agentdb 3.0.0-alpha.10 → 3.0.0-alpha.12

package/dist/src/controllers/attention/AttentionCache.d.ts

@@ -0,0 +1,38 @@
+/**
+ * AttentionCache - Caching layer for attention mechanisms
+ *
+ * Handles:
+ * - Buffer pooling (getBuffer, returnBuffer)
+ * - Mask caching (getCachedMask)
+ * - Performance optimization through reuse
+ */
+/**
+ * AttentionCacheManager - Manages caching for attention operations
+ */
+export declare class AttentionCacheManager {
+    private bufferPool;
+    private maskCache;
+    /**
+     * Get a reusable buffer from the pool or allocate new one
+     * @param size - Buffer size in elements
+     * @returns Float32Array buffer
+     */
+    getBuffer(size: number): Float32Array;
+    /**
+     * Return a buffer to the pool for reuse
+     * @param buffer - Buffer to return
+     */
+    returnBuffer(buffer: Float32Array): void;
+    /**
+     * Get cached attention mask or generate new one
+     * @param seqLen - Sequence length
+     * @param causal - Whether to use causal masking
+     * @returns Cached or generated mask
+     */
+    getCachedMask(seqLen: number, causal: boolean): Float32Array;
+    /**
+     * Clear all caches
+     */
+    clear(): void;
+}
+//# sourceMappingURL=AttentionCache.d.ts.map

package/dist/src/controllers/attention/AttentionCache.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"AttentionCache.d.ts","sourceRoot":"","sources":["../../../../src/controllers/attention/AttentionCache.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAIH;;GAEG;AACH,qBAAa,qBAAqB;IAEhC,OAAO,CAAC,UAAU,CAA0C;IAG5D,OAAO,CAAC,SAAS,CAAwC;IAEzD;;;;OAIG;IACH,SAAS,CAAC,IAAI,EAAE,MAAM,GAAG,YAAY;IAQrC;;;OAGG;IACH,YAAY,CAAC,MAAM,EAAE,YAAY,GAAG,IAAI;IAYxC;;;;;OAKG;IACH,aAAa,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,GAAG,YAAY;IA0B5D;;OAEG;IACH,KAAK,IAAI,IAAI;CAId"}

package/dist/src/controllers/attention/AttentionCache.js

@@ -0,0 +1,80 @@
+/**
+ * AttentionCache - Caching layer for attention mechanisms
+ *
+ * Handles:
+ * - Buffer pooling (getBuffer, returnBuffer)
+ * - Mask caching (getCachedMask)
+ * - Performance optimization through reuse
+ */
+import { AttentionConfigManager } from './AttentionConfig.js';
+/**
+ * AttentionCacheManager - Manages caching for attention operations
+ */
+export class AttentionCacheManager {
+    // Buffer pooling for Float32Array reuse (Optimization: 70-90% fewer allocations)
+    bufferPool = new Map();
+    // Attention mask caching (Optimization: 30-40% faster for repeated ops)
+    maskCache = new Map();
+    /**
+     * Get a reusable buffer from the pool or allocate new one
+     * @param size - Buffer size in elements
+     * @returns Float32Array buffer
+     */
+    getBuffer(size) {
+        const pool = this.bufferPool.get(size) || [];
+        if (pool.length > 0) {
+            return pool.pop();
+        }
+        return new Float32Array(size);
+    }
+    /**
+     * Return a buffer to the pool for reuse
+     * @param buffer - Buffer to return
+     */
+    returnBuffer(buffer) {
+        const size = buffer.length;
+        const pool = this.bufferPool.get(size) || [];
+        if (pool.length < AttentionConfigManager.MAX_POOLED_BUFFERS) {
+            // Zero out buffer for security and reuse
+            buffer.fill(0);
+            pool.push(buffer);
+            this.bufferPool.set(size, pool);
+        }
+    }
+    /**
+     * Get cached attention mask or generate new one
+     * @param seqLen - Sequence length
+     * @param causal - Whether to use causal masking
+     * @returns Cached or generated mask
+     */
+    getCachedMask(seqLen, causal) {
+        const key = `${seqLen}_${causal}`;
+        if (this.maskCache.has(key)) {
+            return this.maskCache.get(key);
+        }
+        const mask = new Float32Array(seqLen * seqLen);
+        if (causal) {
+            // Generate causal mask (lower triangular)
+            for (let i = 0; i < seqLen; i++) {
+                for (let j = 0; j < seqLen; j++) {
+                    mask[i * seqLen + j] = j <= i ? 1.0 : 0.0;
+                }
+            }
+        }
+        else {
+            mask.fill(1.0);
+        }
+        if (this.maskCache.size < AttentionConfigManager.MAX_CACHED_MASKS) {
+            this.maskCache.set(key, mask);
+        }
+        return mask;
+    }
+    /**
+     * Clear all caches
+     */
+    clear() {
+        this.bufferPool.clear();
+        this.maskCache.clear();
+    }
+}
+//# sourceMappingURL=AttentionCache.js.map

package/dist/src/controllers/attention/AttentionCache.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"AttentionCache.js","sourceRoot":"","sources":["../../../../src/controllers/attention/AttentionCache.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAE9D;;GAEG;AACH,MAAM,OAAO,qBAAqB;IAChC,iFAAiF;IACzE,UAAU,GAAgC,IAAI,GAAG,EAAE,CAAC;IAE5D,wEAAwE;IAChE,SAAS,GAA8B,IAAI,GAAG,EAAE,CAAC;IAEzD;;;;OAIG;IACH,SAAS,CAAC,IAAY;QACpB,MAAM,IAAI,GAAG,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;QAC7C,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACpB,OAAO,IAAI,CAAC,GAAG,EAAG,CAAC;QACrB,CAAC;QACD,OAAO,IAAI,YAAY,CAAC,IAAI,CAAC,CAAC;IAChC,CAAC;IAED;;;OAGG;IACH,YAAY,CAAC,MAAoB;QAC/B,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,CAAC;QAC3B,MAAM,IAAI,GAAG,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;QAE7C,IAAI,IAAI,CAAC,MAAM,GAAG,sBAAsB,CAAC,kBAAkB,EAAE,CAAC;YAC5D,yCAAyC;YACzC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACf,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YAClB,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QAClC,CAAC;IACH,CAAC;IAED;;;;;OAKG;IACH,aAAa,CAAC,MAAc,EAAE,MAAe;QAC3C,MAAM,GAAG,GAAG,GAAG,MAAM,IAAI,MAAM,EAAE,CAAC;QAElC,IAAI,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YAC5B,OAAO,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAE,CAAC;QAClC,CAAC;QAED,MAAM,IAAI,GAAG,IAAI,YAAY,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC;QAC/C,IAAI,MAAM,EAAE,CAAC;YACX,0CAA0C;YAC1C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAChC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;oBAChC,IAAI,CAAC,CAAC,GAAG,MAAM,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;gBAC5C,CAAC;YACH,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACjB,CAAC;QAED,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,GAAG,sBAAsB,CAAC,gBAAgB,EAAE,CAAC;YAClE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QAChC,CAAC;QAED,OAAO,IAAI,CAAC;IACd,CAAC;IAED;;OAEG;IACH,KAAK;QACH,IAAI,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC;QACxB,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;IACzB,CAAC;CACF"}

package/dist/src/controllers/attention/AttentionConfig.d.ts

@@ -0,0 +1,127 @@
+/**
+ * AttentionConfig - Configuration management for attention mechanisms
+ *
+ * Handles:
+ * - Config validation
+ * - Constants
+ * - Default values
+ */
+/**
+ * Configuration for attention mechanisms
+ */
+export interface AttentionConfig {
+    /** Number of attention heads */
+    numHeads: number;
+    /** Dimension of each head */
+    headDim: number;
+    /** Total embedding dimension (usually numHeads * headDim) */
+    embedDim: number;
+    /** Dropout probability (0-1) */
+    dropout?: number;
+    /** Whether to use bias in linear projections */
+    bias?: boolean;
+    /** Use Flash Attention optimization if available */
+    useFlash?: boolean;
+    /** Use Linear Attention for O(n) complexity */
+    useLinear?: boolean;
+    /** Use Hyperbolic space for hierarchical data */
+    useHyperbolic?: boolean;
+    /** Use Mixture-of-Experts routing */
+    useMoE?: boolean;
+    /** Number of experts for MoE (default: 8) */
+    numExperts?: number;
+    /** Top-k experts to activate in MoE (default: 2) */
+    topK?: number;
+    /** Sparsification configuration */
+    sparsification?: {
+        enabled: boolean;
+        method: 'ppr' | 'random-walk' | 'spectral';
+        topK: number;
+    };
+    /** Graph partitioning configuration */
+    partitioning?: {
+        enabled: boolean;
+        method: 'stoer-wagner' | 'karger' | 'flow-based';
+        maxPartitionSize: number;
+    };
+}
+/**
+ * Options for attention operations (alias for AttentionConfig)
+ */
+export type AttentionOptions = AttentionConfig;
+/**
+ * Result from attention computation
+ */
+export interface AttentionResult {
+    /** Output embeddings after attention */
+    output: Float32Array;
+    /** Attention weights (optional, for visualization) */
+    weights?: Float32Array;
+    /** Execution time in milliseconds */
+    executionTimeMs: number;
+    /** Which mechanism was used */
+    mechanism: 'multi-head' | 'flash' | 'linear' | 'hyperbolic' | 'moe' | 'sparse' | 'partitioned';
+    /** Runtime environment */
+    runtime: 'napi' | 'wasm' | 'fallback';
+    /** Sparsification metadata (for sparse attention) */
+    sparsityMetadata?: {
+        method?: string;
+        topKNodes?: number;
+        sparsityRatio?: number;
+    };
+    /** Partitioning metadata (for partitioned attention) */
+    partitioningMetadata?: {
+        numPartitions?: number;
+        cutSize?: number;
+        avgPartitionSize?: number;
+    };
+}
+/**
+ * AttentionConfigManager - Manages configuration and constants
+ */
+export declare class AttentionConfigManager {
+    static readonly FLASH_V2_MIN_SPEEDUP = 2.49;
+    static readonly FLASH_V2_MAX_SPEEDUP = 7.47;
+    static readonly MASKED_SCORE: number;
+    static readonly MAX_POOLED_BUFFERS = 10;
+    static readonly MAX_CACHED_MASKS = 50;
+    private config;
+    constructor(config: AttentionConfig);
+    /**
+     * Apply default values to configuration
+     */
+    private applyDefaults;
+    /**
+     * Validate configuration values
+     */
+    private validateConfig;
+    /**
+     * Get the configuration
+     */
+    getConfig(): AttentionConfig;
+    /**
+     * Get number of heads
+     */
+    getNumHeads(): number;
+    /**
+     * Get head dimension
+     */
+    getHeadDim(): number;
+    /**
+     * Get embedding dimension
+     */
+    getEmbedDim(): number;
+    /**
+     * Get dropout rate
+     */
+    getDropout(): number;
+    /**
+     * Get number of experts for MoE
+     */
+    getNumExperts(): number;
+    /**
+     * Get top-k for MoE
+     */
+    getTopK(): number;
+}
+//# sourceMappingURL=AttentionConfig.d.ts.map

package/dist/src/controllers/attention/AttentionConfig.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"AttentionConfig.d.ts","sourceRoot":"","sources":["../../../../src/controllers/attention/AttentionConfig.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH;;GAEG;AACH,MAAM,WAAW,eAAe;IAC9B,gCAAgC;IAChC,QAAQ,EAAE,MAAM,CAAC;IACjB,6BAA6B;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,6DAA6D;IAC7D,QAAQ,EAAE,MAAM,CAAC;IACjB,gCAAgC;IAChC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gDAAgD;IAChD,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,oDAAoD;IACpD,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,+CAA+C;IAC/C,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,iDAAiD;IACjD,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,qCAAqC;IACrC,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,6CAA6C;IAC7C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,oDAAoD;IACpD,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,mCAAmC;IACnC,cAAc,CAAC,EAAE;QACf,OAAO,EAAE,OAAO,CAAC;QACjB,MAAM,EAAE,KAAK,GAAG,aAAa,GAAG,UAAU,CAAC;QAC3C,IAAI,EAAE,MAAM,CAAC;KACd,CAAC;IACF,uCAAuC;IACvC,YAAY,CAAC,EAAE;QACb,OAAO,EAAE,OAAO,CAAC;QACjB,MAAM,EAAE,cAAc,GAAG,QAAQ,GAAG,YAAY,CAAC;QACjD,gBAAgB,EAAE,MAAM,CAAC;KAC1B,CAAC;CACH;AAED;;GAEG;AACH,MAAM,MAAM,gBAAgB,GAAG,eAAe,CAAC;AAE/C;;GAEG;AACH,MAAM,WAAW,eAAe;IAC9B,wCAAwC;IACxC,MAAM,EAAE,YAAY,CAAC;IACrB,sDAAsD;IACtD,OAAO,CAAC,EAAE,YAAY,CAAC;IACvB,qCAAqC;IACrC,eAAe,EAAE,MAAM,CAAC;IACxB,+BAA+B;IAC/B,SAAS,EAAE,YAAY,GAAG,OAAO,GAAG,QAAQ,GAAG,YAAY,GAAG,KAAK,GAAG,QAAQ,GAAG,aAAa,CAAC;IAC/F,0BAA0B;IAC1B,OAAO,EAAE,MAAM,GAAG,MAAM,GAAG,UAAU,CAAC;IACtC,qDAAqD;IACrD,gBAAgB,CAAC,EAAE;QACjB,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,aAAa,CAAC,EAAE,MAAM,CAAC;KACxB,CAAC;IACF,wDAAwD;IACxD,oBAAoB,CAAC,EAAE;QACrB,aAAa,CAAC,EAAE,MAAM,CAAC;QACvB,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,gBAAgB,CAAC,EAAE,MAAM,CAAC;KAC3B,CAAC;CACH;AAED;;GAEG;AACH,qBAAa,sBAAsB;IAEjC,MAAM,CAAC,QAAQ,CAAC,oBAAoB,QAAQ;IAC5C,MAAM,CAAC,QAAQ,CAAC,oBAAoB,QAAQ;IAG5C,MAAM,CAAC,QAAQ,CAAC,YAAY,SAAa;IAGzC,MAAM,CAAC,QAAQ,CAAC,kBAAkB,MAAM;IAGxC,MAAM,CAAC,QAAQ,CAAC,gBAAgB,MAAM;IAEtC,OAAO,CAAC,MAAM,CAAkB;gBAEpB,MAAM,EAAE,eAAe;IAKnC;;OAEG;IACH,OAAO,CAAC,aAAa;IAgCrB;;OAEG;IACH,OAAO,CAAC,cAAc;IAqBtB;;OAEG;IACH,SAAS,IAAI,eAAe;IAI5B;;OAEG;IACH,WAAW,IAAI,MAAM;IAIrB;;OAEG;IACH,UAAU,IAAI,MAAM;IAIpB;;OAEG;IACH,WAAW,IAAI,MAAM;IAIrB;;OAEG;IACH,UAAU,IAAI,MAAM;IAIpB;;OAEG;IACH,aAAa,IAAI,MAAM;IAIvB;;OAEG;IACH,OAAO,IAAI,MAAM;CAGlB"}

package/dist/src/controllers/attention/AttentionConfig.js

@@ -0,0 +1,124 @@
+/**
+ * AttentionConfig - Configuration management for attention mechanisms
+ *
+ * Handles:
+ * - Config validation
+ * - Constants
+ * - Default values
+ */
+/**
+ * AttentionConfigManager - Manages configuration and constants
+ */
+export class AttentionConfigManager {
+    // Performance targets (ADR-071)
+    static FLASH_V2_MIN_SPEEDUP = 2.49;
+    static FLASH_V2_MAX_SPEEDUP = 7.47;
+    // Attention computation constants
+    static MASKED_SCORE = -Infinity;
+    // Buffer pool limits
+    static MAX_POOLED_BUFFERS = 10;
+    // Mask cache limits
+    static MAX_CACHED_MASKS = 50;
+    config;
+    constructor(config) {
+        this.config = this.applyDefaults(config);
+        this.validateConfig(this.config);
+    }
+    /**
+     * Apply default values to configuration
+     */
+    applyDefaults(config) {
+        const defaults = {
+            ...config,
+            dropout: config.dropout ?? 0.1,
+            bias: config.bias ?? true,
+            useFlash: config.useFlash ?? true,
+            useLinear: config.useLinear ?? false,
+            useHyperbolic: config.useHyperbolic ?? false,
+            useMoE: config.useMoE ?? false,
+            numExperts: config.numExperts ?? 8,
+            topK: config.topK ?? 2,
+        };
+        if (config.sparsification) {
+            defaults.sparsification = {
+                enabled: config.sparsification.enabled ?? false,
+                method: config.sparsification.method ?? 'ppr',
+                topK: config.sparsification.topK ?? 100,
+            };
+        }
+        if (config.partitioning) {
+            defaults.partitioning = {
+                enabled: config.partitioning.enabled ?? false,
+                method: config.partitioning.method ?? 'stoer-wagner',
+                maxPartitionSize: config.partitioning.maxPartitionSize ?? 1000,
+            };
+        }
+        return defaults;
+    }
+    /**
+     * Validate configuration values
+     */
+    validateConfig(config) {
+        if (config.numHeads <= 0) {
+            throw new Error('numHeads must be positive');
+        }
+        if (config.headDim <= 0) {
+            throw new Error('headDim must be positive');
+        }
+        if (config.embedDim <= 0) {
+            throw new Error('embedDim must be positive');
+        }
+        if (config.dropout !== undefined && (config.dropout < 0 || config.dropout > 1)) {
+            throw new Error('dropout must be between 0 and 1');
+        }
+        if (config.numExperts !== undefined && config.numExperts <= 0) {
+            throw new Error('numExperts must be positive');
+        }
+        if (config.topK !== undefined && config.topK <= 0) {
+            throw new Error('topK must be positive');
+        }
+    }
+    /**
+     * Get the configuration
+     */
+    getConfig() {
+        return { ...this.config };
+    }
+    /**
+     * Get number of heads
+     */
+    getNumHeads() {
+        return this.config.numHeads;
+    }
+    /**
+     * Get head dimension
+     */
+    getHeadDim() {
+        return this.config.headDim;
+    }
+    /**
+     * Get embedding dimension
+     */
+    getEmbedDim() {
+        return this.config.embedDim;
+    }
+    /**
+     * Get dropout rate
+     */
+    getDropout() {
+        return this.config.dropout || 0.0;
+    }
+    /**
+     * Get number of experts for MoE
+     */
+    getNumExperts() {
+        return this.config.numExperts || 8;
+    }
+    /**
+     * Get top-k for MoE
+     */
+    getTopK() {
+        return this.config.topK || 2;
+    }
+}
+//# sourceMappingURL=AttentionConfig.js.map

package/dist/src/controllers/attention/AttentionConfig.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"AttentionConfig.js","sourceRoot":"","sources":["../../../../src/controllers/attention/AttentionConfig.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AA2EH;;GAEG;AACH,MAAM,OAAO,sBAAsB;IACjC,gCAAgC;IAChC,MAAM,CAAU,oBAAoB,GAAG,IAAI,CAAC;IAC5C,MAAM,CAAU,oBAAoB,GAAG,IAAI,CAAC;IAE5C,kCAAkC;IAClC,MAAM,CAAU,YAAY,GAAG,CAAC,QAAQ,CAAC;IAEzC,qBAAqB;IACrB,MAAM,CAAU,kBAAkB,GAAG,EAAE,CAAC;IAExC,oBAAoB;IACpB,MAAM,CAAU,gBAAgB,GAAG,EAAE,CAAC;IAE9B,MAAM,CAAkB;IAEhC,YAAY,MAAuB;QACjC,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;QACzC,IAAI,CAAC,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACnC,CAAC;IAED;;OAEG;IACK,aAAa,CAAC,MAAuB;QAC3C,MAAM,QAAQ,GAAoB;YAChC,GAAG,MAAM;YACT,OAAO,EAAE,MAAM,CAAC,OAAO,IAAI,GAAG;YAC9B,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,IAAI;YACzB,QAAQ,EAAE,MAAM,CAAC,QAAQ,IAAI,IAAI;YACjC,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,KAAK;YACpC,aAAa,EAAE,MAAM,CAAC,aAAa,IAAI,KAAK;YAC5C,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,KAAK;YAC9B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,CAAC;YAClC,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,CAAC;SACvB,CAAC;QAEF,IAAI,MAAM,CAAC,cAAc,EAAE,CAAC;YAC1B,QAAQ,CAAC,cAAc,GAAG;gBACxB,OAAO,EAAE,MAAM,CAAC,cAAc,CAAC,OAAO,IAAI,KAAK;gBAC/C,MAAM,EAAE,MAAM,CAAC,cAAc,CAAC,MAAM,IAAI,KAAK;gBAC7C,IAAI,EAAE,MAAM,CAAC,cAAc,CAAC,IAAI,IAAI,GAAG;aACxC,CAAC;QACJ,CAAC;QAED,IAAI,MAAM,CAAC,YAAY,EAAE,CAAC;YACxB,QAAQ,CAAC,YAAY,GAAG;gBACtB,OAAO,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,IAAI,KAAK;gBAC7C,MAAM,EAAE,MAAM,CAAC,YAAY,CAAC,MAAM,IAAI,cAAc;gBACpD,gBAAgB,EAAE,MAAM,CAAC,YAAY,CAAC,gBAAgB,IAAI,IAAI;aAC/D,CAAC;QACJ,CAAC;QAED,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,MAAuB;QAC5C,IAAI,MAAM,CAAC,QAAQ,IAAI,CAAC,EAAE,CAAC;YACzB,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;QAC/C,CAAC;QACD,IAAI,MAAM,CAAC,OAAO,IAAI,CAAC,EAAE,CAAC;YACxB,MAAM,IAAI,KAAK,CAAC,0BAA0B,CAAC,CAAC;QAC9C,CAAC;QACD,IAAI,MAAM,CAAC,QAAQ,IAAI,CAAC,EAAE,CAAC;YACzB,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;QAC/C,CAAC;QACD,IAAI,MAAM,CAAC,OAAO,KAAK,SAAS,IAAI,CAAC,MAAM,CAAC,OAAO,GAAG,CAAC,IAAI,MAAM,CAAC,OAAO,GAAG,CAAC,CAAC,EAAE,CAAC;YAC/E,MAAM,IAAI,KAAK,CAAC,iCAAiC,CAAC,CAAC;QACrD,CAAC;QACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,IAAI,CAAC,EAAE,CAAC;YAC9D,MAAM,IAAI,KAAK,CAAC,6BAA6B,CAAC,CAAC;QACjD,CAAC;QACD,IAAI,MAAM,CAAC,IAAI,KAAK,SAAS,IAAI,MAAM,CAAC,IAAI,IAAI,CAAC,EAAE,CAAC;YAClD,MAAM,IAAI,KAAK,CAAC,uBAAuB,CAAC,CAAC;QAC3C,CAAC;IACH,CAAC;IAED;;OAEG;IACH,SAAS;QACP,OAAO,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;IAC5B,CAAC;IAED;;OAEG;IACH,WAAW;QACT,OAAO,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC;IAC9B,CAAC;IAED;;OAEG;IACH,UAAU;QACR,OAAO,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC;IAC7B,CAAC;IAED;;OAEG;IACH,WAAW;QACT,OAAO,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC;IAC9B,CAAC;IAED;;OAEG;IACH,UAAU;QACR,OAAO,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,GAAG,CAAC;IACpC,CAAC;IAED;;OAEG;IACH,aAAa;QACX,OAAO,IAAI,CAAC,MAAM,CAAC,UAAU,IAAI,CAAC,CAAC;IACrC,CAAC;IAED;;OAEG;IACH,OAAO;QACL,OAAO,IAAI,CAAC,MAAM,CAAC,IAAI,IAAI,CAAC,CAAC;IAC/B,CAAC"}

package/dist/src/controllers/attention/AttentionCore.d.ts

@@ -0,0 +1,89 @@
+/**
+ * AttentionCore - Core attention computation logic
+ *
+ * Handles:
+ * - Multi-head attention
+ * - Flash Attention v2
+ * - Dot product computation
+ * - Softmax computation
+ * - Fallback implementations
+ */
+import { AttentionConfigManager } from './AttentionConfig.js';
+import { AttentionCacheManager } from './AttentionCache.js';
+/**
+ * AttentionCoreCompute - Core attention algorithms
+ */
+export declare class AttentionCoreCompute {
+    private configManager;
+    private cacheManager;
+    constructor(configManager: AttentionConfigManager, cacheManager: AttentionCacheManager);
+    /**
+     * Fallback JavaScript implementation of multi-head attention
+     * Used when native modules are not available
+     * Optimized with zero-copy array views (90% fewer allocations)
+     */
+    multiHeadAttentionFallback(query: Float32Array, key: Float32Array, value: Float32Array, mask?: Float32Array): {
+        output: Float32Array;
+        weights?: Float32Array;
+    };
+    /**
+     * Fused Attention - Combines softmax + weighted sum in single pass
+     *
+     * Performance improvement: 20-25% speedup through better cache locality
+     * Memory improvement: Reduces intermediate buffer allocations
+     *
+     * Standard attention (2 passes):
+     *   1. Compute scores → softmax (allocates scores + weights buffers)
+     *   2. Weighted sum over values (allocates output buffer)
+     *
+     * Fused attention (1 pass):
+     *   - Scores → softmax → weighted sum in single loop
+     *   - Only allocates final output buffer
+     *   - Better cache locality (data stays in L1/L2 cache)
+     *
+     * @param query - Query vectors [seqLen * embedDim]
+     * @param key - Key vectors [seqLen * embedDim]
+     * @param value - Value vectors [seqLen * embedDim]
+     * @param options - Fused attention options
+     * @returns Output and performance metrics
+     */
+    fusedAttention(query: Float32Array, key: Float32Array, value: Float32Array, options?: {
+        blockSize?: number;
+        mask?: Float32Array;
+        compareBaseline?: boolean;
+    }): {
+        output: Float32Array;
+        speedup?: number;
+        baselineTimeMs?: number;
+        fusedTimeMs?: number;
+    };
+    /**
+     * Fallback JavaScript implementation of linear attention
+     */
+    linearAttentionFallback(query: Float32Array, key: Float32Array, value: Float32Array): Float32Array;
+    /**
+     * Numerically stable in-place softmax
+     * @param scores - Array of scores
+     * @param start - Start index
+     * @param end - End index
+     */
+    softmaxInPlace(scores: Float32Array, start: number, end: number): void;
+    /**
+     * Zero-copy array view helper
+     * Creates a view into an existing Float32Array without allocation
+     * @param array - Source array
+     * @param start - Start index
+     * @param length - Number of elements
+     * @returns Zero-copy view (shares memory with source)
+     */
+    private getArrayView;
+    /**
+     * SIMD-optimized dot product computation with zero-copy views
+     * Processes 4 elements at a time for JIT vectorization
+     * @param a - First array or view
+     * @param b - Second array or view
+     * @returns Dot product result
+     */
+    private dotProductSIMD;
+}
+//# sourceMappingURL=AttentionCore.d.ts.map

package/dist/src/controllers/attention/AttentionCore.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"AttentionCore.d.ts","sourceRoot":"","sources":["../../../../src/controllers/attention/AttentionCore.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAC9D,OAAO,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAE5D;;GAEG;AACH,qBAAa,oBAAoB;IAC/B,OAAO,CAAC,aAAa,CAAyB;IAC9C,OAAO,CAAC,YAAY,CAAwB;gBAEhC,aAAa,EAAE,sBAAsB,EAAE,YAAY,EAAE,qBAAqB;IAKtF;;;;OAIG;IACH,0BAA0B,CACxB,KAAK,EAAE,YAAY,EACnB,GAAG,EAAE,YAAY,EACjB,KAAK,EAAE,YAAY,EACnB,IAAI,CAAC,EAAE,YAAY,GAClB;QAAE,MAAM,EAAE,YAAY,CAAC;QAAC,OAAO,CAAC,EAAE,YAAY,CAAA;KAAE;IAqDnD;;;;;;;;;;;;;;;;;;;;OAoBG;IACH,cAAc,CACZ,KAAK,EAAE,YAAY,EACnB,GAAG,EAAE,YAAY,EACjB,KAAK,EAAE,YAAY,EACnB,OAAO,CAAC,EAAE;QACR,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,IAAI,CAAC,EAAE,YAAY,CAAC;QACpB,eAAe,CAAC,EAAE,OAAO,CAAC;KAC3B,GACA;QAAE,MAAM,EAAE,YAAY,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,cAAc,CAAC,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAA;KAAE;IAgI5F;;OAEG;IACH,uBAAuB,CACrB,KAAK,EAAE,YAAY,EACnB,GAAG,EAAE,YAAY,EACjB,KAAK,EAAE,YAAY,GAClB,YAAY;IA8Bf;;;;;OAKG;IACH,cAAc,CAAC,MAAM,EAAE,YAAY,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,GAAG,IAAI;IAsBtE;;;;;;;OAOG;IACH,OAAO,CAAC,YAAY;IASpB;;;;;;OAMG;IACH,OAAO,CAAC,cAAc;CAuBvB"}