xithead 1.2.2 → 1.5.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.actrc +14 -0
- package/.actrc.local.example +15 -0
- package/.changelog.md +2 -0
- package/.claude/agents/academic-anthropologist.md +125 -0
- package/.claude/agents/academic-geographer.md +127 -0
- package/.claude/agents/academic-historian.md +123 -0
- package/.claude/agents/academic-narratologist.md +118 -0
- package/.claude/agents/academic-psychologist.md +118 -0
- package/.claude/agents/academic-statistician.md +144 -0
- package/.claude/agents/accounts-payable-agent.md +205 -0
- package/.claude/agents/agentic-identity-trust.md +394 -0
- package/.claude/agents/agents-orchestrator.md +367 -0
- package/.claude/agents/automation-governance-architect.md +216 -0
- package/.claude/agents/blender-addon-engineer.md +243 -0
- package/.claude/agents/business-strategist.md +488 -0
- package/.claude/agents/change-management-consultant.md +497 -0
- package/.claude/agents/chief-financial-officer.md +388 -0
- package/.claude/agents/corporate-training-designer.md +192 -0
- package/.claude/agents/customer-service.md +398 -0
- package/.claude/agents/customer-success-manager.md +460 -0
- package/.claude/agents/data-consolidation-agent.md +60 -0
- package/.claude/agents/data-privacy-officer.md +412 -0
- package/.claude/agents/design-brand-guardian.md +322 -0
- package/.claude/agents/design-image-prompt-engineer.md +236 -0
- package/.claude/agents/design-inclusive-visuals-specialist.md +73 -0
- package/.claude/agents/design-persona-walkthrough.md +272 -0
- package/.claude/agents/design-ui-designer.md +386 -0
- package/.claude/agents/design-ui-finish-gate-reviewer.md +217 -0
- package/.claude/agents/design-ux-architect.md +482 -0
- package/.claude/agents/design-ux-researcher.md +329 -0
- package/.claude/agents/design-visual-storyteller.md +149 -0
- package/.claude/agents/design-whimsy-injector.md +460 -0
- package/.claude/agents/economy-designer.md +156 -0
- package/.claude/agents/engineering-ai-data-remediation-engineer.md +211 -0
- package/.claude/agents/engineering-ai-engineer.md +146 -0
- package/.claude/agents/engineering-api-platform-engineer.md +181 -0
- package/.claude/agents/engineering-ats-validator-architect.md +383 -0
- package/.claude/agents/engineering-autonomous-optimization-architect.md +121 -0
- package/.claude/agents/engineering-backend-architect.md +255 -0
- package/.claude/agents/engineering-china-network-engineer.md +252 -0
- package/.claude/agents/engineering-cms-developer.md +536 -0
- package/.claude/agents/engineering-code-reviewer.md +76 -0
- package/.claude/agents/engineering-codebase-onboarding-engineer.md +173 -0
- package/.claude/agents/engineering-data-engineer.md +349 -0
- package/.claude/agents/engineering-data-visualization-engineer.md +151 -0
- package/.claude/agents/engineering-database-optimizer.md +188 -0
- package/.claude/agents/engineering-database-reliability-engineer.md +201 -0
- package/.claude/agents/engineering-desktop-app-engineer.md +209 -0
- package/.claude/agents/engineering-developer-tooling-engineer.md +153 -0
- package/.claude/agents/engineering-devops-automator.md +385 -0
- package/.claude/agents/engineering-drupal-performance.md +347 -0
- package/.claude/agents/engineering-drupal-shopping-cart.md +360 -0
- package/.claude/agents/engineering-electrical-engineer.md +575 -0
- package/.claude/agents/engineering-email-intelligence-engineer.md +390 -0
- package/.claude/agents/engineering-embedded-firmware-engineer.md +196 -0
- package/.claude/agents/engineering-feishu-integration-developer.md +606 -0
- package/.claude/agents/engineering-filament-optimization-specialist.md +283 -0
- package/.claude/agents/engineering-finops-engineer.md +164 -0
- package/.claude/agents/engineering-frontend-developer.md +257 -0
- package/.claude/agents/engineering-gaussdb-expert.md +334 -0
- package/.claude/agents/engineering-git-workflow-master.md +98 -0
- package/.claude/agents/engineering-i18n-engineer.md +214 -0
- package/.claude/agents/engineering-identity-access-engineer.md +217 -0
- package/.claude/agents/engineering-incident-response-commander.md +444 -0
- package/.claude/agents/engineering-iot-fleet-engineer.md +148 -0
- package/.claude/agents/engineering-it-service-manager.md +561 -0
- package/.claude/agents/engineering-knowledge-graph-engineer.md +372 -0
- package/.claude/agents/engineering-llm-post-training-engineer.md +166 -0
- package/.claude/agents/engineering-minimal-change-engineer.md +207 -0
- package/.claude/agents/engineering-mobile-app-builder.md +493 -0
- package/.claude/agents/engineering-mobile-release-engineer.md +164 -0
- package/.claude/agents/engineering-multi-agent-systems-architect.md +600 -0
- package/.claude/agents/engineering-network-engineer.md +245 -0
- package/.claude/agents/engineering-orgscript-engineer.md +113 -0
- package/.claude/agents/engineering-payments-billing-engineer.md +228 -0
- package/.claude/agents/engineering-pdf-engine-architect.md +696 -0
- package/.claude/agents/engineering-platform-engineer.md +270 -0
- package/.claude/agents/engineering-privacy-engineer.md +152 -0
- package/.claude/agents/engineering-prompt-engineer.md +209 -0
- package/.claude/agents/engineering-rag-pipeline-engineer.md +450 -0
- package/.claude/agents/engineering-rapid-prototyper.md +466 -0
- package/.claude/agents/engineering-realtime-collaboration-engineer.md +200 -0
- package/.claude/agents/engineering-rust-refactoring-specialist.md +313 -0
- package/.claude/agents/engineering-search-relevance-engineer.md +237 -0
- package/.claude/agents/engineering-section-508-specialist.md +339 -0
- package/.claude/agents/engineering-senior-developer.md +176 -0
- package/.claude/agents/engineering-servicenow-developer-mentor.md +180 -0
- package/.claude/agents/engineering-software-architect.md +112 -0
- package/.claude/agents/engineering-solidity-smart-contract-engineer.md +522 -0
- package/.claude/agents/engineering-sre.md +90 -0
- package/.claude/agents/engineering-technical-writer.md +393 -0
- package/.claude/agents/engineering-universal-document-compiler.md +370 -0
- package/.claude/agents/engineering-uswds-developer.md +340 -0
- package/.claude/agents/engineering-video-streaming-engineer.md +151 -0
- package/.claude/agents/engineering-voice-ai-integration-engineer.md +584 -0
- package/.claude/agents/engineering-webassembly-engineer.md +153 -0
- package/.claude/agents/engineering-wechat-mini-program-developer.md +356 -0
- package/.claude/agents/engineering-wordpress-performance.md +346 -0
- package/.claude/agents/engineering-wordpress-shopping-cart.md +346 -0
- package/.claude/agents/esg-sustainability-officer.md +396 -0
- package/.claude/agents/finance-bookkeeper-controller.md +260 -0
- package/.claude/agents/finance-financial-analyst.md +234 -0
- package/.claude/agents/finance-fpa-analyst.md +263 -0
- package/.claude/agents/finance-investment-researcher.md +272 -0
- package/.claude/agents/finance-tax-strategist.md +239 -0
- package/.claude/agents/game-audio-engineer.md +264 -0
- package/.claude/agents/game-designer.md +167 -0
- package/.claude/agents/gis-3d-scene-developer.md +111 -0
- package/.claude/agents/gis-analyst.md +91 -0
- package/.claude/agents/gis-bim-specialist.md +108 -0
- package/.claude/agents/gis-cartography-designer.md +150 -0
- package/.claude/agents/gis-drone-reality-mapping.md +120 -0
- package/.claude/agents/gis-geoai-ml-engineer.md +105 -0
- package/.claude/agents/gis-geoprocessing-specialist.md +97 -0
- package/.claude/agents/gis-qa-engineer.md +133 -0
- package/.claude/agents/gis-solution-engineer.md +101 -0
- package/.claude/agents/gis-spatial-data-engineer.md +97 -0
- package/.claude/agents/gis-spatial-data-scientist.md +111 -0
- package/.claude/agents/gis-technical-consultant.md +86 -0
- package/.claude/agents/gis-web-gis-developer.md +108 -0
- package/.claude/agents/godot-gameplay-scripter.md +338 -0
- package/.claude/agents/godot-multiplayer-engineer.md +301 -0
- package/.claude/agents/godot-shader-developer.md +267 -0
- package/.claude/agents/government-digital-presales-consultant.md +363 -0
- package/.claude/agents/grant-writer.md +511 -0
- package/.claude/agents/healthcare-aging-parent-care-companion.md +414 -0
- package/.claude/agents/healthcare-clinical-evidence-agent.md +231 -0
- package/.claude/agents/healthcare-customer-service.md +389 -0
- package/.claude/agents/healthcare-innovation-strategist.md +433 -0
- package/.claude/agents/healthcare-marketing-compliance.md +395 -0
- package/.claude/agents/healthcare-sovereign-health-systems-agent.md +312 -0
- package/.claude/agents/hospitality-guest-services.md +603 -0
- package/.claude/agents/hr-onboarding.md +451 -0
- package/.claude/agents/identity-graph-operator.md +303 -0
- package/.claude/agents/language-translator.md +264 -0
- package/.claude/agents/legal-billing-time-tracking.md +569 -0
- package/.claude/agents/legal-client-intake.md +492 -0
- package/.claude/agents/legal-document-review.md +454 -0
- package/.claude/agents/level-designer.md +208 -0
- package/.claude/agents/loan-officer-assistant.md +555 -0
- package/.claude/agents/lsp-index-engineer.md +320 -0
- package/.claude/agents/ma-integration-manager.md +427 -0
- package/.claude/agents/macos-spatial-metal-engineer.md +337 -0
- package/.claude/agents/marketing-aeo-foundations.md +264 -0
- package/.claude/agents/marketing-agentic-search-optimizer.md +279 -0
- package/.claude/agents/marketing-ai-citation-strategist.md +172 -0
- package/.claude/agents/marketing-app-store-optimizer.md +321 -0
- package/.claude/agents/marketing-baidu-seo-specialist.md +226 -0
- package/.claude/agents/marketing-bilibili-content-strategist.md +199 -0
- package/.claude/agents/marketing-book-co-author.md +110 -0
- package/.claude/agents/marketing-carousel-growth-engine.md +204 -0
- package/.claude/agents/marketing-china-ecommerce-operator.md +283 -0
- package/.claude/agents/marketing-china-market-localization-strategist.md +283 -0
- package/.claude/agents/marketing-content-creator.md +54 -0
- package/.claude/agents/marketing-cross-border-ecommerce.md +259 -0
- package/.claude/agents/marketing-developer-community-builder.md +285 -0
- package/.claude/agents/marketing-douyin-strategist.md +149 -0
- package/.claude/agents/marketing-email-strategist.md +249 -0
- package/.claude/agents/marketing-global-podcast-strategist.md +206 -0
- package/.claude/agents/marketing-growth-hacker.md +54 -0
- package/.claude/agents/marketing-instagram-curator.md +113 -0
- package/.claude/agents/marketing-kuaishou-strategist.md +223 -0
- package/.claude/agents/marketing-linkedin-content-creator.md +214 -0
- package/.claude/agents/marketing-livestream-commerce-coach.md +305 -0
- package/.claude/agents/marketing-multi-platform-publisher.md +217 -0
- package/.claude/agents/marketing-podcast-strategist.md +277 -0
- package/.claude/agents/marketing-pr-communications-manager.md +473 -0
- package/.claude/agents/marketing-private-domain-operator.md +308 -0
- package/.claude/agents/marketing-reddit-community-builder.md +123 -0
- package/.claude/agents/marketing-seo-specialist.md +370 -0
- package/.claude/agents/marketing-short-video-editing-coach.md +412 -0
- package/.claude/agents/marketing-social-media-strategist.md +125 -0
- package/.claude/agents/marketing-tiktok-strategist.md +125 -0
- package/.claude/agents/marketing-twitter-engager.md +126 -0
- package/.claude/agents/marketing-video-optimization-specialist.md +119 -0
- package/.claude/agents/marketing-wechat-official-account.md +145 -0
- package/.claude/agents/marketing-weibo-strategist.md +240 -0
- package/.claude/agents/marketing-x-twitter-intelligence-analyst.md +161 -0
- package/.claude/agents/marketing-xiaohongshu-specialist.md +138 -0
- package/.claude/agents/marketing-zhihu-strategist.md +162 -0
- package/.claude/agents/medical-billing-coding-specialist.md +491 -0
- package/.claude/agents/narrative-designer.md +243 -0
- package/.claude/agents/operations-manager.md +399 -0
- package/.claude/agents/organizational-psychologist.md +391 -0
- package/.claude/agents/paid-media-auditor.md +71 -0
- package/.claude/agents/paid-media-creative-strategist.md +71 -0
- package/.claude/agents/paid-media-paid-social-strategist.md +71 -0
- package/.claude/agents/paid-media-ppc-strategist.md +71 -0
- package/.claude/agents/paid-media-programmatic-buyer.md +71 -0
- package/.claude/agents/paid-media-search-query-analyst.md +71 -0
- package/.claude/agents/paid-media-tracking-specialist.md +71 -0
- package/.claude/agents/personal-growth-mentor.md +159 -0
- package/.claude/agents/product-behavioral-nudge-engine.md +83 -0
- package/.claude/agents/product-dx-engineer.md +332 -0
- package/.claude/agents/product-feedback-synthesizer.md +119 -0
- package/.claude/agents/product-manager.md +469 -0
- package/.claude/agents/product-sprint-prioritizer.md +154 -0
- package/.claude/agents/product-trend-researcher.md +159 -0
- package/.claude/agents/project-management-experiment-tracker.md +198 -0
- package/.claude/agents/project-management-jira-workflow-steward.md +233 -0
- package/.claude/agents/project-management-meeting-notes-specialist.md +95 -0
- package/.claude/agents/project-management-project-shepherd.md +194 -0
- package/.claude/agents/project-management-studio-operations.md +200 -0
- package/.claude/agents/project-management-studio-producer.md +203 -0
- package/.claude/agents/project-manager-senior.md +135 -0
- package/.claude/agents/real-estate-buyer-seller.md +596 -0
- package/.claude/agents/recruitment-specialist.md +515 -0
- package/.claude/agents/report-distribution-agent.md +65 -0
- package/.claude/agents/research-synthesist.md +136 -0
- package/.claude/agents/resume-tailor.md +230 -0
- package/.claude/agents/retail-customer-returns.md +566 -0
- package/.claude/agents/roblox-avatar-creator.md +297 -0
- package/.claude/agents/roblox-experience-designer.md +306 -0
- package/.claude/agents/roblox-systems-scripter.md +332 -0
- package/.claude/agents/sales-account-strategist.md +227 -0
- package/.claude/agents/sales-coach.md +271 -0
- package/.claude/agents/sales-data-extraction-agent.md +67 -0
- package/.claude/agents/sales-deal-strategist.md +180 -0
- package/.claude/agents/sales-discovery-coach.md +225 -0
- package/.claude/agents/sales-engineer.md +182 -0
- package/.claude/agents/sales-offer-lead-gen-strategist.md +257 -0
- package/.claude/agents/sales-outbound-strategist.md +201 -0
- package/.claude/agents/sales-outreach.md +425 -0
- package/.claude/agents/sales-pipeline-analyst.md +267 -0
- package/.claude/agents/sales-proposal-strategist.md +217 -0
- package/.claude/agents/security-ai-generated-code-auditor.md +207 -0
- package/.claude/agents/security-appsec-engineer.md +536 -0
- package/.claude/agents/security-architect.md +307 -0
- package/.claude/agents/security-blockchain-security-auditor.md +482 -0
- package/.claude/agents/security-cloud-security-architect.md +576 -0
- package/.claude/agents/security-compliance-auditor.md +158 -0
- package/.claude/agents/security-incident-responder.md +450 -0
- package/.claude/agents/security-penetration-tester.md +399 -0
- package/.claude/agents/security-secrets-credential-engineer.md +189 -0
- package/.claude/agents/security-senior-secops.md +760 -0
- package/.claude/agents/security-threat-detection-engineer.md +553 -0
- package/.claude/agents/security-threat-intelligence-analyst.md +661 -0
- package/.claude/agents/specialized-chief-of-staff.md +279 -0
- package/.claude/agents/specialized-civil-engineer.md +356 -0
- package/.claude/agents/specialized-codebase-archaeologist.md +341 -0
- package/.claude/agents/specialized-cultural-intelligence-strategist.md +88 -0
- package/.claude/agents/specialized-developer-advocate.md +317 -0
- package/.claude/agents/specialized-document-generator.md +55 -0
- package/.claude/agents/specialized-fedramp-rmf-compliance.md +378 -0
- package/.claude/agents/specialized-focus-music-architect.md +175 -0
- package/.claude/agents/specialized-french-consulting-market.md +194 -0
- package/.claude/agents/specialized-korean-business-navigator.md +216 -0
- package/.claude/agents/specialized-master-plan-architect.md +157 -0
- package/.claude/agents/specialized-mcp-builder.md +260 -0
- package/.claude/agents/specialized-model-qa.md +528 -0
- package/.claude/agents/specialized-pricing-analyst.md +243 -0
- package/.claude/agents/specialized-salesforce-architect.md +182 -0
- package/.claude/agents/specialized-strategy-duel-agent.md +130 -0
- package/.claude/agents/specialized-workflow-architect.md +597 -0
- package/.claude/agents/study-abroad-advisor.md +282 -0
- package/.claude/agents/supply-chain-strategist.md +591 -0
- package/.claude/agents/support-analytics-reporter.md +387 -0
- package/.claude/agents/support-executive-summary-generator.md +212 -0
- package/.claude/agents/support-finance-tracker.md +480 -0
- package/.claude/agents/support-infrastructure-maintainer.md +628 -0
- package/.claude/agents/support-legal-compliance-checker.md +591 -0
- package/.claude/agents/support-support-responder.md +607 -0
- package/.claude/agents/technical-artist.md +233 -0
- package/.claude/agents/terminal-integration-specialist.md +70 -0
- package/.claude/agents/testing-accessibility-auditor.md +327 -0
- package/.claude/agents/testing-api-tester.md +334 -0
- package/.claude/agents/testing-evidence-collector.md +210 -0
- package/.claude/agents/testing-performance-benchmarker.md +283 -0
- package/.claude/agents/testing-reality-checker.md +249 -0
- package/.claude/agents/testing-test-automation-engineer.md +196 -0
- package/.claude/agents/testing-test-results-analyzer.md +330 -0
- package/.claude/agents/testing-tool-evaluator.md +416 -0
- package/.claude/agents/testing-workflow-optimizer.md +463 -0
- package/.claude/agents/unity-architect.md +271 -0
- package/.claude/agents/unity-editor-tool-developer.md +310 -0
- package/.claude/agents/unity-multiplayer-engineer.md +323 -0
- package/.claude/agents/unity-shader-graph-artist.md +269 -0
- package/.claude/agents/unreal-multiplayer-architect.md +364 -0
- package/.claude/agents/unreal-systems-engineer.md +322 -0
- package/.claude/agents/unreal-technical-artist.md +256 -0
- package/.claude/agents/unreal-world-builder.md +273 -0
- package/.claude/agents/visionos-spatial-engineer.md +54 -0
- package/.claude/agents/xr-cockpit-interaction-specialist.md +32 -0
- package/.claude/agents/xr-immersive-developer.md +32 -0
- package/.claude/agents/xr-interface-architect.md +32 -0
- package/.claude/agents/zk-steward.md +211 -0
- package/.claude-plugin/marketplace.json +30 -0
- package/.commitlintrc.json +26 -0
- package/.devcontainer/Dockerfile +24 -0
- package/.devcontainer/devcontainer.json +70 -0
- package/.devcontainer/docker-compose.memory.yml +42 -0
- package/.devcontainer/memory-stack/devcontainer.json +84 -0
- package/.devcontainer/post-create.sh +83 -0
- package/.dockerignore +23 -0
- package/.env.act.example +15 -0
- package/.env.example +171 -0
- package/.eslintrc.json +16 -0
- package/.git-blame-ignore-revs +10 -0
- package/.gitattributes +2 -0
- package/.gitguardian.yaml +50 -0
- package/.githooks/pre-commit +6 -0
- package/.pre-commit-config.yaml +30 -0
- package/.release-please-config-xithead.json +22 -0
- package/.release-please-config.json +44 -0
- package/.release-please-manifest-xithead.json +3 -0
- package/.release-please-manifest.json +3 -0
- package/CODE_OF_CONDUCT.md +133 -0
- package/CONTRIBUTING.md +133 -0
- package/Cargo.lock +5557 -0
- package/Cargo.toml +121 -0
- package/Dockerfile +38 -0
- package/Dockerfile.proxy +48 -0
- package/ENTERPRISE.md +5 -0
- package/FEATURES_ADVANCED.md +311 -0
- package/Headroom-2.gif +0 -0
- package/HeadroomDemo-Fast.gif +0 -0
- package/MANIFEST.in +4 -0
- package/MODERNIZATION_GUIDE.md +358 -0
- package/Makefile +140 -0
- package/NOTICE +43 -0
- package/NPM_TOKEN_SETUP.md +81 -0
- package/PR.md +170 -0
- package/PUBLISH.md +77 -0
- package/README.md +58 -137
- package/README_FRONTEND.md +196 -0
- package/REALIGNMENT/00-overview.md +53 -0
- package/REALIGNMENT/01-bug-list.md +421 -0
- package/REALIGNMENT/02-architecture.md +322 -0
- package/REALIGNMENT/03-phase-A-lockdown.md +420 -0
- package/REALIGNMENT/04-phase-B-live-zone.md +470 -0
- package/REALIGNMENT/05-phase-C-rust-proxy.md +329 -0
- package/REALIGNMENT/06-phase-D-bedrock-vertex.md +222 -0
- package/REALIGNMENT/07-phase-E-cache-stabilization.md +338 -0
- package/REALIGNMENT/08-phase-F-auth-mode.md +254 -0
- package/REALIGNMENT/09-phase-G-rtk-observability.md +194 -0
- package/REALIGNMENT/10-phase-H-python-retirement.md +214 -0
- package/REALIGNMENT/11-phase-I-test-infra.md +385 -0
- package/REALIGNMENT/12-decisions-needed.md +196 -0
- package/REALIGNMENT/INDEX.md +82 -0
- package/REFACTORING_CAMPAIGN.md +277 -0
- package/RESTRUCTURE_PLAN.md +100 -0
- package/RUST_DEV.md +361 -0
- package/SECURITY.md +65 -0
- package/SETUP_AUTO_PUBLISH.md +164 -0
- package/TESTING-copilot-subscription.md +153 -0
- package/app/api/config/route.ts +22 -0
- package/app/api/export/route.ts +83 -0
- package/app/api/models/route.ts +12 -0
- package/app/api/stats/route.ts +46 -0
- package/app/dashboard/page.tsx +445 -0
- package/app/globals.css +35 -0
- package/app/installation/page.tsx +293 -0
- package/app/layout.tsx +35 -0
- package/app/master-design-system.css +2076 -0
- package/app/master-design-system.source.css +8 -0
- package/app/page.tsx +187 -0
- package/app/rechtliches/agb/page.tsx +165 -0
- package/app/rechtliches/datenschutz/page.tsx +128 -0
- package/app/rechtliches/impressum/page.tsx +122 -0
- package/app/rechtliches/widerruf/page.tsx +185 -0
- package/bin/xithead.js +264 -0
- package/codecov.yml +34 -0
- package/components/layout/AppShell.tsx +25 -0
- package/components/layout/Footer.tsx +120 -0
- package/components/layout/Header.tsx +145 -0
- package/components/marketing/LegalHighlights.tsx +45 -0
- package/components/marketing/LegalPageShell.tsx +49 -0
- package/components/marketing/LegalPanel.tsx +41 -0
- package/components/marketing/index.ts +3 -0
- package/crates/headroom-core/Cargo.toml +165 -0
- package/crates/headroom-core/benches/auth_mode.rs +74 -0
- package/crates/headroom-core/benches/ccr_store.rs +224 -0
- package/crates/headroom-core/benches/tokenizer.rs +52 -0
- package/crates/headroom-core/config/pipeline.toml +151 -0
- package/crates/headroom-core/proptest-regressions/transforms/tag_protector.txt +9 -0
- package/crates/headroom-core/src/auth_mode.rs +250 -0
- package/crates/headroom-core/src/cache_control.rs +414 -0
- package/crates/headroom-core/src/ccr/backends/in_memory.rs +317 -0
- package/crates/headroom-core/src/ccr/backends/mod.rs +152 -0
- package/crates/headroom-core/src/ccr/backends/redis.rs +146 -0
- package/crates/headroom-core/src/ccr/backends/sqlite.rs +205 -0
- package/crates/headroom-core/src/ccr/mod.rs +119 -0
- package/crates/headroom-core/src/compression_policy.rs +505 -0
- package/crates/headroom-core/src/lib.rs +33 -0
- package/crates/headroom-core/src/relevance/base.rs +139 -0
- package/crates/headroom-core/src/relevance/bm25.rs +385 -0
- package/crates/headroom-core/src/relevance/embedding.rs +354 -0
- package/crates/headroom-core/src/relevance/hybrid.rs +364 -0
- package/crates/headroom-core/src/relevance/mod.rs +64 -0
- package/crates/headroom-core/src/signals/keyword_detector.rs +433 -0
- package/crates/headroom-core/src/signals/line_importance.rs +84 -0
- package/crates/headroom-core/src/signals/mod.rs +61 -0
- package/crates/headroom-core/src/signals/tiered.rs +141 -0
- package/crates/headroom-core/src/tokenizer/estimator.rs +152 -0
- package/crates/headroom-core/src/tokenizer/hf_impl.rs +314 -0
- package/crates/headroom-core/src/tokenizer/mod.rs +56 -0
- package/crates/headroom-core/src/tokenizer/registry.rs +342 -0
- package/crates/headroom-core/src/tokenizer/tiktoken_impl.rs +265 -0
- package/crates/headroom-core/src/transforms/adaptive_sizer.rs +610 -0
- package/crates/headroom-core/src/transforms/anchor_selector.rs +1189 -0
- package/crates/headroom-core/src/transforms/content_detector.rs +769 -0
- package/crates/headroom-core/src/transforms/detection.rs +222 -0
- package/crates/headroom-core/src/transforms/diff_compressor.rs +1685 -0
- package/crates/headroom-core/src/transforms/live_zone.rs +2967 -0
- package/crates/headroom-core/src/transforms/log_compressor.rs +1295 -0
- package/crates/headroom-core/src/transforms/magika_detector.rs +395 -0
- package/crates/headroom-core/src/transforms/mod.rs +64 -0
- package/crates/headroom-core/src/transforms/pipeline/config.rs +331 -0
- package/crates/headroom-core/src/transforms/pipeline/mod.rs +103 -0
- package/crates/headroom-core/src/transforms/pipeline/offloads/diff_noise.rs +524 -0
- package/crates/headroom-core/src/transforms/pipeline/offloads/diff_offload.rs +279 -0
- package/crates/headroom-core/src/transforms/pipeline/offloads/json_offload.rs +400 -0
- package/crates/headroom-core/src/transforms/pipeline/offloads/log_offload.rs +284 -0
- package/crates/headroom-core/src/transforms/pipeline/offloads/mod.rs +34 -0
- package/crates/headroom-core/src/transforms/pipeline/offloads/search_offload.rs +331 -0
- package/crates/headroom-core/src/transforms/pipeline/orchestrator.rs +848 -0
- package/crates/headroom-core/src/transforms/pipeline/reformats/json_minifier.rs +175 -0
- package/crates/headroom-core/src/transforms/pipeline/reformats/log_template.rs +545 -0
- package/crates/headroom-core/src/transforms/pipeline/reformats/mod.rs +12 -0
- package/crates/headroom-core/src/transforms/pipeline/traits.rs +338 -0
- package/crates/headroom-core/src/transforms/recommendations.rs +329 -0
- package/crates/headroom-core/src/transforms/safety.rs +215 -0
- package/crates/headroom-core/src/transforms/search_compressor.rs +902 -0
- package/crates/headroom-core/src/transforms/smart_crusher/analyzer.rs +1262 -0
- package/crates/headroom-core/src/transforms/smart_crusher/anchors.rs +407 -0
- package/crates/headroom-core/src/transforms/smart_crusher/builder.rs +263 -0
- package/crates/headroom-core/src/transforms/smart_crusher/classifier.rs +214 -0
- package/crates/headroom-core/src/transforms/smart_crusher/compaction/classifier.rs +283 -0
- package/crates/headroom-core/src/transforms/smart_crusher/compaction/compactor.rs +733 -0
- package/crates/headroom-core/src/transforms/smart_crusher/compaction/formatter.rs +961 -0
- package/crates/headroom-core/src/transforms/smart_crusher/compaction/ir.rs +252 -0
- package/crates/headroom-core/src/transforms/smart_crusher/compaction/mod.rs +127 -0
- package/crates/headroom-core/src/transforms/smart_crusher/compaction/walker.rs +397 -0
- package/crates/headroom-core/src/transforms/smart_crusher/config.rs +173 -0
- package/crates/headroom-core/src/transforms/smart_crusher/constraints.rs +159 -0
- package/crates/headroom-core/src/transforms/smart_crusher/crusher.rs +1665 -0
- package/crates/headroom-core/src/transforms/smart_crusher/crushers.rs +865 -0
- package/crates/headroom-core/src/transforms/smart_crusher/error_keywords.rs +77 -0
- package/crates/headroom-core/src/transforms/smart_crusher/field_detect.rs +393 -0
- package/crates/headroom-core/src/transforms/smart_crusher/hashing.rs +72 -0
- package/crates/headroom-core/src/transforms/smart_crusher/mod.rs +81 -0
- package/crates/headroom-core/src/transforms/smart_crusher/observer.rs +66 -0
- package/crates/headroom-core/src/transforms/smart_crusher/orchestration.rs +459 -0
- package/crates/headroom-core/src/transforms/smart_crusher/outliers.rs +507 -0
- package/crates/headroom-core/src/transforms/smart_crusher/planning.rs +904 -0
- package/crates/headroom-core/src/transforms/smart_crusher/statistics.rs +510 -0
- package/crates/headroom-core/src/transforms/smart_crusher/stats_math.rs +287 -0
- package/crates/headroom-core/src/transforms/smart_crusher/traits.rs +203 -0
- package/crates/headroom-core/src/transforms/smart_crusher/types.rs +263 -0
- package/crates/headroom-core/src/transforms/tag_protector.rs +1272 -0
- package/crates/headroom-core/src/transforms/unidiff_detector.rs +232 -0
- package/crates/headroom-parity/Cargo.toml +26 -0
- package/crates/headroom-parity/examples/diff_fixture.rs +66 -0
- package/crates/headroom-parity/src/bin/parity_run.rs +78 -0
- package/crates/headroom-parity/src/lib.rs +613 -0
- package/crates/headroom-proxy/Cargo.toml +119 -0
- package/crates/headroom-proxy/data/model_prices_and_context_window.json +39767 -0
- package/crates/headroom-proxy/src/bedrock/auth_mode_layer.rs +192 -0
- package/crates/headroom-proxy/src/bedrock/envelope.rs +239 -0
- package/crates/headroom-proxy/src/bedrock/eventstream.rs +769 -0
- package/crates/headroom-proxy/src/bedrock/eventstream_to_sse.rs +459 -0
- package/crates/headroom-proxy/src/bedrock/invoke.rs +735 -0
- package/crates/headroom-proxy/src/bedrock/invoke_streaming.rs +1098 -0
- package/crates/headroom-proxy/src/bedrock/mod.rs +66 -0
- package/crates/headroom-proxy/src/bedrock/sigv4.rs +324 -0
- package/crates/headroom-proxy/src/bedrock/vendor.rs +82 -0
- package/crates/headroom-proxy/src/cache_stabilization/anthropic_cache_control.rs +698 -0
- package/crates/headroom-proxy/src/cache_stabilization/drift_detector.rs +683 -0
- package/crates/headroom-proxy/src/cache_stabilization/mod.rs +61 -0
- package/crates/headroom-proxy/src/cache_stabilization/openai_cache_key.rs +600 -0
- package/crates/headroom-proxy/src/cache_stabilization/tool_def_normalize.rs +514 -0
- package/crates/headroom-proxy/src/cache_stabilization/volatile_detector.rs +720 -0
- package/crates/headroom-proxy/src/compression/anthropic.rs +92 -0
- package/crates/headroom-proxy/src/compression/live_zone_anthropic.rs +1311 -0
- package/crates/headroom-proxy/src/compression/live_zone_openai.rs +660 -0
- package/crates/headroom-proxy/src/compression/live_zone_responses.rs +678 -0
- package/crates/headroom-proxy/src/compression/mod.rs +127 -0
- package/crates/headroom-proxy/src/compression/model_limits.rs +211 -0
- package/crates/headroom-proxy/src/config.rs +646 -0
- package/crates/headroom-proxy/src/error.rs +69 -0
- package/crates/headroom-proxy/src/handlers/chat_completions.rs +100 -0
- package/crates/headroom-proxy/src/handlers/conversations.rs +241 -0
- package/crates/headroom-proxy/src/handlers/mod.rs +13 -0
- package/crates/headroom-proxy/src/handlers/responses.rs +256 -0
- package/crates/headroom-proxy/src/headers.rs +288 -0
- package/crates/headroom-proxy/src/health.rs +41 -0
- package/crates/headroom-proxy/src/lib.rs +21 -0
- package/crates/headroom-proxy/src/main.rs +141 -0
- package/crates/headroom-proxy/src/observability/cache_hit_rate.rs +340 -0
- package/crates/headroom-proxy/src/observability/compression_ratio.rs +158 -0
- package/crates/headroom-proxy/src/observability/metric_names.rs +174 -0
- package/crates/headroom-proxy/src/observability/mod.rs +68 -0
- package/crates/headroom-proxy/src/observability/prometheus.rs +394 -0
- package/crates/headroom-proxy/src/observability/proxy_metrics.rs +368 -0
- package/crates/headroom-proxy/src/proxy.rs +1619 -0
- package/crates/headroom-proxy/src/responses_items.rs +565 -0
- package/crates/headroom-proxy/src/sse/anthropic.rs +415 -0
- package/crates/headroom-proxy/src/sse/framing.rs +366 -0
- package/crates/headroom-proxy/src/sse/mod.rs +40 -0
- package/crates/headroom-proxy/src/sse/openai_chat.rs +231 -0
- package/crates/headroom-proxy/src/sse/openai_responses.rs +458 -0
- package/crates/headroom-proxy/src/vertex/adc.rs +315 -0
- package/crates/headroom-proxy/src/vertex/envelope.rs +167 -0
- package/crates/headroom-proxy/src/vertex/mod.rs +277 -0
- package/crates/headroom-proxy/src/vertex/raw_predict.rs +490 -0
- package/crates/headroom-proxy/src/vertex/stream_raw_predict.rs +49 -0
- package/crates/headroom-proxy/src/websocket.rs +246 -0
- package/crates/headroom-py/Cargo.toml +37 -0
- package/crates/headroom-py/build.rs +77 -0
- package/crates/headroom-py/glibc_compat.c +170 -0
- package/crates/headroom-py/src/lib.rs +1655 -0
- package/deny.toml +32 -0
- package/design-system/base.css +127 -0
- package/design-system/components.css +1943 -0
- package/design-system/fonts.css +17 -0
- package/design-system/tailwind-preset.js +95 -0
- package/design-system/tokens.css +48 -0
- package/design-system/utilities.css +53 -0
- package/docker/differential-network-capture/Dockerfile.runner +14 -0
- package/docker/differential-network-capture/docker-compose.yml +142 -0
- package/docker/differential-network-capture/run-claude-lane.sh +24 -0
- package/docker/docker-compose.native.yml +44 -0
- package/docker-bake.hcl +88 -0
- package/docker-compose.yml +80 -0
- package/e2e/_lib/make_shim.ps1 +24 -0
- package/e2e/_lib/make_shim.sh +28 -0
- package/e2e/docker-native-install.sh +120 -0
- package/e2e/init/Dockerfile +79 -0
- package/e2e/wrap/Dockerfile +115 -0
- package/examples/07-context-compression.ipynb +279 -0
- package/examples/deployment/macos-launchagent/com.headroom.proxy.plist.template +65 -0
- package/examples/deployment/macos-launchagent/install.sh +227 -0
- package/examples/deployment/macos-launchagent/shell-integration.sh +74 -0
- package/examples/deployment/macos-launchagent/uninstall.sh +145 -0
- package/headroom-savings.png +0 -0
- package/headroom_learn.gif +0 -0
- package/lib/config/index.ts +12 -0
- package/lib/hooks/index.ts +4 -0
- package/lib/hooks/use-mounted.ts +13 -0
- package/lib/hooks/use-session.ts +34 -0
- package/lib/hooks/use-settings.ts +49 -0
- package/lib/utils/index.ts +1 -0
- package/lib/utils/settings.ts +162 -0
- package/lib/utils/utils.ts +6 -0
- package/llms.txt +67 -0
- package/middleware.ts +26 -0
- package/mkdocs.yml +136 -0
- package/next-env.d.ts +7 -0
- package/next.config.ts +12 -0
- package/npm/bin/xithead.js +2 -2
- package/npm/bin/xtream-head.js +366 -0
- package/npm/index.js +18 -91
- package/npm/package.json +55 -0
- package/package.json +47 -42
- package/plugins/headroom-agent-hooks/.claude-plugin/plugin.json +17 -0
- package/plugins/headroom-agent-hooks/hooks/hooks.json +29 -0
- package/plugins/headroom-oauth2/LICENSE +3 -0
- package/plugins/headroom-oauth2/SPEC.md +115 -0
- package/plugins/headroom-oauth2/pyproject.toml +48 -0
- package/plugins/hermes/headroom_retrieve/plugin.yaml +6 -0
- package/plugins/openclaw/hook-shim/HOOK.md +21 -0
- package/plugins/openclaw/hook-shim/handler.js +3 -0
- package/plugins/openclaw/openclaw.plugin.json +98 -0
- package/plugins/openclaw/package-lock.json +2692 -0
- package/plugins/openclaw/package.json +54 -0
- package/plugins/openclaw/prepare-dist.mjs +56 -0
- package/plugins/openclaw/src/convert.ts +418 -0
- package/plugins/openclaw/src/engine.ts +284 -0
- package/plugins/openclaw/src/gateway-config.ts +143 -0
- package/plugins/openclaw/src/index.ts +11 -0
- package/plugins/openclaw/src/plugin/index.ts +98 -0
- package/plugins/openclaw/src/proxy-manager.ts +477 -0
- package/plugins/openclaw/src/tools/headroom-retrieve.ts +76 -0
- package/plugins/openclaw/test/convert.test.ts +96 -0
- package/plugins/openclaw/test/engine-normalization.test.ts +21 -0
- package/plugins/openclaw/test/engine.test.ts +101 -0
- package/plugins/openclaw/test/gateway-config.test.ts +278 -0
- package/plugins/openclaw/test/plugin-runtime-routing.test.ts +141 -0
- package/plugins/openclaw/test/proxy-manager.test.ts +343 -0
- package/plugins/openclaw/tsup.config.ts +9 -0
- package/postcss.config.mjs +7 -0
- package/pyproject.toml +438 -0
- package/rust-toolchain.toml +16 -0
- package/sdk/typescript/examples/basic-compress.ts +64 -0
- package/sdk/typescript/examples/ccr-retrieve.ts +80 -0
- package/sdk/typescript/examples/hooks-custom-compression.ts +117 -0
- package/sdk/typescript/examples/middleware-composition.ts +59 -0
- package/sdk/typescript/examples/multi-provider.ts +53 -0
- package/sdk/typescript/examples/openai-anthropic-adapters.ts +73 -0
- package/sdk/typescript/examples/shared-context-multi-agent.ts +74 -0
- package/sdk/typescript/examples/simulation-dry-run.ts +80 -0
- package/sdk/typescript/examples/streaming-chat.ts +71 -0
- package/sdk/typescript/examples/structured-output.ts +99 -0
- package/sdk/typescript/examples/tool-calling-agent.ts +86 -0
- package/sdk/typescript/examples/with-headroom-vercel.ts +57 -0
- package/sdk/typescript/package-lock.json +3437 -0
- package/sdk/typescript/package.json +108 -0
- package/sdk/typescript/src/adapters/anthropic.ts +218 -0
- package/sdk/typescript/src/adapters/gemini.ts +83 -0
- package/sdk/typescript/src/adapters/openai.ts +77 -0
- package/sdk/typescript/src/adapters/vercel-ai.ts +131 -0
- package/sdk/typescript/src/client.ts +632 -0
- package/sdk/typescript/src/compress.ts +87 -0
- package/sdk/typescript/src/errors.ts +114 -0
- package/sdk/typescript/src/hooks.ts +113 -0
- package/sdk/typescript/src/index.ts +140 -0
- package/sdk/typescript/src/paths.ts +272 -0
- package/sdk/typescript/src/shared-context.ts +188 -0
- package/sdk/typescript/src/simulate.ts +52 -0
- package/sdk/typescript/src/types/config.ts +183 -0
- package/sdk/typescript/src/types/models.ts +314 -0
- package/sdk/typescript/src/types.ts +139 -0
- package/sdk/typescript/src/utils/case.ts +34 -0
- package/sdk/typescript/src/utils/format.ts +480 -0
- package/sdk/typescript/src/utils/stream.ts +50 -0
- package/sdk/typescript/test/adapters/anthropic.test.ts +217 -0
- package/sdk/typescript/test/adapters/gemini.test.ts +131 -0
- package/sdk/typescript/test/adapters/openai.test.ts +174 -0
- package/sdk/typescript/test/adapters/vercel-ai.test.ts +306 -0
- package/sdk/typescript/test/client-expanded.test.ts +333 -0
- package/sdk/typescript/test/client.test.ts +266 -0
- package/sdk/typescript/test/compress.test.ts +118 -0
- package/sdk/typescript/test/config.test.ts +275 -0
- package/sdk/typescript/test/errors.test.ts +169 -0
- package/sdk/typescript/test/format-integration.test.ts +252 -0
- package/sdk/typescript/test/hooks.test.ts +265 -0
- package/sdk/typescript/test/integration.test.ts +249 -0
- package/sdk/typescript/test/mdx-examples.test.ts +253 -0
- package/sdk/typescript/test/models.test.ts +318 -0
- package/sdk/typescript/test/parity.test.ts +337 -0
- package/sdk/typescript/test/paths.test.ts +447 -0
- package/sdk/typescript/test/shared-context.test.ts +245 -0
- package/sdk/typescript/test/simulate.test.ts +84 -0
- package/sdk/typescript/test/types.test.ts +244 -0
- package/sdk/typescript/test/utils/case.test.ts +129 -0
- package/sdk/typescript/test/utils/detect-format.test.ts +133 -0
- package/sdk/typescript/test/utils/format.test.ts +434 -0
- package/sdk/typescript/test/vercel-ai-e2e.test.ts +179 -0
- package/sdk/typescript/tsup.config.ts +17 -0
- package/tailwind.master.config.cjs +47 -0
- package/types/api.ts +12 -0
- package/types/config.ts +9 -0
- package/types/index.ts +2 -0
- package/uv.lock +6951 -0
- package/xithead.egg-info/PKG-INFO +22 -0
- package/xithead.egg-info/SOURCES.txt +16 -0
- package/xithead.egg-info/dependency_links.txt +1 -0
- package/xithead.egg-info/entry_points.txt +2 -0
- package/xithead.egg-info/requires.txt +10 -0
- package/xithead.egg-info/top_level.txt +2 -0
- package/xithead.ps1 +136 -0
- package/xithead.sh +129 -0
- package/xithead_dist/pyproject.toml +39 -0
- package/xithead_dist/xithead.egg-info/PKG-INFO +22 -0
- package/xithead_dist/xithead.egg-info/SOURCES.txt +16 -0
- package/xithead_dist/xithead.egg-info/dependency_links.txt +1 -0
- package/xithead_dist/xithead.egg-info/entry_points.txt +3 -0
- package/xithead_dist/xithead.egg-info/requires.txt +10 -0
- package/xithead_dist/xithead.egg-info/top_level.txt +2 -0
|
@@ -0,0 +1,342 @@
|
|
|
1
|
+
//! Model-name → tokenizer dispatch.
|
|
2
|
+
//!
|
|
3
|
+
//! Mirrors `MODEL_PATTERNS` in `headroom/tokenizers/registry.py`. Three
|
|
4
|
+
//! backends in priority order:
|
|
5
|
+
//!
|
|
6
|
+
//! 1. **HuggingFace** — anything the caller has registered via
|
|
7
|
+
//! [`register_hf`] for a given model-name prefix. Real BPE/Unigram/
|
|
8
|
+
//! WordPiece counts. This is opt-in: tokenizer.json files aren't bundled,
|
|
9
|
+
//! so nothing routes here until the embedding application calls
|
|
10
|
+
//! [`register_hf`] at startup. Wins over the rules below when registered.
|
|
11
|
+
//! 2. **Tiktoken** — OpenAI / o-series via `tiktoken-rs`. Byte-identical to
|
|
12
|
+
//! Python `tiktoken`.
|
|
13
|
+
//! 3. **Estimation** — `chars / cpt` fallback for Anthropic Claude (3.5),
|
|
14
|
+
//! Gemini / Cohere / Command without an HF registration (4.0), and
|
|
15
|
+
//! everything else (4.0).
|
|
16
|
+
|
|
17
|
+
use std::collections::HashMap;
|
|
18
|
+
use std::sync::{OnceLock, RwLock};
|
|
19
|
+
|
|
20
|
+
use super::{EstimatingCounter, HfTokenizer, HfTokenizerError, TiktokenCounter, Tokenizer};
|
|
21
|
+
|
|
22
|
+
/// Which family of tokenizer was selected for a model.
|
|
23
|
+
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
|
|
24
|
+
pub enum Backend {
|
|
25
|
+
/// Real BPE via `tiktoken-rs`. Byte-identical to Python `tiktoken`.
|
|
26
|
+
Tiktoken,
|
|
27
|
+
/// HuggingFace `tokenizers` crate (`tokenizer.json` loaded by caller).
|
|
28
|
+
HuggingFace,
|
|
29
|
+
/// Character-density estimation (chars/token formula).
|
|
30
|
+
Estimation,
|
|
31
|
+
}
|
|
32
|
+
|
|
33
|
+
/// Pick a backend purely from the model name, ignoring runtime registrations.
|
|
34
|
+
///
|
|
35
|
+
/// Patterns and ordering match `headroom.tokenizers.registry.MODEL_PATTERNS`
|
|
36
|
+
/// for the families this stage supports. Anything outside the OpenAI BPE
|
|
37
|
+
/// family lands in `Estimation` here — even if [`register_hf`] would route
|
|
38
|
+
/// it to `HuggingFace` at runtime. Use [`get_tokenizer`] for the real
|
|
39
|
+
/// dispatch.
|
|
40
|
+
pub fn detect_backend(model: &str) -> Backend {
|
|
41
|
+
let m = model.to_ascii_lowercase();
|
|
42
|
+
|
|
43
|
+
// OpenAI BPE-tokenized families (gpt-3.5/4/4o + o1/o3 reasoning + embeddings + legacy davinci/curie/babbage/ada + code-).
|
|
44
|
+
if m.starts_with("gpt-4o")
|
|
45
|
+
|| m.starts_with("gpt-4")
|
|
46
|
+
|| m.starts_with("gpt-3.5")
|
|
47
|
+
|| m.starts_with("o1")
|
|
48
|
+
|| m.starts_with("o3")
|
|
49
|
+
|| m.starts_with("text-embedding")
|
|
50
|
+
|| m.starts_with("text-davinci")
|
|
51
|
+
|| m.starts_with("davinci")
|
|
52
|
+
|| m.starts_with("curie")
|
|
53
|
+
|| m.starts_with("babbage")
|
|
54
|
+
|| m.starts_with("ada")
|
|
55
|
+
|| m.starts_with("code-")
|
|
56
|
+
{
|
|
57
|
+
return Backend::Tiktoken;
|
|
58
|
+
}
|
|
59
|
+
|
|
60
|
+
Backend::Estimation
|
|
61
|
+
}
|
|
62
|
+
|
|
63
|
+
/// Return a tokenizer for `model`. Resolution order:
|
|
64
|
+
/// 1. HuggingFace tokenizers registered via [`register_hf`] (longest matching
|
|
65
|
+
/// prefix wins).
|
|
66
|
+
/// 2. Tiktoken for OpenAI / o-series families.
|
|
67
|
+
/// 3. Estimation, with density calibrated per family (Claude → 3.5, Gemini /
|
|
68
|
+
/// Cohere / Command → 4.0, otherwise 4.0).
|
|
69
|
+
pub fn get_tokenizer(model: &str) -> Box<dyn Tokenizer> {
|
|
70
|
+
if let Some(hf) = lookup_hf(model) {
|
|
71
|
+
return Box::new(hf);
|
|
72
|
+
}
|
|
73
|
+
match detect_backend(model) {
|
|
74
|
+
Backend::Tiktoken => match TiktokenCounter::for_model(model) {
|
|
75
|
+
Ok(t) => Box::new(t),
|
|
76
|
+
Err(_) => Box::new(default_estimator_for(model)),
|
|
77
|
+
},
|
|
78
|
+
// Backend::HuggingFace from detect_backend is unreachable — only
|
|
79
|
+
// runtime registrations produce HF, and we already checked above.
|
|
80
|
+
Backend::HuggingFace | Backend::Estimation => Box::new(default_estimator_for(model)),
|
|
81
|
+
}
|
|
82
|
+
}
|
|
83
|
+
|
|
84
|
+
fn default_estimator_for(model: &str) -> EstimatingCounter {
|
|
85
|
+
let m = model.to_ascii_lowercase();
|
|
86
|
+
if m.starts_with("claude-") {
|
|
87
|
+
EstimatingCounter::new(3.5)
|
|
88
|
+
} else if m.starts_with("gemini") || m.starts_with("palm") || m.starts_with("command") {
|
|
89
|
+
EstimatingCounter::new(4.0)
|
|
90
|
+
} else {
|
|
91
|
+
EstimatingCounter::default()
|
|
92
|
+
}
|
|
93
|
+
}
|
|
94
|
+
|
|
95
|
+
// ---- HuggingFace runtime registry --------------------------------------
|
|
96
|
+
//
|
|
97
|
+
// A process-global table mapping a lowercased model-name *prefix* to a loaded
|
|
98
|
+
// `HfTokenizer`. The embedding application owns this — there's no autoloader
|
|
99
|
+
// in core because we don't want to bundle tokenizer.json files (multi-MB) or
|
|
100
|
+
// pull in HF Hub networking here.
|
|
101
|
+
|
|
102
|
+
fn hf_table() -> &'static RwLock<HashMap<String, HfTokenizer>> {
|
|
103
|
+
static TABLE: OnceLock<RwLock<HashMap<String, HfTokenizer>>> = OnceLock::new();
|
|
104
|
+
TABLE.get_or_init(|| RwLock::new(HashMap::new()))
|
|
105
|
+
}
|
|
106
|
+
|
|
107
|
+
/// Register `tokenizer` to handle every model whose lowercased name starts
|
|
108
|
+
/// with `prefix`. Multiple prefixes can coexist; on lookup the longest
|
|
109
|
+
/// matching prefix wins, so registering `"command-r-plus"` overrides a more
|
|
110
|
+
/// general `"command-"` registration for that one model.
|
|
111
|
+
///
|
|
112
|
+
/// Calling this with the same `prefix` twice replaces the previous tokenizer.
|
|
113
|
+
pub fn register_hf(prefix: impl Into<String>, tokenizer: HfTokenizer) {
|
|
114
|
+
let key = prefix.into().to_ascii_lowercase();
|
|
115
|
+
hf_table()
|
|
116
|
+
.write()
|
|
117
|
+
.expect("hf registry poisoned")
|
|
118
|
+
.insert(key, tokenizer);
|
|
119
|
+
}
|
|
120
|
+
|
|
121
|
+
/// Drop every HuggingFace registration. Intended for tests; production code
|
|
122
|
+
/// should register once at startup.
|
|
123
|
+
pub fn clear_hf_registrations() {
|
|
124
|
+
hf_table().write().expect("hf registry poisoned").clear();
|
|
125
|
+
}
|
|
126
|
+
|
|
127
|
+
/// Convenience: download `tokenizer.json` for `repo` from the HuggingFace
|
|
128
|
+
/// Hub and register it under `prefix`. One-line glue around
|
|
129
|
+
/// [`HfTokenizer::from_pretrained`] + [`register_hf`].
|
|
130
|
+
///
|
|
131
|
+
/// Useful for proxy startup code that wants real tokenizers for the major
|
|
132
|
+
/// non-OpenAI families. Each call is independent — failure for one model
|
|
133
|
+
/// (e.g. a gated Llama repo without `HF_TOKEN`) does not affect others.
|
|
134
|
+
///
|
|
135
|
+
/// ```no_run
|
|
136
|
+
/// use headroom_core::tokenizer::try_register_hf;
|
|
137
|
+
/// let _ = try_register_hf("command-", "CohereForAI/c4ai-command-r-v01");
|
|
138
|
+
/// let _ = try_register_hf("mistral-", "mistralai/Mistral-7B-v0.1");
|
|
139
|
+
/// ```
|
|
140
|
+
pub fn try_register_hf(prefix: &str, repo: &str) -> Result<(), HfTokenizerError> {
|
|
141
|
+
let t = HfTokenizer::from_pretrained(repo)?;
|
|
142
|
+
register_hf(prefix, t);
|
|
143
|
+
Ok(())
|
|
144
|
+
}
|
|
145
|
+
|
|
146
|
+
fn lookup_hf(model: &str) -> Option<HfTokenizer> {
|
|
147
|
+
let m = model.to_ascii_lowercase();
|
|
148
|
+
let table = hf_table().read().expect("hf registry poisoned");
|
|
149
|
+
// Longest prefix wins: a `command-r-plus` registration must beat
|
|
150
|
+
// `command-` for that model, regardless of insertion order.
|
|
151
|
+
table
|
|
152
|
+
.iter()
|
|
153
|
+
.filter(|(prefix, _)| m.starts_with(prefix.as_str()))
|
|
154
|
+
.max_by_key(|(prefix, _)| prefix.len())
|
|
155
|
+
.map(|(_, t)| t.clone())
|
|
156
|
+
}
|
|
157
|
+
|
|
158
|
+
#[cfg(test)]
|
|
159
|
+
mod tests {
|
|
160
|
+
use super::*;
|
|
161
|
+
|
|
162
|
+
/// Same minimal tokenizer.json used by hf_impl tests; inlined here so we
|
|
163
|
+
/// can register it without depending on a fixture file.
|
|
164
|
+
const TINY_TOKENIZER_JSON: &str = r#"{
|
|
165
|
+
"version": "1.0",
|
|
166
|
+
"truncation": null,
|
|
167
|
+
"padding": null,
|
|
168
|
+
"added_tokens": [],
|
|
169
|
+
"normalizer": null,
|
|
170
|
+
"pre_tokenizer": {"type": "Whitespace"},
|
|
171
|
+
"post_processor": null,
|
|
172
|
+
"decoder": null,
|
|
173
|
+
"model": {
|
|
174
|
+
"type": "WordLevel",
|
|
175
|
+
"vocab": {"hello": 0, "world": 1, "[UNK]": 2},
|
|
176
|
+
"unk_token": "[UNK]"
|
|
177
|
+
}
|
|
178
|
+
}"#;
|
|
179
|
+
|
|
180
|
+
fn tiny(name: &str) -> HfTokenizer {
|
|
181
|
+
HfTokenizer::from_bytes(name, TINY_TOKENIZER_JSON.as_bytes()).unwrap()
|
|
182
|
+
}
|
|
183
|
+
|
|
184
|
+
/// Tests share a process-global table. `cargo test` runs them in parallel
|
|
185
|
+
/// by default, so without serialization one test's `clear` would wipe
|
|
186
|
+
/// another test's registrations mid-assertion. We serialize the tests
|
|
187
|
+
/// that touch the registry behind a single mutex.
|
|
188
|
+
static REGISTRY_LOCK: std::sync::Mutex<()> = std::sync::Mutex::new(());
|
|
189
|
+
|
|
190
|
+
struct RegistryGuard<'a> {
|
|
191
|
+
_g: std::sync::MutexGuard<'a, ()>,
|
|
192
|
+
}
|
|
193
|
+
impl<'a> RegistryGuard<'a> {
|
|
194
|
+
fn acquire() -> Self {
|
|
195
|
+
// Recover from a poisoned lock — a panic in one test should not
|
|
196
|
+
// break every subsequent test in the file.
|
|
197
|
+
let g = REGISTRY_LOCK.lock().unwrap_or_else(|e| e.into_inner());
|
|
198
|
+
clear_hf_registrations();
|
|
199
|
+
Self { _g: g }
|
|
200
|
+
}
|
|
201
|
+
}
|
|
202
|
+
impl<'a> Drop for RegistryGuard<'a> {
|
|
203
|
+
fn drop(&mut self) {
|
|
204
|
+
clear_hf_registrations();
|
|
205
|
+
}
|
|
206
|
+
}
|
|
207
|
+
|
|
208
|
+
#[test]
|
|
209
|
+
fn openai_models_pick_tiktoken() {
|
|
210
|
+
for m in [
|
|
211
|
+
"gpt-4o",
|
|
212
|
+
"gpt-4o-mini",
|
|
213
|
+
"gpt-4",
|
|
214
|
+
"gpt-4-turbo",
|
|
215
|
+
"gpt-3.5-turbo",
|
|
216
|
+
"o1-preview",
|
|
217
|
+
"o3-mini",
|
|
218
|
+
"text-embedding-3-small",
|
|
219
|
+
"text-davinci-003",
|
|
220
|
+
"davinci",
|
|
221
|
+
"babbage-002",
|
|
222
|
+
"code-davinci-002",
|
|
223
|
+
] {
|
|
224
|
+
assert_eq!(detect_backend(m), Backend::Tiktoken, "{m}");
|
|
225
|
+
}
|
|
226
|
+
}
|
|
227
|
+
|
|
228
|
+
#[test]
|
|
229
|
+
fn non_openai_models_fall_through_to_estimation() {
|
|
230
|
+
for m in [
|
|
231
|
+
"claude-haiku-4-5-20251001",
|
|
232
|
+
"claude-3-opus",
|
|
233
|
+
"gemini-1.5-pro",
|
|
234
|
+
"command-r-plus",
|
|
235
|
+
"llama-3-70b",
|
|
236
|
+
"mistral-large",
|
|
237
|
+
"qwen-72b",
|
|
238
|
+
"made-up-model-name",
|
|
239
|
+
] {
|
|
240
|
+
assert_eq!(detect_backend(m), Backend::Estimation, "{m}");
|
|
241
|
+
}
|
|
242
|
+
}
|
|
243
|
+
|
|
244
|
+
#[test]
|
|
245
|
+
fn case_insensitive() {
|
|
246
|
+
assert_eq!(detect_backend("GPT-4o"), Backend::Tiktoken);
|
|
247
|
+
assert_eq!(detect_backend("Claude-haiku"), Backend::Estimation);
|
|
248
|
+
}
|
|
249
|
+
|
|
250
|
+
#[test]
|
|
251
|
+
fn estimator_density_per_family() {
|
|
252
|
+
let _g = RegistryGuard::acquire();
|
|
253
|
+
// Round-trip through the public dispatch and check that the chosen
|
|
254
|
+
// estimator behaves with the right density. We can't introspect the
|
|
255
|
+
// trait object's chars_per_token directly, so we use a known-length
|
|
256
|
+
// string and back-compute.
|
|
257
|
+
let claude = get_tokenizer("claude-3-opus");
|
|
258
|
+
// 3.5 chars/token: 35 chars -> 10 tokens.
|
|
259
|
+
assert_eq!(claude.count_text(&"a".repeat(35)), 10);
|
|
260
|
+
|
|
261
|
+
let gemini = get_tokenizer("gemini-1.5-pro");
|
|
262
|
+
// 4.0 chars/token: 40 chars -> 10 tokens.
|
|
263
|
+
assert_eq!(gemini.count_text(&"a".repeat(40)), 10);
|
|
264
|
+
}
|
|
265
|
+
|
|
266
|
+
#[test]
|
|
267
|
+
fn registered_hf_wins_over_estimator() {
|
|
268
|
+
let _g = RegistryGuard::acquire();
|
|
269
|
+
register_hf("command-", tiny("cohere"));
|
|
270
|
+
let t = get_tokenizer("command-r-plus");
|
|
271
|
+
// Whitespace tokenization → 2 tokens, *not* the chars/4 estimator.
|
|
272
|
+
assert_eq!(t.count_text("hello world"), 2);
|
|
273
|
+
assert_eq!(t.backend(), Backend::HuggingFace);
|
|
274
|
+
}
|
|
275
|
+
|
|
276
|
+
#[test]
|
|
277
|
+
fn registered_hf_does_not_override_tiktoken() {
|
|
278
|
+
// A user accidentally registers an HF tokenizer with the prefix
|
|
279
|
+
// `gpt-` — should the OpenAI route still win? Per docstring: HF
|
|
280
|
+
// always wins when registered. We document this behavior so the
|
|
281
|
+
// test pins it down: registration is a deliberate override.
|
|
282
|
+
let _g = RegistryGuard::acquire();
|
|
283
|
+
register_hf("gpt-4o", tiny("oops"));
|
|
284
|
+
let t = get_tokenizer("gpt-4o-mini");
|
|
285
|
+
assert_eq!(t.backend(), Backend::HuggingFace);
|
|
286
|
+
}
|
|
287
|
+
|
|
288
|
+
#[test]
|
|
289
|
+
fn longest_prefix_wins() {
|
|
290
|
+
let _g = RegistryGuard::acquire();
|
|
291
|
+
register_hf("command-", tiny("general"));
|
|
292
|
+
register_hf("command-r-plus", tiny("specific"));
|
|
293
|
+
// `command-r-plus` should pick the more specific one.
|
|
294
|
+
let t = get_tokenizer("command-r-plus");
|
|
295
|
+
assert_eq!(t.backend(), Backend::HuggingFace);
|
|
296
|
+
// Counting the same input via both should give identical counts since
|
|
297
|
+
// we used the same tokenizer.json — but the *path* through the
|
|
298
|
+
// registry is different. This locks in that the longer prefix is
|
|
299
|
+
// selected, not the shorter one.
|
|
300
|
+
let count_specific = t.count_text("hello world hello");
|
|
301
|
+
assert_eq!(count_specific, 3);
|
|
302
|
+
|
|
303
|
+
// A model only matched by the shorter prefix still works.
|
|
304
|
+
let t2 = get_tokenizer("command-light");
|
|
305
|
+
assert_eq!(t2.backend(), Backend::HuggingFace);
|
|
306
|
+
}
|
|
307
|
+
|
|
308
|
+
#[test]
|
|
309
|
+
fn case_insensitive_registration() {
|
|
310
|
+
let _g = RegistryGuard::acquire();
|
|
311
|
+
register_hf("Command-", tiny("cohere"));
|
|
312
|
+
let t = get_tokenizer("COMMAND-R-PLUS");
|
|
313
|
+
assert_eq!(t.backend(), Backend::HuggingFace);
|
|
314
|
+
}
|
|
315
|
+
|
|
316
|
+
#[test]
|
|
317
|
+
fn clear_resets_state() {
|
|
318
|
+
let _g = RegistryGuard::acquire();
|
|
319
|
+
register_hf("command-", tiny("cohere"));
|
|
320
|
+
clear_hf_registrations();
|
|
321
|
+
let t = get_tokenizer("command-r-plus");
|
|
322
|
+
// Without an HF registration, falls back to estimator (4.0 cpt).
|
|
323
|
+
assert_eq!(t.backend(), Backend::Estimation);
|
|
324
|
+
}
|
|
325
|
+
|
|
326
|
+
#[test]
|
|
327
|
+
fn unrelated_models_still_estimate() {
|
|
328
|
+
let _g = RegistryGuard::acquire();
|
|
329
|
+
register_hf("command-", tiny("cohere"));
|
|
330
|
+
let t = get_tokenizer("claude-3-opus");
|
|
331
|
+
assert_eq!(t.backend(), Backend::Estimation);
|
|
332
|
+
}
|
|
333
|
+
|
|
334
|
+
#[test]
|
|
335
|
+
fn detect_backend_ignores_runtime_registrations() {
|
|
336
|
+
let _g = RegistryGuard::acquire();
|
|
337
|
+
register_hf("command-", tiny("cohere"));
|
|
338
|
+
// detect_backend is a pure function of the model name; runtime
|
|
339
|
+
// registrations don't show up here.
|
|
340
|
+
assert_eq!(detect_backend("command-r-plus"), Backend::Estimation);
|
|
341
|
+
}
|
|
342
|
+
}
|
|
@@ -0,0 +1,265 @@
|
|
|
1
|
+
//! `tiktoken-rs` adapter implementing [`Tokenizer`].
|
|
2
|
+
//!
|
|
3
|
+
//! `tiktoken-rs` and Python `tiktoken` use the same BPE merge tables; for the
|
|
4
|
+
//! same model and same input, this returns byte-identical token IDs and
|
|
5
|
+
//! therefore byte-identical token *counts*. This is what makes the parity
|
|
6
|
+
//! tests "byte-equal" rather than "approximate".
|
|
7
|
+
//!
|
|
8
|
+
//! Initialization (loading the BPE table) is non-trivial. Each encoding is
|
|
9
|
+
//! built lazily on first use and shared via `LazyLock<Arc<CoreBPE>>`, so the
|
|
10
|
+
//! first `for_model` call pays the cost and every subsequent call is cheap.
|
|
11
|
+
|
|
12
|
+
use std::sync::{Arc, LazyLock};
|
|
13
|
+
|
|
14
|
+
use thiserror::Error;
|
|
15
|
+
use tiktoken_rs::CoreBPE;
|
|
16
|
+
|
|
17
|
+
use super::{Backend, Tokenizer};
|
|
18
|
+
|
|
19
|
+
#[derive(Debug, Error)]
|
|
20
|
+
pub enum TiktokenError {
|
|
21
|
+
/// We don't know which encoding `model` should use. The caller can fall
|
|
22
|
+
/// back to estimation; the registry handles that automatically.
|
|
23
|
+
#[error("unknown encoding for model `{0}`")]
|
|
24
|
+
UnknownEncoding(String),
|
|
25
|
+
}
|
|
26
|
+
|
|
27
|
+
/// Lazy-built shared BPE for the four named encodings. Init failure here would
|
|
28
|
+
/// indicate `tiktoken-rs` itself is broken; we treat that as a programmer error
|
|
29
|
+
/// and panic.
|
|
30
|
+
static O200K: LazyLock<Arc<CoreBPE>> =
|
|
31
|
+
LazyLock::new(|| Arc::new(tiktoken_rs::o200k_base().expect("o200k_base init")));
|
|
32
|
+
static CL100K: LazyLock<Arc<CoreBPE>> =
|
|
33
|
+
LazyLock::new(|| Arc::new(tiktoken_rs::cl100k_base().expect("cl100k_base init")));
|
|
34
|
+
static P50K: LazyLock<Arc<CoreBPE>> =
|
|
35
|
+
LazyLock::new(|| Arc::new(tiktoken_rs::p50k_base().expect("p50k_base init")));
|
|
36
|
+
static R50K: LazyLock<Arc<CoreBPE>> =
|
|
37
|
+
LazyLock::new(|| Arc::new(tiktoken_rs::r50k_base().expect("r50k_base init")));
|
|
38
|
+
|
|
39
|
+
/// BPE token counter for OpenAI / o-series models.
|
|
40
|
+
pub struct TiktokenCounter {
|
|
41
|
+
model: String,
|
|
42
|
+
encoding_name: &'static str,
|
|
43
|
+
bpe: Arc<CoreBPE>,
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
impl std::fmt::Debug for TiktokenCounter {
|
|
47
|
+
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
|
|
48
|
+
f.debug_struct("TiktokenCounter")
|
|
49
|
+
.field("model", &self.model)
|
|
50
|
+
.field("encoding", &self.encoding_name)
|
|
51
|
+
.finish()
|
|
52
|
+
}
|
|
53
|
+
}
|
|
54
|
+
|
|
55
|
+
impl TiktokenCounter {
|
|
56
|
+
/// Build a counter for `model`. Returns `UnknownEncoding` if the model
|
|
57
|
+
/// doesn't fall into any of the supported BPE families.
|
|
58
|
+
pub fn for_model(model: &str) -> Result<Self, TiktokenError> {
|
|
59
|
+
let encoding_name = encoding_for(model)?;
|
|
60
|
+
let bpe = match encoding_name {
|
|
61
|
+
"o200k_base" => O200K.clone(),
|
|
62
|
+
"cl100k_base" => CL100K.clone(),
|
|
63
|
+
"p50k_base" => P50K.clone(),
|
|
64
|
+
"r50k_base" => R50K.clone(),
|
|
65
|
+
// unreachable: encoding_for only returns the four names above.
|
|
66
|
+
_ => return Err(TiktokenError::UnknownEncoding(model.to_string())),
|
|
67
|
+
};
|
|
68
|
+
Ok(Self {
|
|
69
|
+
model: model.to_string(),
|
|
70
|
+
encoding_name,
|
|
71
|
+
bpe,
|
|
72
|
+
})
|
|
73
|
+
}
|
|
74
|
+
|
|
75
|
+
pub fn model(&self) -> &str {
|
|
76
|
+
&self.model
|
|
77
|
+
}
|
|
78
|
+
|
|
79
|
+
pub fn encoding_name(&self) -> &'static str {
|
|
80
|
+
self.encoding_name
|
|
81
|
+
}
|
|
82
|
+
}
|
|
83
|
+
|
|
84
|
+
impl Tokenizer for TiktokenCounter {
|
|
85
|
+
fn count_text(&self, text: &str) -> usize {
|
|
86
|
+
if text.is_empty() {
|
|
87
|
+
// Match Python `TiktokenCounter.count_text`: short-circuit empty.
|
|
88
|
+
return 0;
|
|
89
|
+
}
|
|
90
|
+
// For ORDINARY input (no literal special-token strings) `encode_ordinary`
|
|
91
|
+
// here and `encoding.encode(text)` in Python yield identical token IDs
|
|
92
|
+
// and counts — that's the byte-equality the parity harness verifies.
|
|
93
|
+
//
|
|
94
|
+
// Divergence (rare in practice): if `text` contains a literal
|
|
95
|
+
// `<|endoftext|>` (or any other special-token string), Python's default
|
|
96
|
+
// `encode` raises (because `disallowed_special="all"`) while we treat
|
|
97
|
+
// it as ordinary text. We chose tolerance over panic since proxy users
|
|
98
|
+
// can legitimately send those substrings; document for future readers.
|
|
99
|
+
self.bpe.encode_ordinary(text).len()
|
|
100
|
+
}
|
|
101
|
+
|
|
102
|
+
fn backend(&self) -> Backend {
|
|
103
|
+
Backend::Tiktoken
|
|
104
|
+
}
|
|
105
|
+
}
|
|
106
|
+
|
|
107
|
+
/// Map model → encoding name. Mirrors `MODEL_ENCODINGS` and the prefix
|
|
108
|
+
/// fallbacks in `headroom/tokenizers/tiktoken_counter.py`.
|
|
109
|
+
fn encoding_for(model: &str) -> Result<&'static str, TiktokenError> {
|
|
110
|
+
let m = model.to_ascii_lowercase();
|
|
111
|
+
|
|
112
|
+
// o200k_base: GPT-4o + o1/o3 reasoning families.
|
|
113
|
+
if m.starts_with("gpt-4o") || m.starts_with("o1") || m.starts_with("o3") {
|
|
114
|
+
return Ok("o200k_base");
|
|
115
|
+
}
|
|
116
|
+
|
|
117
|
+
// cl100k_base: GPT-4, GPT-3.5-turbo, embeddings.
|
|
118
|
+
if m.starts_with("gpt-4") || m.starts_with("gpt-3.5") || m.starts_with("text-embedding") {
|
|
119
|
+
return Ok("cl100k_base");
|
|
120
|
+
}
|
|
121
|
+
|
|
122
|
+
// p50k_base: code-* and the davinci-002/003 text-completion line.
|
|
123
|
+
if m.starts_with("code-")
|
|
124
|
+
|| m.starts_with("text-davinci-002")
|
|
125
|
+
|| m.starts_with("text-davinci-003")
|
|
126
|
+
{
|
|
127
|
+
return Ok("p50k_base");
|
|
128
|
+
}
|
|
129
|
+
|
|
130
|
+
// r50k_base: legacy davinci-001 and earlier completion families.
|
|
131
|
+
if m.starts_with("text-davinci")
|
|
132
|
+
|| m.starts_with("davinci")
|
|
133
|
+
|| m.starts_with("curie")
|
|
134
|
+
|| m.starts_with("babbage")
|
|
135
|
+
|| m.starts_with("ada")
|
|
136
|
+
{
|
|
137
|
+
return Ok("r50k_base");
|
|
138
|
+
}
|
|
139
|
+
|
|
140
|
+
Err(TiktokenError::UnknownEncoding(model.to_string()))
|
|
141
|
+
}
|
|
142
|
+
|
|
143
|
+
#[cfg(test)]
|
|
144
|
+
mod tests {
|
|
145
|
+
use super::*;
|
|
146
|
+
|
|
147
|
+
#[test]
|
|
148
|
+
fn empty_string_is_zero() {
|
|
149
|
+
let t = TiktokenCounter::for_model("gpt-4o-mini").unwrap();
|
|
150
|
+
assert_eq!(t.count_text(""), 0);
|
|
151
|
+
}
|
|
152
|
+
|
|
153
|
+
#[test]
|
|
154
|
+
fn nonempty_text_is_at_least_one_token() {
|
|
155
|
+
let t = TiktokenCounter::for_model("gpt-4o-mini").unwrap();
|
|
156
|
+
assert!(t.count_text("a") >= 1);
|
|
157
|
+
}
|
|
158
|
+
|
|
159
|
+
#[test]
|
|
160
|
+
fn known_token_counts_for_o200k() {
|
|
161
|
+
// These constants are the o200k_base BPE token counts produced by
|
|
162
|
+
// both Python `tiktoken.encoding_for_model("gpt-4o-mini")` and
|
|
163
|
+
// `tiktoken-rs::o200k_base()` for the given strings. They lock in
|
|
164
|
+
// byte-equal parity so a future tiktoken-rs upgrade that subtly
|
|
165
|
+
// changes BPE behavior would fail this test.
|
|
166
|
+
let t = TiktokenCounter::for_model("gpt-4o-mini").unwrap();
|
|
167
|
+
assert_eq!(t.count_text("hello"), 1);
|
|
168
|
+
assert_eq!(t.count_text("Hello, world!"), 4);
|
|
169
|
+
assert_eq!(
|
|
170
|
+
t.count_text("the quick brown fox jumps over the lazy dog"),
|
|
171
|
+
9
|
|
172
|
+
);
|
|
173
|
+
}
|
|
174
|
+
|
|
175
|
+
#[test]
|
|
176
|
+
fn determinism() {
|
|
177
|
+
let t = TiktokenCounter::for_model("gpt-4o").unwrap();
|
|
178
|
+
let s = "Determinism check across many calls.";
|
|
179
|
+
let first = t.count_text(s);
|
|
180
|
+
for _ in 0..1000 {
|
|
181
|
+
assert_eq!(t.count_text(s), first);
|
|
182
|
+
}
|
|
183
|
+
}
|
|
184
|
+
|
|
185
|
+
#[test]
|
|
186
|
+
fn unicode_input_does_not_panic() {
|
|
187
|
+
let t = TiktokenCounter::for_model("gpt-4o-mini").unwrap();
|
|
188
|
+
// Each call should produce a reasonable count (>=1 for non-empty),
|
|
189
|
+
// not panic, and not return absurd values.
|
|
190
|
+
for s in [
|
|
191
|
+
"héllo wörld", // accented
|
|
192
|
+
"你好世界", // CJK
|
|
193
|
+
"مرحبا بالعالم", // Arabic (RTL)
|
|
194
|
+
"🦀 ferris the crab", // emoji
|
|
195
|
+
"\n\t\r\x07", // control chars
|
|
196
|
+
] {
|
|
197
|
+
let n = t.count_text(s);
|
|
198
|
+
assert!(n >= 1, "{s:?}");
|
|
199
|
+
assert!(n < s.len() * 4 + 10, "absurd count {n} for {s:?}");
|
|
200
|
+
}
|
|
201
|
+
}
|
|
202
|
+
|
|
203
|
+
#[test]
|
|
204
|
+
fn very_long_input() {
|
|
205
|
+
let t = TiktokenCounter::for_model("gpt-4o-mini").unwrap();
|
|
206
|
+
let s = "the quick brown fox ".repeat(50_000); // ~1MB
|
|
207
|
+
let n = t.count_text(&s);
|
|
208
|
+
// 50k repeats * ~5 tokens per repeat = ~250k tokens, sanity bound.
|
|
209
|
+
assert!(n > 100_000 && n < 1_000_000, "n={n}");
|
|
210
|
+
}
|
|
211
|
+
|
|
212
|
+
#[test]
|
|
213
|
+
fn encoding_dispatch() {
|
|
214
|
+
for (model, expected) in [
|
|
215
|
+
("gpt-4o", "o200k_base"),
|
|
216
|
+
("gpt-4o-mini", "o200k_base"),
|
|
217
|
+
("gpt-4o-2024-08-06", "o200k_base"),
|
|
218
|
+
("o1-preview", "o200k_base"),
|
|
219
|
+
("o3-mini", "o200k_base"),
|
|
220
|
+
("gpt-4", "cl100k_base"),
|
|
221
|
+
("gpt-4-turbo", "cl100k_base"),
|
|
222
|
+
("gpt-3.5-turbo", "cl100k_base"),
|
|
223
|
+
("text-embedding-3-small", "cl100k_base"),
|
|
224
|
+
("code-davinci-002", "p50k_base"),
|
|
225
|
+
("text-davinci-002", "p50k_base"),
|
|
226
|
+
("text-davinci-003", "p50k_base"),
|
|
227
|
+
("text-davinci-001", "r50k_base"),
|
|
228
|
+
("davinci", "r50k_base"),
|
|
229
|
+
("curie", "r50k_base"),
|
|
230
|
+
("babbage", "r50k_base"),
|
|
231
|
+
("ada", "r50k_base"),
|
|
232
|
+
] {
|
|
233
|
+
let t = TiktokenCounter::for_model(model)
|
|
234
|
+
.unwrap_or_else(|e| panic!("for_model({model}) failed: {e}"));
|
|
235
|
+
assert_eq!(t.encoding_name(), expected, "{model}");
|
|
236
|
+
}
|
|
237
|
+
}
|
|
238
|
+
|
|
239
|
+
#[test]
|
|
240
|
+
fn unknown_model_returns_error() {
|
|
241
|
+
let r = TiktokenCounter::for_model("claude-3-opus");
|
|
242
|
+
assert!(matches!(r, Err(TiktokenError::UnknownEncoding(_))));
|
|
243
|
+
}
|
|
244
|
+
|
|
245
|
+
#[test]
|
|
246
|
+
fn case_insensitive_dispatch() {
|
|
247
|
+
let t = TiktokenCounter::for_model("GPT-4o-Mini").unwrap();
|
|
248
|
+
assert_eq!(t.encoding_name(), "o200k_base");
|
|
249
|
+
}
|
|
250
|
+
|
|
251
|
+
#[test]
|
|
252
|
+
fn shared_bpe_instances() {
|
|
253
|
+
// Two counters for the same encoding should share the underlying BPE
|
|
254
|
+
// (same Arc), proving the LazyLock cache works.
|
|
255
|
+
let a = TiktokenCounter::for_model("gpt-4o").unwrap();
|
|
256
|
+
let b = TiktokenCounter::for_model("gpt-4o-mini").unwrap();
|
|
257
|
+
assert!(Arc::ptr_eq(&a.bpe, &b.bpe));
|
|
258
|
+
}
|
|
259
|
+
|
|
260
|
+
#[test]
|
|
261
|
+
fn backend_is_tiktoken() {
|
|
262
|
+
let t = TiktokenCounter::for_model("gpt-4o-mini").unwrap();
|
|
263
|
+
assert_eq!(t.backend(), Backend::Tiktoken);
|
|
264
|
+
}
|
|
265
|
+
}
|