[WIP] refactor: Implement modular candle-binding architecture (#254) #266
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Restructure codebase into modular layers (core/, ffi/, model_architectures/, classifiers/) - Add unified error handling and configuration loading systems - Implement dual-path architecture for traditional and LoRA models - Add comprehensive FFI layer with memory safety Maintains backward compatibility while enabling future model integrations. refactor: Implement modular candle-binding architecture - Restructure codebase into modular layers (core/, ffi/, model_architectures/, classifiers/) - Add unified error handling and configuration loading systems - Implement dual-path architecture for traditional and LoRA models - Add comprehensive FFI layer with memory safety Maintains backward compatibility while enabling future model integrations. Signed-off-by: OneZero-Y <[email protected]>
✅ Deploy Preview for vllm-semantic-router ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
👥 vLLM Semantic Team NotificationThe following members have been identified for the changed files in this PR and have been automatically assigned: 📁
|
|
@OneZero-Y @Xunzhuo Let's have the following resolved before merging
|
feat:unit tests for candle refactoring feat:unit tests for candle refactoring Signed-off-by: OneZero-Y <[email protected]>
|
@OneZero-Y now since we work on the feature branch, how about you use this branch for both refactoring and new embedding models? |
Closed
|
@rootfs OK, I'll advance the embedded model on this branch |
|
@OneZero-Y that's great! I'll switch to this work as soon as i can. |
Comment on lines
+89
to
+104
| let handles = vec![ | ||
| self.spawn_intent_task(texts_owned.clone(), Arc::clone(&intent_results)), | ||
| self.spawn_pii_task(texts_owned.clone(), Arc::clone(&pii_results)), | ||
| self.spawn_security_task(texts_owned, Arc::clone(&security_results)), | ||
| ]; | ||
|
|
||
| // Wait for all threads to complete | ||
| for handle in handles { | ||
| handle.join().map_err(|_| { | ||
| let unified_err = concurrency_error( | ||
| "thread join", | ||
| "Failed to join parallel classification thread", | ||
| ); | ||
| candle_core::Error::from(unified_err) | ||
| })?; | ||
| } |
There was a problem hiding this comment.
This could be simplified a bit. Something like
let intent_handle = thread::spawn(|| intent_task(texts)); // slice is fine, no need to own the data.
let pii_handle = ... same
let security_handle = ... same
let intent_results = intent_handle.join()?; // map_err omitted
let pii_results = pii_handle.join()?;
let security_results = security_handle.join()?;
There was a problem hiding this comment.
Since we're on the topic of threads - you may like some of the abstractions that the rayon crate provides
There was a problem hiding this comment.
thank you @ivarflakstad
Comment on lines
+162
to
+168
| pub fn parallel_detect(&self, texts: &[&str]) -> Result<Vec<PIIResult>> { | ||
| let mut results = Vec::new(); | ||
| for text in texts { | ||
| results.push(self.detect_pii(text)?); | ||
| } | ||
| Ok(results) | ||
| } |
There was a problem hiding this comment.
If you want this to be in parallel you could do something like
// add `use rayon::prelude::*;` at top of file
Ok(texts.par_iter().map(|text| self.detect_pii(text)?).collect())There was a problem hiding this comment.
Though I'm starting to suspect that what you actually want, for the long term, is an async runtime.
There was a problem hiding this comment.
@ivarflakstad thanks for looking into this. On a separate note, for async to run most efficiently, would you help look at the if locking is done the right way?
There was a problem hiding this comment.
Sure :)
Are you thinking about any specific locks in particular? (pr is fairly large 😉 )
There was a problem hiding this comment.
thank you @ivarflakstad
The classify_text is currently protected under lock. This could get us performance hit, would you help share your ideas? Thanks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Maintains backward compatibility while enabling future model integrations.
refactor: Implement modular candle-binding architecture
Maintains backward compatibility while enabling future model integrations.
What type of PR is this?
What this PR does / why we need it:
Which issue(s) this PR fixes:
Fixes #
Release Notes: Yes/No