webinfer 0.0.1

package/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/dist/attention/block-sparse/format.d.ts

@@ -0,0 +1,52 @@
+/**
+ * Block-Sparse CSR Format
+ * Enables a single kernel to support all attention variants
+ */
+/**
+ * Block-Sparse CSR representation of attention mask
+ */
+export interface BlockSparseCSR {
+    blockSize: number;
+    rowPtr: Uint32Array;
+    colIdx: Uint32Array;
+    blockMask?: Uint8Array;
+    numRows: number;
+    numCols: number;
+    numBlockRows: number;
+    numBlockCols: number;
+    nnzBlocks: number;
+}
+/**
+ * Attention pattern types
+ */
+export type AttentionPattern = {
+    type: "dense";
+} | {
+    type: "causal";
+} | {
+    type: "sliding";
+    windowSize: number;
+} | {
+    type: "global-local";
+    globalTokens: number[];
+    localWindow: number;
+} | {
+    type: "custom";
+    mask: boolean[][];
+};
+/**
+ * Build BS-CSR from attention pattern
+ */
+export declare function buildBlockSparseCSR(seqLen: number, pattern: AttentionPattern, blockSize?: number): BlockSparseCSR;
+/**
+ * Calculate sparsity ratio of the mask
+ */
+export declare function getSparsityRatio(csr: BlockSparseCSR): number;
+/**
+ * Estimate memory savings from sparsity
+ */
+export declare function estimateMemorySavings(csr: BlockSparseCSR): {
+    denseBytes: number;
+    sparseBytes: number;
+    savingsRatio: number;
+};

package/dist/attention/block-sparse/patterns/causal.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Causal (autoregressive) attention pattern
+ * Used in GPT, Llama, and most decoder-only models
+ */
+import { type BlockSparseCSR } from "../format.ts";
+/**
+ * Build causal attention mask in BS-CSR format
+ * Each query position can only attend to positions <= its own position
+ */
+export declare function buildCausalMask(seqLen: number, blockSize?: number): BlockSparseCSR;
+/**
+ * Get the theoretical sparsity of causal attention
+ * For a sequence of length N, causal attention has N*(N+1)/2 non-zero elements
+ * out of N*N total, giving ~50% sparsity for large N
+ */
+export declare function getCausalSparsity(seqLen: number): number;

package/dist/attention/block-sparse/patterns/sliding.d.ts

@@ -0,0 +1,22 @@
+/**
+ * Sliding window attention pattern
+ * Used in Mistral and other efficient attention models
+ */
+import { type BlockSparseCSR } from "../format.ts";
+/**
+ * Build sliding window attention mask in BS-CSR format
+ * Each query can only attend to the previous `windowSize` positions
+ */
+export declare function buildSlidingWindowMask(seqLen: number, windowSize: number, blockSize?: number): BlockSparseCSR;
+/**
+ * Get the theoretical sparsity of sliding window attention
+ * For window size W and sequence length N:
+ * - First W positions have triangular attention (like causal)
+ * - Remaining N-W positions have W+1 attention each
+ */
+export declare function getSlidingWindowSparsity(seqLen: number, windowSize: number): number;
+/**
+ * Sliding window with causal constraint
+ * This is what Mistral uses - combines sliding window with causal masking
+ */
+export declare function buildCausalSlidingWindowMask(seqLen: number, windowSize: number, blockSize?: number): BlockSparseCSR;

package/dist/attention/flash-attention.d.ts

@@ -0,0 +1,30 @@
+/**
+ * FlashAttention Implementation for WebGPU
+ * Memory-efficient attention using online softmax and tiling
+ */
+import type { WebInferDevice } from "../core/device.ts";
+import { Tensor } from "../core/tensor.ts";
+import type { AttentionPattern } from "./block-sparse/format.ts";
+export interface AttentionConfig {
+    numHeads: number;
+    headDim: number;
+    seqLen: number;
+    scale?: number;
+    pattern?: AttentionPattern;
+    blockSize?: number;
+}
+/**
+ * FlashAttention forward pass
+ * Computes: softmax(Q @ K^T / sqrt(d)) @ V
+ *
+ * @param device WebInfer device
+ * @param q Query tensor [batch, seqLen, numHeads, headDim]
+ * @param k Key tensor [batch, seqLen, numHeads, headDim]
+ * @param v Value tensor [batch, seqLen, numHeads, headDim]
+ * @param config Attention configuration
+ */
+export declare function flashAttention(device: WebInferDevice, q: Tensor, k: Tensor, v: Tensor, config: AttentionConfig): Promise<Tensor>;
+/**
+ * CPU reference implementation for verification
+ */
+export declare function attentionCPU(q: Float32Array, k: Float32Array, v: Float32Array, seqLen: number, numHeads: number, headDim: number, causal?: boolean): Float32Array;

package/dist/attention/index.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Attention Module Exports
+ */
+export { flashAttention, attentionCPU, type AttentionConfig, } from "./flash-attention.ts";
+export { buildBlockSparseCSR, getSparsityRatio, estimateMemorySavings, type BlockSparseCSR, type AttentionPattern, } from "./block-sparse/format.ts";
+export { buildCausalMask, getCausalSparsity } from "./block-sparse/patterns/causal.ts";
+export { buildSlidingWindowMask, buildCausalSlidingWindowMask, getSlidingWindowSparsity, } from "./block-sparse/patterns/sliding.ts";
+export { AttentionScheduler, type ChunkPlan } from "./scheduler.ts";
+export { PagedKVCache, type PagedKVCacheConfig, type SequenceEntry, BlockManager, ContinuousBatchScheduler, type BlockManagerConfig, type AllocationPolicy, type AllocationRequest, } from "./paged-kv/index.ts";

package/dist/attention/paged-kv/block-manager.d.ts

@@ -0,0 +1,102 @@
+/**
+ * Block Manager - High-level memory management for PagedKVCache
+ */
+import type { WebInferDevice } from "../../core/device.ts";
+import { PagedKVCache, type PagedKVCacheConfig } from "./page-table.ts";
+/**
+ * Block allocation policy
+ */
+export type AllocationPolicy = "greedy" | "best-fit" | "first-fit";
+/**
+ * Block Manager configuration
+ */
+export interface BlockManagerConfig extends PagedKVCacheConfig {
+    policy?: AllocationPolicy;
+    reservedPages?: number;
+}
+/**
+ * Request for KV cache allocation
+ */
+export interface AllocationRequest {
+    seqId?: number;
+    numTokens: number;
+    priority?: number;
+}
+/**
+ * Block Manager - Manages KV cache allocation across multiple sequences
+ */
+export declare class BlockManager {
+    private cache;
+    private config;
+    private priorities;
+    constructor(device: WebInferDevice, config: BlockManagerConfig);
+    /**
+     * Check if allocation is possible
+     */
+    canAllocate(request: AllocationRequest): boolean;
+    /**
+     * Allocate or extend a sequence
+     */
+    allocate(request: AllocationRequest): number;
+    /**
+     * Free a sequence
+     */
+    free(seqId: number): void;
+    /**
+     * Evict lowest priority sequences until we have enough free pages
+     */
+    evict(neededPages: number): number[];
+    /**
+     * Get memory utilization
+     */
+    getUtilization(): number;
+    /**
+     * Get the underlying cache
+     */
+    getCache(): PagedKVCache;
+    /**
+     * Get statistics
+     */
+    getStats(): {
+        totalPages: number;
+        usedPages: number;
+        freePages: number;
+        numSequences: number;
+        memoryUsedBytes: number;
+        memoryTotalBytes: number;
+    };
+    /**
+     * Dispose resources
+     */
+    dispose(): void;
+}
+/**
+ * Scheduler for continuous batching with PagedAttention
+ */
+export declare class ContinuousBatchScheduler {
+    private blockManager;
+    private runningSequences;
+    private waitingQueue;
+    constructor(blockManager: BlockManager);
+    /**
+     * Add a new request to the scheduler
+     */
+    addRequest(request: AllocationRequest): void;
+    /**
+     * Complete a sequence
+     */
+    completeSequence(seqId: number): void;
+    /**
+     * Extend a running sequence
+     */
+    extendSequence(seqId: number, numNewTokens: number): boolean;
+    private scheduleWaiting;
+    /**
+     * Get running sequence count
+     */
+    getRunningCount(): number;
+    /**
+     * Get waiting request count
+     */
+    getWaitingCount(): number;
+}

package/dist/attention/paged-kv/index.d.ts

@@ -0,0 +1,5 @@
+/**
+ * PagedKV Module Exports
+ */
+export { PagedKVCache, type PagedKVCacheConfig, type SequenceEntry, } from "./page-table.ts";
+export { BlockManager, ContinuousBatchScheduler, type BlockManagerConfig, type AllocationPolicy, type AllocationRequest, } from "./block-manager.ts";

package/dist/attention/paged-kv/page-table.d.ts

@@ -0,0 +1,99 @@
+/**
+ * Paged KV Cache - Software page table for efficient memory management
+ * Inspired by vLLM's PagedAttention
+ */
+import type { WebInferDevice } from "../../core/device.ts";
+/**
+ * Configuration for PagedKVCache
+ */
+export interface PagedKVCacheConfig {
+    numLayers: number;
+    numHeads: number;
+    headDim: number;
+    pageSize: number;
+    maxPages: number;
+    dtype?: "f32" | "f16";
+}
+/**
+ * Entry in the page table for a sequence
+ */
+export interface SequenceEntry {
+    seqId: number;
+    pages: number[];
+    length: number;
+}
+/**
+ * PagedKVCache - Manages KV cache with paging for efficient memory use
+ *
+ * Benefits:
+ * 1. No memory fragmentation - pages are fixed size
+ * 2. Efficient memory sharing - multiple sequences can share cache
+ * 3. Dynamic allocation - only allocate pages as needed
+ * 4. Easy defragmentation - just remap logical to physical pages
+ */
+export declare class PagedKVCache {
+    private device;
+    private config;
+    private keyCache;
+    private valueCache;
+    private pageTable;
+    private freePages;
+    private nextSeqId;
+    constructor(device: WebInferDevice, config: PagedKVCacheConfig);
+    /**
+     * Allocate pages for a new sequence
+     */
+    allocateSequence(initialLength?: number): number;
+    /**
+     * Extend a sequence with new tokens
+     */
+    extendSequence(seqId: number, numNewTokens: number): void;
+    /**
+     * Free a sequence and its pages
+     */
+    freeSequence(seqId: number): void;
+    /**
+     * Get page indices for a sequence
+     */
+    getSequencePages(seqId: number): number[] | null;
+    /**
+     * Get sequence length
+     */
+    getSequenceLength(seqId: number): number;
+    /**
+     * Get the physical page index for a given sequence position
+     */
+    getPageForPosition(seqId: number, position: number): number | null;
+    /**
+     * Get offset within a page for a given position
+     */
+    getOffsetInPage(position: number): number;
+    private allocatePage;
+    private freePage;
+    /**
+     * Get cache statistics
+     */
+    getStats(): {
+        totalPages: number;
+        usedPages: number;
+        freePages: number;
+        numSequences: number;
+        memoryUsedBytes: number;
+        memoryTotalBytes: number;
+    };
+    /**
+     * Get GPU buffers for kernel binding
+     */
+    getBuffers(): {
+        keyCache: GPUBuffer;
+        valueCache: GPUBuffer;
+    };
+    /**
+     * Get configuration
+     */
+    getConfig(): PagedKVCacheConfig;
+    /**
+     * Dispose GPU resources
+     */
+    dispose(): void;
+}

package/dist/attention/scheduler.d.ts

@@ -0,0 +1,40 @@
+/**
+ * Attention Scheduler - Prevents TDR (GPU timeout) by splitting long sequences
+ */
+import type { WebInferDevice } from "../core/device.ts";
+export interface ChunkPlan {
+    numChunks: number;
+    chunkSize: number;
+    estimatedTimeMs: number;
+}
+/**
+ * Attention Scheduler for TDR prevention
+ * Splits long sequences into chunks to avoid GPU timeout
+ */
+export declare class AttentionScheduler {
+    private device;
+    private tdrLimit;
+    constructor(device: WebInferDevice);
+    private detectTDRLimit;
+    /**
+     * Estimate execution time for attention operation
+     * Based on empirical formula: time ∝ seqLen² × numHeads × headDim
+     */
+    estimateExecutionTime(seqLen: number, numHeads: number, headDim: number): number;
+    /**
+     * Compute chunk plan for given sequence length
+     */
+    computeChunkPlan(seqLen: number, numHeads: number, headDim: number): ChunkPlan;
+    /**
+     * Yield to main thread to prevent TDR
+     */
+    yieldToMain(): Promise<void>;
+    /**
+     * Check if sequence might cause TDR
+     */
+    mightCauseTDR(seqLen: number, numHeads: number, headDim: number): boolean;
+    /**
+     * Get recommended maximum sequence length for single-pass execution
+     */
+    getMaxSinglePassSeqLen(numHeads: number, headDim: number): number;
+}

package/dist/core/buffer-pool.d.ts

@@ -0,0 +1,18 @@
+/**
+ * GPU Buffer Pool - Reduces allocation overhead and memory fragmentation
+ */
+export declare class BufferPool {
+    private device;
+    private pools;
+    private sizeClasses;
+    constructor(device: GPUDevice);
+    private getSizeClass;
+    acquire(size: number, usage: GPUBufferUsageFlags): GPUBuffer;
+    release(buffer: GPUBuffer): void;
+    getStats(): {
+        totalBuffers: number;
+        inUse: number;
+        totalBytes: number;
+    };
+    dispose(): void;
+}

package/dist/core/device.d.ts

@@ -0,0 +1,23 @@
+/**
+ * WebGPU Device Management
+ */
+export interface DeviceInfo {
+    vendor: "apple" | "nvidia" | "intel" | "amd" | "unknown";
+    architecture: string;
+    maxWorkgroupSize: number;
+    maxComputeInvocationsPerWorkgroup: number;
+    maxStorageBufferBindingSize: number;
+}
+export declare class WebInferDevice {
+    private _device;
+    private _info;
+    private constructor();
+    static create(): Promise<WebInferDevice>;
+    private static detectDeviceInfo;
+    get device(): GPUDevice;
+    get info(): DeviceInfo;
+    get limits(): GPUSupportedLimits;
+    createCommandEncoder(): GPUCommandEncoder;
+    submit(commandBuffers: GPUCommandBuffer[]): void;
+    dispose(): void;
+}

package/dist/core/tensor.d.ts

@@ -0,0 +1,25 @@
+/**
+ * Tensor - GPU-backed multidimensional array
+ */
+import type { WebInferDevice } from "./device.ts";
+export type DType = "f32" | "f16" | "i32" | "u32";
+export declare class Tensor {
+    private _device;
+    private _shape;
+    private _dtype;
+    private _buffer;
+    private _disposed;
+    constructor(device: WebInferDevice, shape: number[], dtype?: DType, data?: Float32Array | Uint32Array | Int32Array);
+    static fromArray(device: WebInferDevice, shape: number[], data: Float32Array, dtype?: DType): Promise<Tensor>;
+    static zeros(device: WebInferDevice, shape: number[], dtype?: DType): Tensor;
+    static rand(device: WebInferDevice, shape: number[], dtype?: DType): Tensor;
+    get shape(): readonly number[];
+    get dtype(): DType;
+    get numel(): number;
+    get byteSize(): number;
+    get buffer(): GPUBuffer;
+    get device(): WebInferDevice;
+    toArray(): Promise<Float32Array>;
+    reshape(newShape: number[]): Tensor;
+    dispose(): void;
+}