node-llama-cpp 3.0.0-beta.13 → 3.0.0-beta.15

package/dist/evaluator/TokenMeter.js

@@ -0,0 +1,86 @@
+/**
+ * Tracks the evaluation usage of tokens.
+ */
+export class TokenMeter {
+    _inputTokens = 0;
+    _outputTokens = 0;
+    _restoreStateTokens = 0;
+    /**
+     * The number of input tokens used
+     */
+    get usedInputTokens() {
+        return this._inputTokens;
+    }
+    /**
+     * The number of tokens generated by a model
+     */
+    get usedOutputTokens() {
+        return this._outputTokens;
+    }
+    /**
+     * The number of tokens used as input to restore a context sequence state to continue previous evaluation.
+     * This may be consumed by virtual context sequences.
+     */
+    get usedRestoreStateTokens() {
+        return this._restoreStateTokens;
+    }
+    /**
+     * Get the current state of the token meter
+     */
+    getState() {
+        return {
+            usedInputTokens: this.usedInputTokens,
+            usedOutputTokens: this.usedOutputTokens,
+            usedRestoreStateTokens: this.usedRestoreStateTokens
+        };
+    }
+    /**
+     * Log the usage of tokens
+     */
+    useTokens(tokens, type) {
+        if (tokens < 0)
+            throw new RangeError("Tokens cannot be negative");
+        else if (tokens === 0)
+            return;
+        if (type === "input")
+            this._inputTokens += tokens;
+        else if (type === "output")
+            this._outputTokens += tokens;
+        else if (type === "restoreState")
+            this._restoreStateTokens += tokens;
+        else {
+            void (type);
+            throw new TypeError(`Unknown token type: ${type}`);
+        }
+    }
+    /**
+     * Get the difference between the current meter and another meter
+     */
+    diff(meter) {
+        return TokenMeter.diff(this, meter);
+    }
+    /**
+     * Log the usage of tokens on multiple meters
+     */
+    static useTokens(meters, tokens, type) {
+        if (meters == null)
+            return;
+        if (meters instanceof TokenMeter)
+            meters.useTokens(tokens, type);
+        else {
+            for (const meter of meters)
+                meter.useTokens(tokens, type);
+        }
+    }
+    /**
+     * Get the difference between two meters
+     */
+    static diff(meter1, meter2) {
+        return {
+            usedInputTokens: meter1.usedInputTokens - meter2.usedInputTokens,
+            usedOutputTokens: meter1.usedOutputTokens - meter2.usedOutputTokens,
+            usedRestoreStateTokens: meter1.usedRestoreStateTokens - meter2.usedRestoreStateTokens
+        };
+    }
+}
+//# sourceMappingURL=TokenMeter.js.map

package/dist/evaluator/TokenMeter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"TokenMeter.js","sourceRoot":"","sources":["../../src/evaluator/TokenMeter.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,MAAM,OAAO,UAAU;IACX,YAAY,GAAW,CAAC,CAAC;IACzB,aAAa,GAAW,CAAC,CAAC;IAC1B,mBAAmB,GAAW,CAAC,CAAC;IAExC;;OAEG;IACH,IAAW,eAAe;QACtB,OAAO,IAAI,CAAC,YAAY,CAAC;IAC7B,CAAC;IAED;;OAEG;IACH,IAAW,gBAAgB;QACvB,OAAO,IAAI,CAAC,aAAa,CAAC;IAC9B,CAAC;IAED;;;OAGG;IACH,IAAW,sBAAsB;QAC7B,OAAO,IAAI,CAAC,mBAAmB,CAAC;IACpC,CAAC;IAED;;OAEG;IACI,QAAQ;QACX,OAAO;YACH,eAAe,EAAE,IAAI,CAAC,eAAe;YACrC,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;YACvC,sBAAsB,EAAE,IAAI,CAAC,sBAAsB;SACtD,CAAC;IACN,CAAC;IAED;;OAEG;IACI,SAAS,CAAC,MAAc,EAAE,IAAyC;QACtE,IAAI,MAAM,GAAG,CAAC;YACV,MAAM,IAAI,UAAU,CAAC,2BAA2B,CAAC,CAAC;aACjD,IAAI,MAAM,KAAK,CAAC;YACjB,OAAO;QAEX,IAAI,IAAI,KAAK,OAAO;YAChB,IAAI,CAAC,YAAY,IAAI,MAAM,CAAC;aAC3B,IAAI,IAAI,KAAK,QAAQ;YACtB,IAAI,CAAC,aAAa,IAAI,MAAM,CAAC;aAC5B,IAAI,IAAI,KAAK,cAAc;YAC5B,IAAI,CAAC,mBAAmB,IAAI,MAAM,CAAC;aAClC;YACD,KAAK,CAAC,IAAoB,CAAC,CAAC;YAC5B,MAAM,IAAI,SAAS,CAAC,uBAAuB,IAAI,EAAE,CAAC,CAAC;SACtD;IACL,CAAC;IAED;;OAEG;IACI,IAAI,CAAC,KAAmC;QAC3C,OAAO,UAAU,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IACxC,CAAC;IAED;;OAEG;IACI,MAAM,CAAC,SAAS,CACnB,MAAuF,EACvF,MAAc,EACd,IAAyC;QAEzC,IAAI,MAAM,IAAI,IAAI;YACd,OAAO;QAEX,IAAI,MAAM,YAAY,UAAU;YAC5B,MAAM,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;aAC9B;YACD,KAAK,MAAM,KAAK,IAAI,MAAM;gBACtB,KAAK,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;SACrC;IACL,CAAC;IAED;;OAEG;IACI,MAAM,CAAC,IAAI,CACd,MAAoC,EACpC,MAAoC;QAEpC,OAAO;YACH,eAAe,EAAE,MAAM,CAAC,eAAe,GAAG,MAAM,CAAC,eAAe;YAChE,gBAAgB,EAAE,MAAM,CAAC,gBAAgB,GAAG,MAAM,CAAC,gBAAgB;YACnE,sBAAsB,EAAE,MAAM,CAAC,sBAAsB,GAAG,MAAM,CAAC,sBAAsB;SACxF,CAAC;IACN,CAAC;CACJ"}

package/dist/gguf/GgufInsights.d.ts

@@ -0,0 +1,40 @@
+import { Llama } from "../bindings/Llama.js";
+import { GgufFileInfo } from "./types/GgufFileInfoTypes.js";
+export type GgufInsightsResourceRequirements = {
+    cpuRam: number;
+    gpuVram: number;
+};
+export declare class GgufInsights {
+    readonly ggufFileInfo: GgufFileInfo;
+    private constructor();
+    /** The context size the model was trained on */
+    get trainContextSize(): number | undefined;
+    /** The size of an embedding vector the model can produce */
+    get embeddingVectorSize(): number | undefined;
+    get totalLayers(): number;
+    get modelSize(): number;
+    estimateModelResourceRequirements({ gpuLayers }: {
+        gpuLayers: number;
+    }): GgufInsightsResourceRequirements;
+    /**
+     * Estimates the memory required to create a context of the given parameters based on the implementation details of `llama.cpp`.
+     * The calculation doesn't include a precise estimation of the graph overhead memory, so it uses a rough estimate for that.
+     * The estimation for the graph overhead memory will be improved in the future to be more precise, but it's good enough for now.
+     */
+    estimateContextResourceRequirements({ contextSize, modelGpuLayers, batchSize, sequences, isEmbeddingContext, includeGraphOverhead }: {
+        contextSize: number;
+        modelGpuLayers: number;
+        batchSize?: number;
+        sequences?: number;
+        isEmbeddingContext?: boolean;
+        includeGraphOverhead?: boolean;
+    }): GgufInsightsResourceRequirements;
+    /**
+     * @param ggufFileInfo
+     * @param llama - If you already have a `Llama` instance, pass it to reuse it for the `GgufInsights` instance.
+     * If you don't pass a `Llama` instance, a basic `Llama` instance is created as a fallback - it's a slim instance that
+     * doesn't instantiate a `llama.cpp` backend, so it won't utilize the GPU at all, and be shared with other `GgufInsights` instances
+     * that need a fallback `Llama` instance.
+     */
+    static from(ggufFileInfo: GgufFileInfo, llama?: Llama): Promise<GgufInsights>;
+}

package/dist/gguf/GgufInsights.js

@@ -0,0 +1,350 @@
+import { getLlamaWithoutBackend } from "../bindings/utils/getLlamaWithoutBackend.js";
+import { getDefaultContextBatchSize, getDefaultContextSequences } from "../evaluator/LlamaContext/LlamaContext.js";
+import { GgufArchitectureType } from "./types/GgufMetadataTypes.js";
+export class GgufInsights {
+    /** @internal */ _llama;
+    /** @internal */ _modelSize;
+    /** @internal */ _totalLayers = null;
+    ggufFileInfo;
+    constructor(ggufFileInfo, llama) {
+        this._llama = llama;
+        this.ggufFileInfo = ggufFileInfo;
+        this._modelSize = calculateTensorsSize(ggufFileInfo.tensorInfo ?? [], llama);
+    }
+    /** The context size the model was trained on */
+    get trainContextSize() {
+        return this.ggufFileInfo.architectureMetadata.context_length;
+    }
+    /** The size of an embedding vector the model can produce */
+    get embeddingVectorSize() {
+        return this.ggufFileInfo.architectureMetadata.embedding_length;
+    }
+    get totalLayers() {
+        if (this._totalLayers != null)
+            return this._totalLayers;
+        const outputLayers = 1;
+        this._totalLayers = this._getFileLayers() + outputLayers;
+        return this._totalLayers;
+    }
+    get modelSize() {
+        return this._modelSize;
+    }
+    estimateModelResourceRequirements({ gpuLayers }) {
+        const { cpu, gpu } = this._getTensorResourceSplit(gpuLayers);
+        return {
+            cpuRam: calculateTensorsSize(cpu, this._llama),
+            gpuVram: calculateTensorsSize(gpu, this._llama)
+        };
+    }
+    /**
+     * Estimates the memory required to create a context of the given parameters based on the implementation details of `llama.cpp`.
+     * The calculation doesn't include a precise estimation of the graph overhead memory, so it uses a rough estimate for that.
+     * The estimation for the graph overhead memory will be improved in the future to be more precise, but it's good enough for now.
+     */
+    estimateContextResourceRequirements({ contextSize, modelGpuLayers, batchSize, sequences, isEmbeddingContext = false, includeGraphOverhead = true }) {
+        if (sequences == null)
+            sequences = getDefaultContextSequences();
+        if (batchSize == null)
+            batchSize = getDefaultContextBatchSize({ contextSize, sequences });
+        const actualContextSize = contextSize * sequences;
+        const totalLayers = this.totalLayers;
+        const finalGpuLayers = Math.max(0, Math.min(modelGpuLayers ?? totalLayers, totalLayers));
+        const finalCpuLayers = totalLayers - finalGpuLayers;
+        const llmData = this.ggufFileInfo.architectureMetadata;
+        const vocabularySize = llmData.vocab_size ?? this.ggufFileInfo.metadata.tokenizer.ggml.tokens.length;
+        const logitsSize = vocabularySize * batchSize;
+        const embedSize = isEmbeddingContext
+            ? (llmData.embedding_length ?? 0) * batchSize
+            : 0;
+        const sizeTBytes = 8; // sizeof(size_t)
+        const floatBytes = 4; // sizeof(float)
+        const uint32TBytes = 4; // sizeof(uint32_t)
+        const int32TBytes = 4; // sizeof(int32_t)
+        // source: `llama_get_state_size` in `llama.cpp`
+        const sRngSize = sizeTBytes;
+        const sRng = this._llama._consts.llamaMaxRngState;
+        const sNOutputs = sizeTBytes;
+        const sNOutputPos = batchSize * int32TBytes;
+        const sLogitsSize = sizeTBytes;
+        const sLogits = logitsSize * floatBytes;
+        const sEmbeddingSize = sizeTBytes;
+        const sEmbedding = embedSize * floatBytes;
+        const sKvBufSize = sizeTBytes;
+        const sKvHead = uint32TBytes;
+        const sKvSize = uint32TBytes;
+        const sKvUsed = uint32TBytes;
+        const sKv = 2 * int32TBytes * modelGpuLayers * this._llama._consts.ggmlTensorOverhead;
+        const sKvCell = this._llama._consts.llamaPosSize + sizeTBytes + this._llama._consts.llamaSeqIdSize;
+        const kvSelfLength = this.ggufFileInfo.metadata.general.architecture === GgufArchitectureType.mamba
+            ? Math.max(1, sequences)
+            : actualContextSize;
+        const sKvCells = kvSelfLength * sKvCell;
+        const overheadMemory = (sRngSize +
+            sRng +
+            sNOutputs +
+            sNOutputPos +
+            sLogitsSize +
+            sLogits +
+            sEmbeddingSize +
+            sEmbedding +
+            sKvBufSize +
+            sKvHead +
+            sKvSize +
+            sKvUsed +
+            sKv +
+            sKvCells);
+        // Estimates the memory allocated by `ggml_backend_sched_reserve` in `llama_new_context_with_model` in `llama.cpp`.
+        // If you read this line and have better insights on how to estimate this memory, please open a PR to improve it :)
+        const estimateGraphOverheadMemory = () => {
+            const s1MB = Math.pow(1024, 2);
+            const tensorInfo = this.ggufFileInfo.tensorInfo ?? [];
+            let defaultCalculationAdjustment = 0;
+            if (batchSize == null)
+                return 0;
+            if (this.ggufFileInfo.metadata.general.architecture === GgufArchitectureType.llama) {
+                const expertCount = this.ggufFileInfo.architectureMetadata.expert_count ?? 0;
+                const headCount = this.ggufFileInfo.architectureMetadata.attention?.head_count ?? 0;
+                const embeddingLength = llmData.embedding_length ?? 0;
+                if (expertCount > 0) {
+                    const expertsUsedCount = this.ggufFileInfo.architectureMetadata.expert_used_count ?? 2;
+                    return int32TBytes * batchSize * (((expertsUsedCount + 1) * embeddingLength) + (actualContextSize * headCount));
+                }
+                return int32TBytes * batchSize * (embeddingLength + (actualContextSize * headCount));
+            }
+            else if (this.ggufFileInfo.metadata.general.architecture === GgufArchitectureType.qwen2) {
+                if (modelGpuLayers === this.totalLayers) {
+                    defaultCalculationAdjustment -= (s1MB * 340) * (this.trainContextSize == null
+                        ? 1
+                        : actualContextSize / this.trainContextSize);
+                }
+                else {
+                    defaultCalculationAdjustment -= (s1MB * 250) + ((s1MB * 50) * (this.trainContextSize == null
+                        ? 1
+                        : actualContextSize / this.trainContextSize));
+                }
+            }
+            else if (this.ggufFileInfo.metadata.general.architecture === GgufArchitectureType.gemma) {
+                // only works properly when all layers are on the GPU, which is why it's commented out:
+                // return int32TBytes * batchSize * ((llmData.embedding_length ?? 0));
+                if (modelGpuLayers === this.totalLayers) {
+                    defaultCalculationAdjustment += (s1MB * 40) - ((s1MB * 270) * (this.trainContextSize == null
+                        ? 1
+                        : actualContextSize / this.trainContextSize));
+                }
+                else {
+                    defaultCalculationAdjustment += -(s1MB * 550) + ((s1MB * 150) * (this.trainContextSize == null
+                        ? 1
+                        : Math.max(0, (1 - (actualContextSize / this.trainContextSize)))));
+                }
+            }
+            else if (this.ggufFileInfo.metadata.general.architecture === GgufArchitectureType.stablelm) {
+                const headCount = this.ggufFileInfo.architectureMetadata.attention?.head_count ?? 0;
+                return (int32TBytes * batchSize * actualContextSize * headCount) - (50 * s1MB);
+                // if (modelGpuLayers === this.totalLayers) {
+                //     defaultCalculationAdjustment += -(s1MB * 20) + (
+                //         (s1MB * 250) * (
+                //             this.trainContextSize == null
+                //                 ? 1
+                //                 : actualContextSize / this.trainContextSize
+                //         )
+                //     );
+                // } else {
+                //     defaultCalculationAdjustment += -(s1MB * 40) + (
+                //         (s1MB * 300) * (
+                //             this.trainContextSize == null
+                //                 ? 1
+                //                 : actualContextSize / this.trainContextSize
+                //         )
+                //     );
+                // }
+            }
+            const totalElements = tensorInfo.length === 0
+                ? this.totalLayers * (((llmData.embedding_length ?? 0) +
+                    (llmData.feed_forward_length ?? 0)) / 2)
+                : tensorInfo.reduce((res, tensor) => {
+                    return res + tensor.dimensions.reduce((res, dim) => res + Number(dim), 0);
+                }, 0);
+            // magic numbers for estimation. will be improved in the future
+            return (totalElements * 77.655 * (actualContextSize / 4096)) + defaultCalculationAdjustment;
+        };
+        const graphOverheadMemory = !includeGraphOverhead
+            ? 0
+            : estimateGraphOverheadMemory();
+        const usingGpu = finalGpuLayers !== 0;
+        const cpuRam = (!usingGpu
+            ? (overheadMemory + graphOverheadMemory)
+            : 0) +
+            this._estimateKvMemorySizeInBytes(actualContextSize, finalCpuLayers);
+        const gpuVram = usingGpu
+            ? (overheadMemory +
+                graphOverheadMemory +
+                this._estimateKvMemorySizeInBytes(actualContextSize, finalGpuLayers < totalLayers
+                    ? (finalGpuLayers + 1)
+                    : finalGpuLayers))
+            : 0;
+        return {
+            cpuRam,
+            gpuVram
+        };
+    }
+    /**
+     * Get the split tensor resources for CPU and GPU based on the number of GPU layers
+     * @internal
+     */
+    _getTensorResourceSplit(gpuLayers) {
+        const tensorInfo = this.ggufFileInfo.tensorInfo ?? [];
+        if (gpuLayers === 0) {
+            return {
+                cpu: tensorInfo,
+                gpu: []
+            };
+        }
+        const gpuTensors = [];
+        const cpuTensors = [];
+        for (const singleTensorInfo of tensorInfo) {
+            const { layerNumber } = parseTensorName(singleTensorInfo.name);
+            if (gpuLayers !== this.totalLayers) {
+                const architecture = this.ggufFileInfo.metadata?.general?.architecture;
+                if (architecture === GgufArchitectureType.qwen2 || architecture === GgufArchitectureType.gemma) {
+                    if (layerNumber != null && layerNumber < gpuLayers)
+                        gpuTensors.push(singleTensorInfo);
+                    else
+                        cpuTensors.push(singleTensorInfo);
+                    continue;
+                }
+            }
+            if (layerNumber == null || layerNumber < gpuLayers)
+                gpuTensors.push(singleTensorInfo);
+            else
+                cpuTensors.push(singleTensorInfo);
+        }
+        return {
+            cpu: cpuTensors,
+            gpu: gpuTensors
+        };
+    }
+    /** @internal */
+    _determineNumberOfLayersFromTensorInfo() {
+        const layerNumbers = new Set();
+        for (const singleTensorInfo of (this.ggufFileInfo.tensorInfo ?? [])) {
+            const { layerNumber } = parseTensorName(singleTensorInfo.name);
+            if (layerNumber != null)
+                layerNumbers.add(layerNumber);
+        }
+        return layerNumbers.size;
+    }
+    /** @internal */
+    _getFileLayers() {
+        return this.ggufFileInfo.architectureMetadata.block_count ?? this._determineNumberOfLayersFromTensorInfo();
+    }
+    /** @internal */
+    _estimateKvMemorySizeInBytes(contextSize, layers) {
+        // source: `llama_kv_cache_init` in `llama.cpp`
+        const nHead = this.ggufFileInfo.architectureMetadata.attention?.head_count ?? 0;
+        const nEmbd = this.ggufFileInfo.architectureMetadata.embedding_length ?? 0;
+        const nEmbdHeadK = this.ggufFileInfo.architectureMetadata.attention?.key_length ?? ((nHead == 0) ? 0 : (nEmbd / nHead));
+        const nHeadKv = this.ggufFileInfo.architectureMetadata.attention?.head_count_kv ?? nHead;
+        const modelNEmbdKGqa = nEmbdHeadK * nHeadKv;
+        const ssmDConv = this.ggufFileInfo.architectureMetadata.ssm?.conv_kernel ?? 0;
+        const ssmDInner = this.ggufFileInfo.architectureMetadata.ssm?.inner_size ?? 0;
+        const modelNEmbdKS = (ssmDConv > 0 ? (ssmDConv - 1) : 0) * ssmDInner;
+        const nEmbdHeadV = this.ggufFileInfo.architectureMetadata.attention?.value_length ?? ((nHead == 0) ? 0 : nEmbd / nHead);
+        const modelNEmbdVGqa = nEmbdHeadV * nHeadKv;
+        const ssmDState = this.ggufFileInfo.architectureMetadata.ssm?.state_size ?? 0;
+        const modelNEmbdVS = ssmDState * ssmDInner;
+        const totalNEmbdKGqa = modelNEmbdKGqa + modelNEmbdKS;
+        const totalNEmbdVGqa = modelNEmbdVGqa + modelNEmbdVS;
+        const keyTypeSize = this.ggufFileInfo.metadata.general.architecture === GgufArchitectureType.mamba
+            // if `type_k` of `llama_context_params` changes to be configurable in `LlamaContext`,
+            // this would have to depend on that value
+            ? this._llama._consts.ggmlTypeF32Size
+            : this._llama._consts.ggmlTypeF16Size;
+        const valueTypeSize = this.ggufFileInfo.metadata.general.architecture === GgufArchitectureType.mamba
+            // if `type_v` of `llama_context_params` changes to be configurable in `LlamaContext`,
+            // this would have to depend on that value
+            ? this._llama._consts.ggmlTypeF32Size
+            : this._llama._consts.ggmlTypeF16Size;
+        const keyTensorsSize = layers * totalNEmbdKGqa * contextSize * keyTypeSize;
+        const valueTensorsSize = layers * totalNEmbdVGqa * contextSize * valueTypeSize;
+        return keyTensorsSize + valueTensorsSize;
+    }
+    /**
+     * @param ggufFileInfo
+     * @param llama - If you already have a `Llama` instance, pass it to reuse it for the `GgufInsights` instance.
+     * If you don't pass a `Llama` instance, a basic `Llama` instance is created as a fallback - it's a slim instance that
+     * doesn't instantiate a `llama.cpp` backend, so it won't utilize the GPU at all, and be shared with other `GgufInsights` instances
+     * that need a fallback `Llama` instance.
+     */
+    static async from(ggufFileInfo, llama) {
+        let resolvedLlama = llama;
+        if (resolvedLlama == null)
+            resolvedLlama = await getLlamaWithoutBackend();
+        return new GgufInsights(ggufFileInfo, resolvedLlama);
+    }
+}
+function parseTensorName(tensorName) {
+    if (tensorName == null)
+        return { layerNumber: undefined };
+    const layerTensorPrefix = "blk.";
+    if (!tensorName.startsWith(layerTensorPrefix))
+        return { layerNumber: undefined };
+    const dotIndex = tensorName.indexOf(".", layerTensorPrefix.length);
+    const layerNumberString = tensorName.slice(layerTensorPrefix.length, dotIndex < 0
+        ? tensorName.length
+        : dotIndex);
+    const layerNumber = parseInt(layerNumberString);
+    if (Number.isFinite(layerNumber))
+        return { layerNumber };
+    return { layerNumber: undefined };
+}
+function calculateTensorsSize(tensorsInfo, llama) {
+    let size = 0;
+    for (const tensorInfo of tensorsInfo)
+        size += calculateTensorSize(tensorInfo, llama);
+    return size;
+}
+function calculateTensorSize(tensor, llama) {
+    const typeSize = llama._bindings.getTypeSizeForGgmlType(tensor.ggmlType);
+    const blockSize = llama._bindings.getBlockSizeForGgmlType(tensor.ggmlType);
+    const ggmlMaxDims = llama._consts.ggmlMaxDims;
+    if (typeSize == null || blockSize == null)
+        throw new Error("Invalid type or block size");
+    const { ne, nb } = getTensorNeAndNb(tensor, { typeSize, blockSize, ggmlMaxDims });
+    if (blockSize === 1) {
+        let totalBytes = typeSize;
+        for (let i = 0; i < ggmlMaxDims; i++) {
+            totalBytes += (ne[i] - 1) * nb[i];
+        }
+        return totalBytes;
+    }
+    else {
+        let totalBytes = Math.floor((ne[0] * nb[0]) / blockSize);
+        for (let i = 1; i < ggmlMaxDims; i++) {
+            totalBytes += (ne[i] - 1) * nb[i];
+        }
+        return totalBytes;
+    }
+}
+function getTensorNeAndNb(tensor, { typeSize, blockSize, ggmlMaxDims }) {
+    // number of elements
+    // source: `ggml_new_tensor_impl` in `ggml.c`
+    const ne = [
+        ...tensor.dimensions,
+        ...(Array(Math.max(0, ggmlMaxDims - tensor.dimensions.length)).fill(1))
+    ].slice(0, ggmlMaxDims);
+    // number of bytes
+    // source: `ggml_new_tensor_impl` in `ggml.c`
+    const nb = [
+        typeSize,
+        Math.floor(typeSize * (ne[0] / blockSize)),
+        ...Array(ggmlMaxDims - 2).fill(0)
+    ];
+    for (let i = 2; i < ggmlMaxDims; i++) {
+        nb[i] = nb[i - 1] * ne[i - 1];
+    }
+    return {
+        ne,
+        nb
+    };
+}
+//# sourceMappingURL=GgufInsights.js.map

package/dist/gguf/GgufInsights.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"GgufInsights.js","sourceRoot":"","sources":["../../src/gguf/GgufInsights.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,sBAAsB,EAAC,MAAM,6CAA6C,CAAC;AACnF,OAAO,EAAC,0BAA0B,EAAE,0BAA0B,EAAC,MAAM,2CAA2C,CAAC;AAGjH,OAAO,EAAC,oBAAoB,EAAC,MAAM,8BAA8B,CAAC;AAOlE,MAAM,OAAO,YAAY;IACrB,gBAAgB,CAAkB,MAAM,CAAQ;IAChD,gBAAgB,CAAkB,UAAU,CAAS;IACrD,gBAAgB,CAAS,YAAY,GAAkB,IAAI,CAAC;IAC5C,YAAY,CAAe;IAE3C,YAAoB,YAA0B,EAAE,KAAY;QACxD,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC;QACpB,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QAEjC,IAAI,CAAC,UAAU,GAAG,oBAAoB,CAAC,YAAY,CAAC,UAAU,IAAI,EAAE,EAAE,KAAK,CAAC,CAAC;IACjF,CAAC;IAED,gDAAgD;IAChD,IAAW,gBAAgB;QACvB,OAAO,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,cAAc,CAAC;IACjE,CAAC;IAED,4DAA4D;IAC5D,IAAW,mBAAmB;QAC1B,OAAO,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,gBAAgB,CAAC;IACnE,CAAC;IAED,IAAW,WAAW;QAClB,IAAI,IAAI,CAAC,YAAY,IAAI,IAAI;YACzB,OAAO,IAAI,CAAC,YAAY,CAAC;QAE7B,MAAM,YAAY,GAAG,CAAC,CAAC;QACvB,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,cAAc,EAAE,GAAG,YAAY,CAAC;QAEzD,OAAO,IAAI,CAAC,YAAY,CAAC;IAC7B,CAAC;IAED,IAAW,SAAS;QAChB,OAAO,IAAI,CAAC,UAAU,CAAC;IAC3B,CAAC;IAEM,iCAAiC,CAAC,EAAC,SAAS,EAAsB;QACrE,MAAM,EAAC,GAAG,EAAE,GAAG,EAAC,GAAG,IAAI,CAAC,uBAAuB,CAAC,SAAS,CAAC,CAAC;QAE3D,OAAO;YACH,MAAM,EAAE,oBAAoB,CAAC,GAAG,EAAE,IAAI,CAAC,MAAM,CAAC;YAC9C,OAAO,EAAE,oBAAoB,CAAC,GAAG,EAAE,IAAI,CAAC,MAAM,CAAC;SAClD,CAAC;IACN,CAAC;IAED;;;;OAIG;IACI,mCAAmC,CAAC,EACvC,WAAW,EAAE,cAAc,EAAE,SAAS,EAAE,SAAS,EAAE,kBAAkB,GAAG,KAAK,EAAE,oBAAoB,GAAG,IAAI,EAI7G;QACG,IAAI,SAAS,IAAI,IAAI;YAAE,SAAS,GAAG,0BAA0B,EAAE,CAAC;QAChE,IAAI,SAAS,IAAI,IAAI;YAAE,SAAS,GAAG,0BAA0B,CAAC,EAAC,WAAW,EAAE,SAAS,EAAC,CAAC,CAAC;QAExF,MAAM,iBAAiB,GAAG,WAAW,GAAG,SAAS,CAAC;QAElD,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,CAAC;QACrC,MAAM,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,cAAc,IAAI,WAAW,EAAE,WAAW,CAAC,CAAC,CAAC;QACzF,MAAM,cAAc,GAAG,WAAW,GAAG,cAAc,CAAC;QACpD,MAAM,OAAO,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC;QAEvD,MAAM,cAAc,GAAG,OAAO,CAAC,UAAU,IAAI,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC;QACrG,MAAM,UAAU,GAAG,cAAc,GAAG,SAAS,CAAC;QAC9C,MAAM,SAAS,GAAG,kBAAkB;YAChC,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB,IAAI,CAAC,CAAC,GAAG,SAAS;YAC7C,CAAC,CAAC,CAAC,CAAC;QAER,MAAM,UAAU,GAAG,CAAC,CAAC,CAAC,iBAAiB;QACvC,MAAM,UAAU,GAAG,CAAC,CAAC,CAAC,gBAAgB;QACtC,MAAM,YAAY,GAAG,CAAC,CAAC,CAAC,mBAAmB;QAC3C,MAAM,WAAW,GAAG,CAAC,CAAC,CAAC,kBAAkB;QAEzC,gDAAgD;QAChD,MAAM,QAAQ,GAAG,UAAU,CAAC;QAC5B,MAAM,IAAI,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,gBAAgB,CAAC;QAClD,MAAM,SAAS,GAAG,UAAU,CAAC;QAC7B,MAAM,WAAW,GAAG,SAAS,GAAG,WAAW,CAAC;QAC5C,MAAM,WAAW,GAAG,UAAU,CAAC;QAC/B,MAAM,OAAO,GAAG,UAAU,GAAG,UAAU,CAAC;QACxC,MAAM,cAAc,GAAG,UAAU,CAAC;QAClC,MAAM,UAAU,GAAG,SAAS,GAAG,UAAU,CAAC;QAC1C,MAAM,UAAU,GAAG,UAAU,CAAC;QAC9B,MAAM,OAAO,GAAG,YAAY,CAAC;QAC7B,MAAM,OAAO,GAAG,YAAY,CAAC;QAC7B,MAAM,OAAO,GAAG,YAAY,CAAC;QAC7B,MAAM,GAAG,GAAG,CAAC,GAAG,WAAW,GAAG,cAAc,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,kBAAkB,CAAC;QACtF,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,YAAY,GAAG,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,cAAc,CAAC;QACnG,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,YAAY,KAAK,oBAAoB,CAAC,KAAK;YAC/F,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,CAAC;YACxB,CAAC,CAAC,iBAAiB,CAAC;QACxB,MAAM,QAAQ,GAAG,YAAY,GAAG,OAAO,CAAC;QAExC,MAAM,cAAc,GAAG,CACnB,QAAQ;YACR,IAAI;YACJ,SAAS;YACT,WAAW;YACX,WAAW;YACX,OAAO;YACP,cAAc;YACd,UAAU;YACV,UAAU;YACV,OAAO;YACP,OAAO;YACP,OAAO;YACP,GAAG;YACH,QAAQ,CACX,CAAC;QAEF,mHAAmH;QACnH,mHAAmH;QACnH,MAAM,2BAA2B,GAAG,GAAG,EAAE;YACrC,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;YAC/B,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,CAAC,UAAU,IAAI,EAAE,CAAC;YAEtD,IAAI,4BAA4B,GAAG,CAAC,CAAC;YAErC,IAAI,SAAS,IAAI,IAAI;gBACjB,OAAO,CAAC,CAAC;YAEb,IAAI,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,YAAY,KAAK,oBAAoB,CAAC,KAAK,EAAE;gBAChF,MAAM,WAAW,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,YAAY,IAAI,CAAC,CAAC;gBAC7E,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,SAAS,EAAE,UAAU,IAAI,CAAC,CAAC;gBACpF,MAAM,eAAe,GAAG,OAAO,CAAC,gBAAgB,IAAI,CAAC,CAAC;gBAEtD,IAAI,WAAW,GAAG,CAAC,EAAE;oBACjB,MAAM,gBAAgB,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,iBAAiB,IAAI,CAAC,CAAC;oBAEvF,OAAO,WAAW,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,gBAAgB,GAAG,CAAC,CAAC,GAAG,eAAe,CAAC,GAAG,CAAC,iBAAiB,GAAG,SAAS,CAAC,CAAC,CAAC;iBACnH;gBAED,OAAO,WAAW,GAAG,SAAS,GAAG,CAAC,eAAe,GAAG,CAAC,iBAAiB,GAAG,SAAS,CAAC,CAAC,CAAC;aACxF;iBAAM,IAAI,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,YAAY,KAAK,oBAAoB,CAAC,KAAK,EAAE;gBACvF,IAAI,cAAc,KAAK,IAAI,CAAC,WAAW,EAAE;oBACrC,4BAA4B,IAAI,CAAC,IAAI,GAAG,GAAG,CAAC,GAAG,CAC3C,IAAI,CAAC,gBAAgB,IAAI,IAAI;wBACzB,CAAC,CAAC,CAAC;wBACH,CAAC,CAAC,iBAAiB,GAAG,IAAI,CAAC,gBAAgB,CAClD,CAAC;iBACL;qBAAM;oBACH,4BAA4B,IAAI,CAAC,IAAI,GAAG,GAAG,CAAC,GAAG,CAC3C,CAAC,IAAI,GAAG,EAAE,CAAC,GAAG,CACV,IAAI,CAAC,gBAAgB,IAAI,IAAI;wBACzB,CAAC,CAAC,CAAC;wBACH,CAAC,CAAC,iBAAiB,GAAG,IAAI,CAAC,gBAAgB,CAClD,CACJ,CAAC;iBACL;aACJ;iBAAM,IAAI,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,YAAY,KAAK,oBAAoB,CAAC,KAAK,EAAE;gBACvF,uFAAuF;gBACvF,sEAAsE;gBAEtE,IAAI,cAAc,KAAK,IAAI,CAAC,WAAW,EAAE;oBACrC,4BAA4B,IAAI,CAAC,IAAI,GAAG,EAAE,CAAC,GAAG,CAC1C,CAAC,IAAI,GAAG,GAAG,CAAC,GAAG,CACX,IAAI,CAAC,gBAAgB,IAAI,IAAI;wBACzB,CAAC,CAAC,CAAC;wBACH,CAAC,CAAC,iBAAiB,GAAG,IAAI,CAAC,gBAAgB,CAClD,CACJ,CAAC;iBACL;qBAAM;oBACH,4BAA4B,IAAI,CAAC,CAAC,IAAI,GAAG,GAAG,CAAC,GAAG,CAC5C,CAAC,IAAI,GAAG,GAAG,CAAC,GAAG,CACX,IAAI,CAAC,gBAAgB,IAAI,IAAI;wBACzB,CAAC,CAAC,CAAC;wBACH,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,iBAAiB,GAAG,IAAI,CAAC,gBAAgB,CAAC,CAAC,CAAC,CACvE,CACJ,CAAC;iBACL;aACJ;iBAAM,IAAI,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,YAAY,KAAK,oBAAoB,CAAC,QAAQ,EAAE;gBAC1F,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,SAAS,EAAE,UAAU,IAAI,CAAC,CAAC;gBAEpF,OAAO,CAAC,WAAW,GAAG,SAAS,GAAG,iBAAiB,GAAG,SAAS,CAAC,GAAG,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC;gBAE/E,6CAA6C;gBAC7C,uDAAuD;gBACvD,2BAA2B;gBAC3B,4CAA4C;gBAC5C,sBAAsB;gBACtB,8DAA8D;gBAC9D,YAAY;gBACZ,SAAS;gBACT,WAAW;gBACX,uDAAuD;gBACvD,2BAA2B;gBAC3B,4CAA4C;gBAC5C,sBAAsB;gBACtB,8DAA8D;gBAC9D,YAAY;gBACZ,SAAS;gBACT,IAAI;aACP;YAED,MAAM,aAAa,GAAG,UAAU,CAAC,MAAM,KAAK,CAAC;gBACzC,CAAC,CAAC,IAAI,CAAC,WAAW,GAAG,CACjB,CACI,CAAC,OAAO,CAAC,gBAAgB,IAAI,CAAC,CAAC;oBAC/B,CAAC,OAAO,CAAC,mBAAmB,IAAI,CAAC,CAAC,CACrC,GAAG,CAAC,CACR;gBACD,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE;oBAChC,OAAO,GAAG,GAAG,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,GAAW,EAAE,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;gBACtF,CAAC,EAAE,CAAC,CAAC,CAAC;YAEV,+DAA+D;YAC/D,OAAO,CAAC,aAAa,GAAG,MAAM,GAAG,CAAC,iBAAiB,GAAG,IAAI,CAAC,CAAC,GAAG,4BAA4B,CAAC;QAChG,CAAC,CAAC;QAEF,MAAM,mBAAmB,GAAG,CAAC,oBAAoB;YAC7C,CAAC,CAAC,CAAC;YACH,CAAC,CAAC,2BAA2B,EAAE,CAAC;QAEpC,MAAM,QAAQ,GAAG,cAAc,KAAK,CAAC,CAAC;QAEtC,MAAM,MAAM,GAAG,CACX,CAAC,QAAQ;YACL,CAAC,CAAC,CAAC,cAAc,GAAG,mBAAmB,CAAC;YACxC,CAAC,CAAC,CAAC,CACV;YACG,IAAI,CAAC,4BAA4B,CAAC,iBAAiB,EAAE,cAAc,CAAC,CAAC;QACzE,MAAM,OAAO,GAAG,QAAQ;YACpB,CAAC,CAAC,CACE,cAAc;gBACd,mBAAmB;gBACnB,IAAI,CAAC,4BAA4B,CAC7B,iBAAiB,EACjB,cAAc,GAAG,WAAW;oBACxB,CAAC,CAAC,CAAC,cAAc,GAAG,CAAC,CAAC;oBACtB,CAAC,CAAC,cAAc,CACvB,CACJ;YACD,CAAC,CAAC,CAAC,CAAC;QAER,OAAO;YACH,MAAM;YACN,OAAO;SACV,CAAC;IACN,CAAC;IAED;;;OAGG;IACI,uBAAuB,CAAC,SAAiB;QAI5C,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,CAAC,UAAU,IAAI,EAAE,CAAC;QAEtD,IAAI,SAAS,KAAK,CAAC,EAAE;YACjB,OAAO;gBACH,GAAG,EAAE,UAAU;gBACf,GAAG,EAAE,EAAE;aACV,CAAC;SACL;QAED,MAAM,UAAU,GAAqB,EAAE,CAAC;QACxC,MAAM,UAAU,GAAqB,EAAE,CAAC;QAExC,KAAK,MAAM,gBAAgB,IAAI,UAAU,EAAE;YACvC,MAAM,EAAC,WAAW,EAAC,GAAG,eAAe,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC;YAE7D,IAAI,SAAS,KAAK,IAAI,CAAC,WAAW,EAAE;gBAChC,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,CAAC,QAAQ,EAAE,OAAO,EAAE,YAAY,CAAC;gBAEvE,IAAI,YAAY,KAAK,oBAAoB,CAAC,KAAK,IAAI,YAAY,KAAK,oBAAoB,CAAC,KAAK,EAAE;oBAC5F,IAAI,WAAW,IAAI,IAAI,IAAI,WAAW,GAAG,SAAS;wBAC9C,UAAU,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;;wBAElC,UAAU,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;oBAEtC,SAAS;iBACZ;aACJ;YAED,IAAI,WAAW,IAAI,IAAI,IAAI,WAAW,GAAG,SAAS;gBAC9C,UAAU,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;;gBAElC,UAAU,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;SACzC;QAED,OAAO;YACH,GAAG,EAAE,UAAU;YACf,GAAG,EAAE,UAAU;SAClB,CAAC;IACN,CAAC;IAED,gBAAgB;IACT,sCAAsC;QACzC,MAAM,YAAY,GAAG,IAAI,GAAG,EAAU,CAAC;QAEvC,KAAK,MAAM,gBAAgB,IAAI,CAAC,IAAI,CAAC,YAAY,CAAC,UAAU,IAAI,EAAE,CAAC,EAAE;YACjE,MAAM,EAAC,WAAW,EAAC,GAAG,eAAe,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC;YAE7D,IAAI,WAAW,IAAI,IAAI;gBACnB,YAAY,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC;SACrC;QAED,OAAO,YAAY,CAAC,IAAI,CAAC;IAC7B,CAAC;IAED,gBAAgB;IACT,cAAc;QACjB,OAAO,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,WAAW,IAAI,IAAI,CAAC,sCAAsC,EAAE,CAAC;IAC/G,CAAC;IAED,gBAAgB;IACT,4BAA4B,CAAC,WAAmB,EAAE,MAAc;QACnE,+CAA+C;QAC/C,MAAM,KAAK,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,SAAS,EAAE,UAAU,IAAI,CAAC,CAAC;QAChF,MAAM,KAAK,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,gBAAgB,IAAI,CAAC,CAAC;QAC3E,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,SAAS,EAAE,UAAU,IAAI,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC;QACxH,MAAM,OAAO,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,SAAS,EAAE,aAAa,IAAI,KAAK,CAAC;QACzF,MAAM,cAAc,GAAG,UAAU,GAAG,OAAO,CAAC;QAE5C,MAAM,QAAQ,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,GAAG,EAAE,WAAW,IAAI,CAAC,CAAC;QAC9E,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,GAAG,EAAE,UAAU,IAAI,CAAC,CAAC;QAC9E,MAAM,YAAY,GAAG,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,SAAS,CAAC;QAErE,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,SAAS,EAAE,YAAY,IAAI,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC;QACxH,MAAM,cAAc,GAAG,UAAU,GAAG,OAAO,CAAC;QAE5C,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,GAAG,EAAE,UAAU,IAAI,CAAC,CAAC;QAC9E,MAAM,YAAY,GAAG,SAAS,GAAG,SAAS,CAAC;QAE3C,MAAM,cAAc,GAAG,cAAc,GAAG,YAAY,CAAC;QACrD,MAAM,cAAc,GAAG,cAAc,GAAG,YAAY,CAAC;QAErD,MAAM,WAAW,GAAG,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,YAAY,KAAK,oBAAoB,CAAC,KAAK;YAC9F,sFAAsF;YACtF,0CAA0C;YAC1C,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,eAAe;YACrC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC;QAC1C,MAAM,aAAa,GAAG,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,YAAY,KAAK,oBAAoB,CAAC,KAAK;YAChG,sFAAsF;YACtF,0CAA0C;YAC1C,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,eAAe;YACrC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC;QAE1C,MAAM,cAAc,GAAG,MAAM,GAAG,cAAc,GAAG,WAAW,GAAG,WAAW,CAAC;QAC3E,MAAM,gBAAgB,GAAG,MAAM,GAAG,cAAc,GAAG,WAAW,GAAG,aAAa,CAAC;QAE/E,OAAO,cAAc,GAAG,gBAAgB,CAAC;IAC7C,CAAC;IAED;;;;;;OAMG;IACI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,YAA0B,EAAE,KAAa;QAC9D,IAAI,aAAa,GAAG,KAAK,CAAC;QAC1B,IAAI,aAAa,IAAI,IAAI;YACrB,aAAa,GAAG,MAAM,sBAAsB,EAAE,CAAC;QAEnD,OAAO,IAAI,YAAY,CAAC,YAAY,EAAE,aAAa,CAAC,CAAC;IACzD,CAAC;CACJ;AAED,SAAS,eAAe,CAAC,UAAmB;IAGxC,IAAI,UAAU,IAAI,IAAI;QAClB,OAAO,EAAC,WAAW,EAAE,SAAS,EAAC,CAAC;IAEpC,MAAM,iBAAiB,GAAG,MAAM,CAAC;IACjC,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,iBAAiB,CAAC;QACzC,OAAO,EAAC,WAAW,EAAE,SAAS,EAAC,CAAC;IAEpC,MAAM,QAAQ,GAAG,UAAU,CAAC,OAAO,CAAC,GAAG,EAAE,iBAAiB,CAAC,MAAM,CAAC,CAAC;IACnE,MAAM,iBAAiB,GAAG,UAAU,CAAC,KAAK,CACtC,iBAAiB,CAAC,MAAM,EACxB,QAAQ,GAAG,CAAC;QACR,CAAC,CAAC,UAAU,CAAC,MAAM;QACnB,CAAC,CAAC,QAAQ,CACjB,CAAC;IAEF,MAAM,WAAW,GAAG,QAAQ,CAAC,iBAAiB,CAAC,CAAC;IAChD,IAAI,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAC;QAC5B,OAAO,EAAC,WAAW,EAAC,CAAC;IAEzB,OAAO,EAAC,WAAW,EAAE,SAAS,EAAC,CAAC;AACpC,CAAC;AAED,SAAS,oBAAoB,CAAC,WAA6B,EAAE,KAAY;IACrE,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,KAAK,MAAM,UAAU,IAAI,WAAW;QAChC,IAAI,IAAI,mBAAmB,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC;IAEnD,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,SAAS,mBAAmB,CAAC,MAAsB,EAAE,KAAY;IAC7D,MAAM,QAAQ,GAAG,KAAK,CAAC,SAAS,CAAC,sBAAsB,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACzE,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,CAAC,uBAAuB,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IAC3E,MAAM,WAAW,GAAG,KAAK,CAAC,OAAO,CAAC,WAAW,CAAC;IAE9C,IAAI,QAAQ,IAAI,IAAI,IAAI,SAAS,IAAI,IAAI;QACrC,MAAM,IAAI,KAAK,CAAC,4BAA4B,CAAC,CAAC;IAElD,MAAM,EAAC,EAAE,EAAE,EAAE,EAAC,GAAG,gBAAgB,CAAC,MAAM,EAAE,EAAC,QAAQ,EAAE,SAAS,EAAE,WAAW,EAAC,CAAC,CAAC;IAE9E,IAAI,SAAS,KAAK,CAAC,EAAE;QACjB,IAAI,UAAU,GAAG,QAAQ,CAAC;QAC1B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,WAAW,EAAE,CAAC,EAAE,EAAE;YAClC,UAAU,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;SACrC;QAED,OAAO,UAAU,CAAC;KACrB;SAAM;QACH,IAAI,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,GAAG,SAAS,CAAC,CAAC;QACzD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,WAAW,EAAE,CAAC,EAAE,EAAE;YAClC,UAAU,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;SACrC;QAED,OAAO,UAAU,CAAC;KACrB;AACL,CAAC;AAED,SAAS,gBAAgB,CAAC,MAAsB,EAAE,EAC9C,QAAQ,EAAE,SAAS,EAAE,WAAW,EAGnC;IACG,qBAAqB;IACrB,6CAA6C;IAC7C,MAAM,EAAE,GAAG;QACP,GAAG,MAAM,CAAC,UAAU;QACpB,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,WAAW,GAAG,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;KAC1E,CAAC,KAAK,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC;IAExB,kBAAkB;IAClB,6CAA6C;IAC7C,MAAM,EAAE,GAAG;QACP,QAAQ;QACR,IAAI,CAAC,KAAK,CAAC,QAAQ,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,SAAS,CAAC,CAAC;QAC1C,GAAG,KAAK,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;KACpC,CAAC;IACF,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,WAAW,EAAE,CAAC,EAAE,EAAE;QAClC,EAAE,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;KACjC;IAED,OAAO;QACH,EAAE;QACF,EAAE;KACL,CAAC;AACN,CAAC"}

package/dist/gguf/consts.d.ts

@@ -0,0 +1,3 @@
+import retry from "async-retry";
+export declare const ggufDefaultFetchRetryOptions: retry.Options;
+export declare const defaultExtraAllocationSize: number;

package/dist/gguf/consts.js

@@ -0,0 +1,8 @@
+export const ggufDefaultFetchRetryOptions = {
+    retries: 10,
+    factor: 2,
+    minTimeout: 1000,
+    maxTimeout: 1000 * 16
+};
+export const defaultExtraAllocationSize = 1024 * 1024 * 1.5; // 1.5MB
+//# sourceMappingURL=consts.js.map

package/dist/gguf/consts.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"consts.js","sourceRoot":"","sources":["../../src/gguf/consts.ts"],"names":[],"mappings":"AAEA,MAAM,CAAC,MAAM,4BAA4B,GAAkB;IACvD,OAAO,EAAE,EAAE;IACX,MAAM,EAAE,CAAC;IACT,UAAU,EAAE,IAAI;IAChB,UAAU,EAAE,IAAI,GAAG,EAAE;CACf,CAAC;AAEX,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,GAAG,IAAI,GAAG,GAAG,CAAC,CAAC,QAAQ"}

package/dist/gguf/errors/InvalidGgufMagicError.d.ts

@@ -0,0 +1,3 @@
+export declare class InvalidGgufMagicError extends Error {
+    constructor(expectedGgufMagic: string, actualGgufMagic: string);
+}

package/dist/gguf/errors/InvalidGgufMagicError.js

@@ -0,0 +1,6 @@
+export class InvalidGgufMagicError extends Error {
+    constructor(expectedGgufMagic, actualGgufMagic) {
+        super(`Invalid GGUF magic. Expected "${expectedGgufMagic}" but got "${actualGgufMagic}".`);
+    }
+}
+//# sourceMappingURL=InvalidGgufMagicError.js.map

package/dist/gguf/errors/InvalidGgufMagicError.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"InvalidGgufMagicError.js","sourceRoot":"","sources":["../../../src/gguf/errors/InvalidGgufMagicError.ts"],"names":[],"mappings":"AAAA,MAAM,OAAO,qBAAsB,SAAQ,KAAK;IAC5C,YAAmB,iBAAyB,EAAE,eAAuB;QACjE,KAAK,CAAC,iCAAiC,iBAAiB,cAAc,eAAe,IAAI,CAAC,CAAC;IAC/F,CAAC;CACJ"}

package/dist/gguf/errors/UnsupportedGgufValueTypeError.d.ts

@@ -0,0 +1,4 @@
+export declare class UnsupportedGgufValueTypeError extends Error {
+    readonly ggufValueType: number;
+    constructor(ggufValueType: number);
+}

package/dist/gguf/errors/UnsupportedGgufValueTypeError.js

@@ -0,0 +1,9 @@
+export class UnsupportedGgufValueTypeError extends Error {
+    ggufValueType;
+    constructor(ggufValueType) {
+        super(`Unsupported GGUF value type "${ggufValueType}"`);
+        Object.defineProperty(this, "ggufValueType", { enumerable: false });
+        this.ggufValueType = ggufValueType;
+    }
+}
+//# sourceMappingURL=UnsupportedGgufValueTypeError.js.map

package/dist/gguf/errors/UnsupportedGgufValueTypeError.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"UnsupportedGgufValueTypeError.js","sourceRoot":"","sources":["../../../src/gguf/errors/UnsupportedGgufValueTypeError.ts"],"names":[],"mappings":"AAAA,MAAM,OAAO,6BAA8B,SAAQ,KAAK;IACpC,aAAa,CAAS;IAEtC,YAAmB,aAAqB;QACpC,KAAK,CAAC,gCAAgC,aAAa,GAAG,CAAC,CAAC;QAExD,MAAM,CAAC,cAAc,CAAC,IAAI,EAAE,eAAoC,EAAE,EAAC,UAAU,EAAE,KAAK,EAAC,CAAC,CAAC;QAEvF,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;IACvC,CAAC;CACJ"}

package/dist/gguf/fileReaders/GgufFileReader.d.ts

@@ -0,0 +1,33 @@
+/// <reference types="node" />
+import { GgufReadOffset } from "../utils/GgufReadOffset.js";
+export declare const valueTypeToBytesToRead: {
+    readonly uint8: 1;
+    readonly uint16: 2;
+    readonly uint32: 4;
+    readonly uint64: 8;
+    readonly int8: 1;
+    readonly int16: 2;
+    readonly int32: 4;
+    readonly int64: 8;
+    readonly float32: 4;
+    readonly float64: 8;
+    readonly bool: 1;
+};
+export declare abstract class GgufFileReader {
+    protected _buffer: Buffer;
+    abstract readByteRange(offset: number | GgufReadOffset, length: number): Promise<Buffer>;
+    readUint8(offset: number | GgufReadOffset): Promise<number>;
+    readUint16(offset: number | GgufReadOffset): Promise<number>;
+    readUint32(offset: number | GgufReadOffset): Promise<number>;
+    readUint64(offset: number | GgufReadOffset): Promise<bigint>;
+    readInt8(offset: number | GgufReadOffset): Promise<number>;
+    readInt16(offset: number | GgufReadOffset): Promise<number>;
+    readInt32(offset: number | GgufReadOffset): Promise<number>;
+    readInt64(offset: number | GgufReadOffset): Promise<bigint>;
+    readFloat32(offset: number | GgufReadOffset): Promise<number>;
+    readFloat64(offset: number | GgufReadOffset): Promise<number>;
+    readBool(offset: number | GgufReadOffset): Promise<boolean>;
+    protected _addToBuffer(buffer: Buffer): void;
+    private _readByteRangeAndUpdateOffset;
+    static castNumberIfSafe(value: bigint): number | bigint;
+}