PyTorch 2.13 to 2.14 required evaluating roughly 4000 changed tests instead of re-analyzing tens of thousands of tests in the full suite. _daily.dev helps teams stay ahead of PyTorch version upgrades and their downstream CI impact._ ## Similar posts on daily.dev - [Introducing Cross-Repository CI Relay: Scalable CI for PyTorch’s Out-of-Tree Backends – PyTorch](https://daily.dev/posts/introducing-cross-repository-ci-relay-scalable-ci-for-pytorch-s-out-of-tree-backends-pytorch-ql8or2hmm) · PyTorch · 0 upvotes · 0 comments - [PyTorch Hardware Enablement: Updates from the Acceleration Integration Working Group – PyTorch](https://daily.dev/posts/pytorch-hardware-enablement-updates-from-the-acceleration-integration-working-group-pytorch-dlbmxosy4) · PyTorch · 0 upvotes · 0 comments - [Agents have made CI the bottleneck. Faster pipelines are the wrong fix.](https://daily.dev/posts/agents-have-made-ci-the-bottleneck-faster-pipelines-are-the-wrong-fix--fott4jdmy) · The New Stack · 2 upvotes · 0 comments - [How Red Hat has redefined continuous performance testing](https://daily.dev/posts/how-red-hat-has-redefined-continuous-performance-testing-m6oh6tqug) · Red Hat Developer · 0 upvotes · 0 comments - [Harnessing AI for Day-One Model Enablement – PyTorch](https://daily.dev/posts/harnessing-ai-for-day-one-model-enablement-pytorch-dbx5qmyih) · PyTorch · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#testing](https://daily.dev/tags/testing), [#cicd](https://daily.dev/tags/cicd), [#github-actions](https://daily.dev/tags/github-actions), [#pytorch](https://daily.dev/tags/pytorch) [View this post on daily.dev](https://daily.dev/posts/from-upstream-changes-to-downstream-confidence-inside-torch-spyre-s-integration-with-pytorch-crcr--1inp5j2xe) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"From Upstream Changes to Downstream Confidence: Inside Torch Spyre’s Integration with PyTorch CRCR – PyTorch","url":"https://daily.dev/posts/from-upstream-changes-to-downstream-confidence-inside-torch-spyre-s-integration-with-pytorch-crcr--1inp5j2xe","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/from-upstream-changes-to-downstream-confidence-inside-torch-spyre-s-integration-with-pytorch-crcr--1inp5j2xe"},"datePublished":"2026-09-30T13:45:08.646Z","dateModified":"2026-09-30T13:45:36.327Z","description":"PyTorch's Cross-Repository CI Relay (CRCR) lets out-of-tree accelerator backends plug into upstream CI, but leaves each backend to decide which of PyTorch's...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6c6e8ee318693c4de6f5f85f45160b68?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6c6e8ee318693c4de6f5f85f45160b68?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"PyTorch","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"PyTorch","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/5fdab3d75a834a97a52d433d9c7e5ff9","url":"https://daily.dev/sources/pytorch"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/from-upstream-changes-to-downstream-confidence-inside-torch-spyre-s-integration-with-pytorch-crcr--1inp5j2xe","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,testing,cicd,github-actions,pytorch","timeRequired":"PT16M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"PyTorch","item":"https://daily.dev/sources/pytorch"},{"@type":"ListItem","position":3,"name":"From Upstream Changes to Downstream Confidence: Inside Torch Spyre’s Integration with PyTorch CRCR – PyTorch"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/from-upstream-changes-to-downstream-confidence-inside-torch-spyre-s-integration-with-pytorch-crcr--1inp5j2xe#faq","mainEntity":[{"@type":"Question","name":"How does Torch Spyre decide which PyTorch tests to run for the IBM Spyre accelerator backend?","acceptedAnswer":{"@type":"Answer","text":"An agentic four-stage pipeline handles it: a first agent narrows the test tree to relevant folders based on which PyTorch extension points the backend hooks into, a repository memory generator builds a queryable index of symbols and embeddings, a second agent buckets tests into mandatory_success or skip using backend code and metadata, and a final pass refines buckets using real hardware execution logs to catch runtime failures static analysis misses. daily.dev surfaces engineering deep dives like this for teams building custom hardware CI pipelines."}},{"@type":"Question","name":"How can I adapt PyTorch's CUDA-oriented test suite for a custom accelerator backend without patching upstream test files?","acceptedAnswer":{"@type":"Answer","text":"Torch Spyre uses a declarative YAML framework that patches upstream's @ops, @modules, and @dtypes decorators at collection time rather than editing test source. It provides parameter-level knobs to exclude specific dtypes or shapes, four outcome buckets (mandatory_success, xfail, xfail_strict, skip), and capability declarations so enabling a new op automatically admits every test waiting on it, keeping the upstream test tree pristine. developers weighing test-reuse strategies for hardware backends can track patterns like this on daily.dev."}},{"@type":"Question","name":"How much test-selection rework is needed when upgrading to a new PyTorch version under an agentic test-selection pipeline?","acceptedAnswer":{"@type":"Answer","text":"Version upgrades become config diffs rather than full reprocessing: the low-level selection agent runs only on the delta of newly added and modified tests. For example, upgrading from PyTorch 2.13 to 2.14 required evaluating roughly 4000 changed tests instead of re-analyzing tens of thousands of tests in the full suite. daily.dev helps teams stay ahead of PyTorch version upgrades and their downstream CI impact."}}]} ```