Repository navigation
[Docs] [templates] Reorganize module design documentation - #5139
Conversation
Signed-off-by: hsliuustc0106 <liuhongsheng4@huawei.com>
727ae8b to
80fd96c
Compare
Signed-off-by: hsliuustc0106 <liuhongsheng4@huawei.com>
hsliuustc0106
added
the
refactor
ChatGPT Codex Connector
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 69766960f2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| - design/module/dit_module.md | ||
| - design/module/entrypoint_module.md | ||
| - design/module/async_omni_architecture.md | ||
| - Entrypoints: design/module/entrypoints.md |
There was a problem hiding this comment.
This rename deletes the previously published module pages (entrypoint_module.md, ar_module.md, dit_module.md, and async_omni_architecture.md) while moving navigation to new filenames, and I found no redirect/alias configuration in mkdocs.yml or docs/. After the next docs deploy, existing links such as /design/module/entrypoint_module/ will 404 even though the content was only reorganized; please keep lightweight compatibility pages or add redirects for the old URLs before removing them.
Useful? React with 👍 / 👎.
Gaohan123
left a comment
Gaohan123
left a comment
There was a problem hiding this comment.
Overall LGTM. And there are several modules uncovered:
- hardware platforms
- tests
| **Rule:** Public protocol values MUST be validated and converted to internal | ||
| request contracts before engine submission. | ||
|
|
||
| ### ENTRY-INV-003: Streaming preserves request identity |
There was a problem hiding this comment.
Maybe we should add Entrypoints should not contain model specific logic outside of common abstractions here as well. Too much model specific code in the entrypoints makes them harder to read and can cause the behavior to be inconsistent, both across models in a given entrypoints, and across the online / offline paths for similar calls.
For example serving speech has tons of model specific code (example 1, example 2, etc). Isolating the model specific code in good abstractions that can called generically is more ideal since it makes the behaviors easier to maintain, and also lets us add patterns for good unit tests instead of having to e2e everything
There was a problem hiding this comment.
I agree. This PR scope only provides the templates for different modules. After refactoring, the module maintainers will be responsible for providing details
| **Rule:** Defaults, files, environment variables, and CLI overrides MUST have a | ||
| documented and deterministic precedence. | ||
|
|
||
| ### CONFIG-INV-003: Runtime modules consume validated configuration |
There was a problem hiding this comment.
I think we should also consider something like environment variables should be written to config objects at initialization time.
Especially with the direction we are going in the the Omni config, the object should be the source of truth for the correct value after everything is parsed. for example, if we read a value into the config at init time, and then write + read to the corresponding env var at inference time instead of using the config object, it will cause bad behaviors.
This would also help ensure that env vars that are missing from configs are added where needed
hsliuustc0106
changed the title
tests will be listed in an independent folder |
| @@ -0,0 +1,48 @@ | |||
| --- | |||
There was a problem hiding this comment.
@Gaohan123 this is designed for hardware
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
…ct#5139) Signed-off-by: hsliuustc0106 <liuhongsheng4@huawei.com> Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
3 participants
Summary
This PR implements the module-oriented design documentation layout proposed in #5137 against current
main.engine_orchestration.mdstage_runtime.mderror_contracts.mdentrypoints.mdinput_output_modality_contracts.mddocs/design/module/archive/and removes them from active navigation; no redirect stubs are addedvllm_omni_config.mdexplicitly deferred and draft pending configuration refactoring, without assigning a stable invariant namespaceAll module contracts remain
draft. Open refactors and RFCs are cited as in-flight context, not presented as current or normative behavior.Confirmed technical owners
Technical owners are intentionally separate from each contract's independent required reviewers.
Review guide
Please focus review on:
Boundary follow-ups remain tracked in #5227, #5441, #5453, and #5570.
Impact and non-goals
Documentation only. This PR does not change runtime behavior, public APIs, deployment configuration, or error behavior. It does not make in-flight refactors normative and does not introduce compatibility redirects for archived pages.
Validation
origin/mainthrough a normal merge; no history rewritemkdocs build --strictgit diff --checkAll checks passed. Runtime and device tests were not run because the change is documentation-only.
Related to #5137.