-
Notifications
You must be signed in to change notification settings - Fork 2.8k
feat: Enable on-disk storage for Qdrant vectors and HNSW index #7182
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
- Configure vectors with on_disk: true to reduce memory usage - Configure HNSW index with on_disk: true for low memory footprint - Improve precision with m=64 and ef_construct=512 parameters - Increase hnsw_ef from 128 to 256 for better search precision - Remove error throwing in deletePointsByMultipleFilePaths to prevent disruption
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you for your contribution! I've reviewed the changes and have some suggestions for improvement. The on-disk storage configuration is a valuable addition for reducing memory usage, but there are a few considerations around performance trade-offs and configuration flexibility that would be worth addressing.
| on_disk: true, // Store vectors on disk for low memory usage | ||
| }, | ||
| hnsw_config: { | ||
| m: 64, // Increased for better precision |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The HNSW parameters (m=64, ef_construct=512) are significantly higher than typical defaults. Could you add a comment explaining why these specific values were chosen? This would help future maintainers understand the precision vs. performance trade-offs.
| }) | ||
|
|
||
| throw error | ||
| } |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Removing the here silently swallows errors, which might hide important issues. While preventing disruption is good, consider:
- Logging at a higher severity level (console.error is already there, which is good)
- Implementing a retry mechanism for transient failures
- Only catching specific non-critical errors and re-throwing critical ones
For example:
- Update test expectations for createCollection to include on_disk: true for vectors - Add hnsw_config expectations with m: 64, ef_construct: 512, and on_disk: true - Update search test expectations to use hnsw_ef: 256 instead of 128 These changes align the tests with the PR's implementation of on-disk storage for reduced memory usage and improved precision.
Remove inline comments that describe what was changed/increased/added as they pollute the codebase. The configuration values remain unchanged.
The PR changed hnsw_ef from 256 to 128 for improved search precision, but the tests were still expecting the old value. This commit updates all test expectations to match the new hnsw_ef value of 128.
* Follow symlinks in rooignore checks (RooCodeInc#7405) * Sonic -> Grok Code Fast (RooCodeInc#7426) * chore: add changeset for v3.26.0 (RooCodeInc#7428) * Changeset version bump (RooCodeInc#7429) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * feat: Add Vercel AI Gateway provider integration (RooCodeInc#7396) Co-authored-by: daniel-lxs <[email protected]> Co-authored-by: cte <[email protected]> * feat: Enable on-disk storage for Qdrant vectors and HNSW index (RooCodeInc#7182) * fix: use anthropic protocol for token counting when using anthropic models via Vercel AI Gateway (RooCodeInc#7433) - Added condition in getApiProtocol to return 'anthropic' for vercel-ai-gateway when modelId starts with 'anthropic/' - Added tests for Vercel AI Gateway provider protocol detection This ensures proper token counting for Anthropic models accessed through Vercel AI Gateway, as Anthropic and OpenAI count tokens differently (Anthropic excludes cache tokens from input count, OpenAI includes them). * fix: remove duplicate cache display in task header (RooCodeInc#7443) * Random chat text area cleanup (RooCodeInc#7436) * Update @roo-code/cloud to enable roomote control for cloud agents (RooCodeInc#7446) Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com> * Always set remoteControlEnabled to true for cloud agents (RooCodeInc#7448) * chore: add changeset for v3.26.1 (RooCodeInc#7459) * feat: show model ID in API configuration dropdown (RooCodeInc#7423) * feat: update tooltip component to match native VSCode tooltip shadow styling (RooCodeInc#7457) Co-authored-by: Roo Code <[email protected]> Co-authored-by: cte <[email protected]> * Add support for Vercel embeddings (RooCodeInc#7445) Co-authored-by: daniel-lxs <[email protected]> * Remove dot before model display (RooCodeInc#7461) * Update contributors list (RooCodeInc#7109) Co-authored-by: mrubens <[email protected]> * Update 3.26.1 changeset (RooCodeInc#7463) * Changeset version bump (RooCodeInc#7460) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * Add type for RooCodeEventName.TaskSpawned (RooCodeInc#7465) * fix: hide .rooignore'd files from environment details by default (RooCodeInc#7369) * fix: change default showRooIgnoredFiles to false to hide ignored files - Changed default value from true to false across all files - Updated tests to reflect the new default behavior - This prevents ignored files from appearing in environment details Fixes RooCodeInc#7368 * fix: update tests to match new showRooIgnoredFiles default * fix: update test expectation to match new showRooIgnoredFiles default value The PR changed the default value of showRooIgnoredFiles from true to false, so the test needs to expect false instead of true when calling formatFilesList. --------- Co-authored-by: Roo Code <[email protected]> Co-authored-by: daniel-lxs <[email protected]> * fix: exclude browser scroll actions from repetition detection (RooCodeInc#7471) - Modified ToolRepetitionDetector to skip repetition detection for browser_action scroll_down and scroll_up actions - Added isBrowserScrollAction() helper method to identify scroll actions - Added comprehensive tests for the new behavior - Fixes issue where multiple scroll actions were incorrectly flagged as being stuck in a loop Resolves: RooCodeInc#7470 Co-authored-by: Roo Code <[email protected]> * Fix GPT-5 Responses API issues with condensing and image support (RooCodeInc#7067) Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com> Co-authored-by: Roo Code <[email protected]> Co-authored-by: Hannes Rudolph <[email protected]> * Bump cloud to 0.25.0 (RooCodeInc#7475) * feat: add image generation tool with OpenRouter integration (RooCodeInc#7474) Co-authored-by: Matt Rubens <[email protected]> Co-authored-by: cte <[email protected]> * Make the default image filename more generic (RooCodeInc#7479) * Release v3.26.2 (RooCodeInc#7490) * Support free imagegen (RooCodeInc#7493) * feat: update OpenRouter API to support input/output modalities and filter image generation models (RooCodeInc#7492) * Add padding to image model picker (RooCodeInc#7494) * fix: prevent dirty state on initial mount in ImageGenerationSettings (RooCodeInc#7495) * Changeset version bump (RooCodeInc#7491) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * Show console logging in vitests when the --no-silent flag is set (RooCodeInc#7467) By default, all of the tests run in silent mode with monkey-patched the console logging so no console logging will ever appear in test output. This confuses the agent- sometimes it will add console logging to help it debug things, and it won't see the logs that it expects. Adds src/utils/vitest-verbosity.ts to handle verbosity resolution and console logging. Modifies src/vitest.config.ts and webview-ui/vitest.config.ts to integrate the new verbosity control. Removes manual console suppression from src/vitest.setup.ts and webview-ui/vitest.setup.ts as it's now handled dynamically. Co-authored-by: Chris Hasson <[email protected]> * Move @roo-code/cloud to the Roo-Code repo (RooCodeInc#7503) * Refactor the extension bridge (RooCodeInc#7515) * Implement deferred task subscriptions (RooCodeInc#7517) * feat: add optional input image parameter to image generation tool (RooCodeInc#7525) Co-authored-by: Roo Code <[email protected]> Co-authored-by: Daniel Riccio <[email protected]> * feat: sync extension bridge settings with cloud (RooCodeInc#7535) - Use CloudService.getUserSettings() for remoteControlEnabled instead of global state - Update CloudService.updateUserSettings when toggling remote control - Add BridgeOrchestrator.connectOrDisconnect handling in settings update handler - Remove dependency on contentProxy/globalSettings for remote control state --------- Co-authored-by: Roo Code <[email protected]> Co-authored-by: John Richmond <[email protected]> * refactor: flatten image generation settings structure (RooCodeInc#7536) * chore: add changeset for v3.26.3 (RooCodeInc#7541) * Changeset version bump (RooCodeInc#7542) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * Mode and provider profile selector (RooCodeInc#7545) * Putting the Roo in Roo-leases (RooCodeInc#7546) * Fix evals (RooCodeInc#7547) * fix: special tokens should not break task processing (RooCodeInc#7540) * docs: update PR template branding from Roo Code to Costrict * feat: optimize memory usage for image handling in webview (RooCodeInc#7556) * feat: optimize memory usage for image handling in webview - Replace base64 image data with webview URIs to reduce memory footprint - Add proper resource roots to webview for workspace file access - Implement convertToWebviewUri method for safe file-to-URI conversion - Update ImageViewer to handle both webview URIs and file paths separately - Add image message type for proper image rendering in chat - Improve error handling and display for failed image loads - Add comprehensive tests for ImageViewer component - Format display paths as relative for better readability This change significantly reduces memory usage by avoiding base64 encoding of images and instead using VSCode's webview URI system for direct file access. Images are now loaded on-demand from disk rather than being held in memory as base64 strings. * fix: address PR review comments - Use safeJsonParse instead of JSON.parse in ChatRow.tsx - Add type definition for parsed image info - Add more specific error types in ClineProvider.ts - Add comprehensive JSDoc comments to ImageBlock.tsx - Improve error handling and type safety * fix: address MrUbens' review comments - Remove hardcoded 'rc1' pattern in formatDisplayPath, use generic workspace detection - Internationalize 'No image data' text using i18n system * chore: remove useless comment * chore(i18n): add image.noData to all locales to fix translation check * test: update ImageViewer.spec to align with i18n key and flexible path formatting * feat: rename Account tab to Cloud tab (RooCodeInc#7558) Co-authored-by: Roo Code <[email protected]> Co-authored-by: Matt Rubens <[email protected]> * feat: add Ollama API key support for Turbo mode (RooCodeInc#7425) * feat: add Ollama API key support for Turbo mode - Add ollamaApiKey field to ProviderSettings schema - Add ollamaApiKey to SECRET_STATE_KEYS for secure storage - Update Ollama and NativeOllama providers to use API key for authentication - Add UI field for Ollama API key (shown when custom base URL is provided) - Add test coverage for API key functionality This enables users to use Ollama Turbo with datacenter-grade hardware by providing an API key for authenticated Ollama instances or cloud services. * fix: use VSCodeTextField for Ollama API key field Remove non-existent ApiKeyField import and use standard VSCodeTextField with password type, matching other provider implementations * Add missing translation keys for Ollama API key support - Add providers.ollama.apiKey and providers.ollama.apiKeyHelp to all 18 language files - Support for authenticated Ollama instances and cloud services - Relates to PR RooCodeInc#7425 * refactor: improve type safety for Ollama client configuration - Replace 'any' type with proper OllamaOptions (Config) type - Import Config type from ollama package for better type checking --------- Co-authored-by: Roo Code <[email protected]> Co-authored-by: Daniel Riccio <[email protected]> --------- Co-authored-by: Matt Rubens <[email protected]> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Josh <[email protected]> Co-authored-by: daniel-lxs <[email protected]> Co-authored-by: cte <[email protected]> Co-authored-by: Daniel <[email protected]> Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com> Co-authored-by: roomote[bot] <219738659+roomote[bot]@users.noreply.github.com> Co-authored-by: Roo Code <[email protected]> Co-authored-by: mrubens <[email protected]> Co-authored-by: Hannes Rudolph <[email protected]> Co-authored-by: Chris Hasson <[email protected]> Co-authored-by: Chris Hasson <[email protected]> Co-authored-by: John Richmond <[email protected]> Co-authored-by: Piotr Wilkin (ilintar) <[email protected]>
* Follow symlinks in rooignore checks (RooCodeInc#7405) * Sonic -> Grok Code Fast (RooCodeInc#7426) * chore: add changeset for v3.26.0 (RooCodeInc#7428) * Changeset version bump (RooCodeInc#7429) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * feat: Add Vercel AI Gateway provider integration (RooCodeInc#7396) Co-authored-by: daniel-lxs <[email protected]> Co-authored-by: cte <[email protected]> * feat: Enable on-disk storage for Qdrant vectors and HNSW index (RooCodeInc#7182) * fix: use anthropic protocol for token counting when using anthropic models via Vercel AI Gateway (RooCodeInc#7433) - Added condition in getApiProtocol to return 'anthropic' for vercel-ai-gateway when modelId starts with 'anthropic/' - Added tests for Vercel AI Gateway provider protocol detection This ensures proper token counting for Anthropic models accessed through Vercel AI Gateway, as Anthropic and OpenAI count tokens differently (Anthropic excludes cache tokens from input count, OpenAI includes them). * fix: remove duplicate cache display in task header (RooCodeInc#7443) * Random chat text area cleanup (RooCodeInc#7436) * Update @roo-code/cloud to enable roomote control for cloud agents (RooCodeInc#7446) Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com> * Always set remoteControlEnabled to true for cloud agents (RooCodeInc#7448) * chore: add changeset for v3.26.1 (RooCodeInc#7459) * feat: show model ID in API configuration dropdown (RooCodeInc#7423) * feat: update tooltip component to match native VSCode tooltip shadow styling (RooCodeInc#7457) Co-authored-by: Roo Code <[email protected]> Co-authored-by: cte <[email protected]> * Add support for Vercel embeddings (RooCodeInc#7445) Co-authored-by: daniel-lxs <[email protected]> * Remove dot before model display (RooCodeInc#7461) * Update contributors list (RooCodeInc#7109) Co-authored-by: mrubens <[email protected]> * Update 3.26.1 changeset (RooCodeInc#7463) * Changeset version bump (RooCodeInc#7460) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * Add type for RooCodeEventName.TaskSpawned (RooCodeInc#7465) * fix: hide .rooignore'd files from environment details by default (RooCodeInc#7369) * fix: change default showRooIgnoredFiles to false to hide ignored files - Changed default value from true to false across all files - Updated tests to reflect the new default behavior - This prevents ignored files from appearing in environment details Fixes RooCodeInc#7368 * fix: update tests to match new showRooIgnoredFiles default * fix: update test expectation to match new showRooIgnoredFiles default value The PR changed the default value of showRooIgnoredFiles from true to false, so the test needs to expect false instead of true when calling formatFilesList. --------- Co-authored-by: Roo Code <[email protected]> Co-authored-by: daniel-lxs <[email protected]> * fix: exclude browser scroll actions from repetition detection (RooCodeInc#7471) - Modified ToolRepetitionDetector to skip repetition detection for browser_action scroll_down and scroll_up actions - Added isBrowserScrollAction() helper method to identify scroll actions - Added comprehensive tests for the new behavior - Fixes issue where multiple scroll actions were incorrectly flagged as being stuck in a loop Resolves: RooCodeInc#7470 Co-authored-by: Roo Code <[email protected]> * Fix GPT-5 Responses API issues with condensing and image support (RooCodeInc#7067) Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com> Co-authored-by: Roo Code <[email protected]> Co-authored-by: Hannes Rudolph <[email protected]> * Bump cloud to 0.25.0 (RooCodeInc#7475) * feat: add image generation tool with OpenRouter integration (RooCodeInc#7474) Co-authored-by: Matt Rubens <[email protected]> Co-authored-by: cte <[email protected]> * Make the default image filename more generic (RooCodeInc#7479) * Release v3.26.2 (RooCodeInc#7490) * Support free imagegen (RooCodeInc#7493) * feat: update OpenRouter API to support input/output modalities and filter image generation models (RooCodeInc#7492) * Add padding to image model picker (RooCodeInc#7494) * fix: prevent dirty state on initial mount in ImageGenerationSettings (RooCodeInc#7495) * Changeset version bump (RooCodeInc#7491) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * Show console logging in vitests when the --no-silent flag is set (RooCodeInc#7467) By default, all of the tests run in silent mode with monkey-patched the console logging so no console logging will ever appear in test output. This confuses the agent- sometimes it will add console logging to help it debug things, and it won't see the logs that it expects. Adds src/utils/vitest-verbosity.ts to handle verbosity resolution and console logging. Modifies src/vitest.config.ts and webview-ui/vitest.config.ts to integrate the new verbosity control. Removes manual console suppression from src/vitest.setup.ts and webview-ui/vitest.setup.ts as it's now handled dynamically. Co-authored-by: Chris Hasson <[email protected]> * Move @roo-code/cloud to the Roo-Code repo (RooCodeInc#7503) * Refactor the extension bridge (RooCodeInc#7515) * Implement deferred task subscriptions (RooCodeInc#7517) * feat: add optional input image parameter to image generation tool (RooCodeInc#7525) Co-authored-by: Roo Code <[email protected]> Co-authored-by: Daniel Riccio <[email protected]> * feat: sync extension bridge settings with cloud (RooCodeInc#7535) - Use CloudService.getUserSettings() for remoteControlEnabled instead of global state - Update CloudService.updateUserSettings when toggling remote control - Add BridgeOrchestrator.connectOrDisconnect handling in settings update handler - Remove dependency on contentProxy/globalSettings for remote control state --------- Co-authored-by: Roo Code <[email protected]> Co-authored-by: John Richmond <[email protected]> * refactor: flatten image generation settings structure (RooCodeInc#7536) * chore: add changeset for v3.26.3 (RooCodeInc#7541) * Changeset version bump (RooCodeInc#7542) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * Mode and provider profile selector (RooCodeInc#7545) * Putting the Roo in Roo-leases (RooCodeInc#7546) * Fix evals (RooCodeInc#7547) * fix: special tokens should not break task processing (RooCodeInc#7540) * docs: update PR template branding from Roo Code to Costrict * feat: optimize memory usage for image handling in webview (RooCodeInc#7556) * feat: optimize memory usage for image handling in webview - Replace base64 image data with webview URIs to reduce memory footprint - Add proper resource roots to webview for workspace file access - Implement convertToWebviewUri method for safe file-to-URI conversion - Update ImageViewer to handle both webview URIs and file paths separately - Add image message type for proper image rendering in chat - Improve error handling and display for failed image loads - Add comprehensive tests for ImageViewer component - Format display paths as relative for better readability This change significantly reduces memory usage by avoiding base64 encoding of images and instead using VSCode's webview URI system for direct file access. Images are now loaded on-demand from disk rather than being held in memory as base64 strings. * fix: address PR review comments - Use safeJsonParse instead of JSON.parse in ChatRow.tsx - Add type definition for parsed image info - Add more specific error types in ClineProvider.ts - Add comprehensive JSDoc comments to ImageBlock.tsx - Improve error handling and type safety * fix: address MrUbens' review comments - Remove hardcoded 'rc1' pattern in formatDisplayPath, use generic workspace detection - Internationalize 'No image data' text using i18n system * chore: remove useless comment * chore(i18n): add image.noData to all locales to fix translation check * test: update ImageViewer.spec to align with i18n key and flexible path formatting * feat: rename Account tab to Cloud tab (RooCodeInc#7558) Co-authored-by: Roo Code <[email protected]> Co-authored-by: Matt Rubens <[email protected]> * feat: add Ollama API key support for Turbo mode (RooCodeInc#7425) * feat: add Ollama API key support for Turbo mode - Add ollamaApiKey field to ProviderSettings schema - Add ollamaApiKey to SECRET_STATE_KEYS for secure storage - Update Ollama and NativeOllama providers to use API key for authentication - Add UI field for Ollama API key (shown when custom base URL is provided) - Add test coverage for API key functionality This enables users to use Ollama Turbo with datacenter-grade hardware by providing an API key for authenticated Ollama instances or cloud services. * fix: use VSCodeTextField for Ollama API key field Remove non-existent ApiKeyField import and use standard VSCodeTextField with password type, matching other provider implementations * Add missing translation keys for Ollama API key support - Add providers.ollama.apiKey and providers.ollama.apiKeyHelp to all 18 language files - Support for authenticated Ollama instances and cloud services - Relates to PR RooCodeInc#7425 * refactor: improve type safety for Ollama client configuration - Replace 'any' type with proper OllamaOptions (Config) type - Import Config type from ollama package for better type checking --------- Co-authored-by: Roo Code <[email protected]> Co-authored-by: Daniel Riccio <[email protected]> * Disconnect extension bridge on logout (RooCodeInc#7563) * Disconnect extension bridge on logout * Remove bad test * Cleanup * Fix claudeCode.notFound translation key (RooCodeInc#7571) I think I saw someone post a screenshot where "errors.claudeCode.notFound" was shown untranslated. * v3.26.4 (RooCodeInc#7579) * Update contributors list (RooCodeInc#7462) Co-authored-by: mrubens <[email protected]> * Changeset version bump (RooCodeInc#7580) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * feat: add configurable embedding batch size for code indexing (RooCodeInc#7464) Co-authored-by: Roo Code <[email protected]> Co-authored-by: Daniel Riccio <[email protected]> * Shows a pill with the base Roo Code Cloud URL when not pointing to pr… (RooCodeInc#7555) Co-authored-by: Roo Code <[email protected]> Co-authored-by: Matt Rubens <[email protected]> Co-authored-by: roomote[bot] <219738659+roomote[bot]@users.noreply.github.com> --------- Co-authored-by: Matt Rubens <[email protected]> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Josh <[email protected]> Co-authored-by: daniel-lxs <[email protected]> Co-authored-by: cte <[email protected]> Co-authored-by: Daniel <[email protected]> Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com> Co-authored-by: roomote[bot] <219738659+roomote[bot]@users.noreply.github.com> Co-authored-by: Roo Code <[email protected]> Co-authored-by: mrubens <[email protected]> Co-authored-by: Hannes Rudolph <[email protected]> Co-authored-by: Chris Hasson <[email protected]> Co-authored-by: Chris Hasson <[email protected]> Co-authored-by: John Richmond <[email protected]> Co-authored-by: Piotr Wilkin (ilintar) <[email protected]> Co-authored-by: Christiaan Arnoldus <[email protected]> Co-authored-by: Bruno Bergher <[email protected]>
* Follow symlinks in rooignore checks (RooCodeInc#7405) * Sonic -> Grok Code Fast (RooCodeInc#7426) * chore: add changeset for v3.26.0 (RooCodeInc#7428) * Changeset version bump (RooCodeInc#7429) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Matt Rubens <[email protected]> * feat: Add Vercel AI Gateway provider integration (
Summary
This PR configures Qdrant to store vectors and HNSW index on disk for high precision with low memory usage.
Based on https://qdrant.tech/documentation/guides/optimize/#2-high-precision-with-low-memory-usage
Changes
on_disk: trueto reduce memory footprintm=64andef_construct=512for better search precisionhnsw_effrom 128 to 256 for improved search accuracydeletePointsByMultipleFilePathsto prevent disruptionBenefits
Testing
Important
Enable on-disk storage for Qdrant vectors and HNSW index, adjust HNSW parameters for precision, and improve error handling.
on_disk: truefor vectors and HNSW index inqdrant-client.tsandqdrant-client.spec.tsto reduce memory usage.m=64andef_construct=512inqdrant-client.tsandqdrant-client.spec.tsfor improved precision.hnsw_effrom 128 to 256 inqdrant-client.tsfor better search accuracy.deletePointsByMultipleFilePathsinqdrant-client.tsto prevent task disruption.This description was created by
for 74fe2d1. You can customize this summary. It will automatically update as commits are pushed.