node-llama-cpp 3.0.0-beta.2 → 3.0.0-beta.20

package/dist/gguf/readGgufFileInfo.js

@@ -0,0 +1,66 @@
+import { isUrl } from "../utils/isUrl.js";
+import { parseGguf } from "./parser/parseGguf.js";
+import { GgufNetworkFetchFileReader } from "./fileReaders/GgufNetworkFetchFileReader.js";
+import { GgufFsFileReader } from "./fileReaders/GgufFsFileReader.js";
+import { ggufDefaultFetchRetryOptions } from "./consts.js";
+import { normalizeGgufDownloadUrl } from "./utils/normalizeGgufDownloadUrl.js";
+import { resolveSplitGgufParts } from "./utils/resolveSplitGgufParts.js";
+/**
+ * Read a GGUF file and return its metadata and tensor info (unless `readTensorInfo` is set to `false`).
+ * Only the parts of the file required for the metadata and tensor info are read.
+ */
+export async function readGgufFileInfo(pathOrUrl, { readTensorInfo = true, sourceType, ignoreKeys = [], logWarnings = true, fetchRetryOptions = ggufDefaultFetchRetryOptions, fetchHeaders = {}, spliceSplitFiles = true, signal } = {}) {
+    const useNetworkReader = sourceType === "network" || (sourceType == null && isUrl(pathOrUrl));
+    function createFileReader(pathOrUrl) {
+        if (useNetworkReader) {
+            return new GgufNetworkFetchFileReader({
+                url: normalizeGgufDownloadUrl(pathOrUrl),
+                retryOptions: fetchRetryOptions,
+                headers: fetchHeaders,
+                signal
+            });
+        }
+        else if (sourceType === "filesystem" || sourceType == null) {
+            return new GgufFsFileReader({
+                filePath: pathOrUrl,
+                signal
+            });
+        }
+        void (sourceType);
+        throw new Error(`Unsupported sourceType: ${sourceType}`);
+    }
+    async function readSingleFile(pathOrUrl) {
+        const fileReader = createFileReader(pathOrUrl);
+        return await parseGguf({
+            fileReader,
+            ignoreKeys,
+            readTensorInfo,
+            logWarnings
+        });
+    }
+    if (!spliceSplitFiles)
+        return await readSingleFile(pathOrUrl);
+    const allSplitPartPaths = resolveSplitGgufParts(pathOrUrl);
+    if (allSplitPartPaths.length === 1)
+        return await readSingleFile(allSplitPartPaths[0]);
+    const [first, ...rest] = await Promise.all(allSplitPartPaths.map((partPath) => readSingleFile(partPath)));
+    return {
+        version: first.version,
+        tensorCount: first.tensorCount,
+        metadata: first.metadata,
+        architectureMetadata: first.architectureMetadata,
+        tensorInfo: first.tensorInfo,
+        metadataSize: first.metadataSize,
+        splicedParts: allSplitPartPaths.length,
+        totalTensorInfoSize: first.totalTensorInfoSize == null
+            ? undefined
+            : (first.totalTensorInfoSize + rest.reduce((acc, part) => (acc + (part.totalTensorInfoSize ?? 0)), 0)),
+        totalTensorCount: Number(first.totalTensorCount) + rest.reduce((acc, part) => acc + Number(part.totalTensorCount), 0),
+        totalMetadataSize: first.totalMetadataSize + rest.reduce((acc, part) => acc + part.totalMetadataSize, 0),
+        fullTensorInfo: first.fullTensorInfo == null
+            ? undefined
+            : [first, ...rest].flatMap(part => (part.fullTensorInfo ?? [])),
+        tensorInfoSize: first.tensorInfoSize
+    };
+}
+//# sourceMappingURL=readGgufFileInfo.js.map

package/dist/gguf/readGgufFileInfo.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"readGgufFileInfo.js","sourceRoot":"","sources":["../../src/gguf/readGgufFileInfo.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,KAAK,EAAC,MAAM,mBAAmB,CAAC;AACxC,OAAO,EAAC,SAAS,EAAC,MAAM,uBAAuB,CAAC;AAChD,OAAO,EAAC,0BAA0B,EAAC,MAAM,6CAA6C,CAAC;AACvF,OAAO,EAAC,gBAAgB,EAAC,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAC,4BAA4B,EAAC,MAAM,aAAa,CAAC;AACzD,OAAO,EAAC,wBAAwB,EAAC,MAAM,qCAAqC,CAAC;AAC7E,OAAO,EAAC,qBAAqB,EAAC,MAAM,kCAAkC,CAAC;AAIvE;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,SAAiB,EAAE,EACtD,cAAc,GAAG,IAAI,EACrB,UAAU,EACV,UAAU,GAAG,EAAE,EACf,WAAW,GAAG,IAAI,EAClB,iBAAiB,GAAG,4BAA4B,EAChD,YAAY,GAAG,EAAE,EACjB,gBAAgB,GAAG,IAAI,EACvB,MAAM,KAiCN,EAAE;IACF,MAAM,gBAAgB,GAAG,UAAU,KAAK,SAAS,IAAI,CAAC,UAAU,IAAI,IAAI,IAAI,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC;IAE9F,SAAS,gBAAgB,CAAC,SAAiB;QACvC,IAAI,gBAAgB,EAAE;YAClB,OAAO,IAAI,0BAA0B,CAAC;gBAClC,GAAG,EAAE,wBAAwB,CAAC,SAAS,CAAC;gBACxC,YAAY,EAAE,iBAAiB;gBAC/B,OAAO,EAAE,YAAY;gBACrB,MAAM;aACT,CAAC,CAAC;SACN;aAAM,IAAI,UAAU,KAAK,YAAY,IAAI,UAAU,IAAI,IAAI,EAAE;YAC1D,OAAO,IAAI,gBAAgB,CAAC;gBACxB,QAAQ,EAAE,SAAS;gBACnB,MAAM;aACT,CAAC,CAAC;SACN;QAED,KAAK,CAAC,UAA0B,CAAC,CAAC;QAClC,MAAM,IAAI,KAAK,CAAC,2BAA2B,UAAU,EAAE,CAAC,CAAC;IAC7D,CAAC;IAED,KAAK,UAAU,cAAc,CAAC,SAAiB;QAC3C,MAAM,UAAU,GAAG,gBAAgB,CAAC,SAAS,CAAC,CAAC;QAC/C,OAAO,MAAM,SAAS,CAAC;YACnB,UAAU;YACV,UAAU;YACV,cAAc;YACd,WAAW;SACd,CAAC,CAAC;IACP,CAAC;IAED,IAAI,CAAC,gBAAgB;QACjB,OAAO,MAAM,cAAc,CAAC,SAAS,CAAC,CAAC;IAE3C,MAAM,iBAAiB,GAAG,qBAAqB,CAAC,SAAS,CAAC,CAAC;IAE3D,IAAI,iBAAiB,CAAC,MAAM,KAAK,CAAC;QAC9B,OAAO,MAAM,cAAc,CAAC,iBAAiB,CAAC,CAAC,CAAC,CAAC,CAAC;IAEtD,MAAM,CAAC,KAAK,EAAE,GAAG,IAAI,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CACtC,iBAAiB,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC,CAChE,CAAC;IAEF,OAAO;QACH,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,WAAW,EAAE,KAAK,CAAC,WAAW;QAC9B,QAAQ,EAAE,KAAK,CAAC,QAAQ;QACxB,oBAAoB,EAAE,KAAK,CAAC,oBAAoB;QAChD,UAAU,EAAE,KAAK,CAAC,UAAU;QAC5B,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,YAAY,EAAE,iBAAiB,CAAC,MAAM;QACtC,mBAAmB,EAAE,KAAK,CAAC,mBAAmB,IAAI,IAAI;YAClD,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,CAAC,KAAK,CAAC,mBAAmB,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,GAAG,CAAC,IAAI,CAAC,mBAAmB,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QAC1G,gBAAgB,EAAE,MAAM,CAAC,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,EAAE,CAAC,CAAC;QACrH,iBAAiB,EAAE,KAAK,CAAC,iBAAiB,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,iBAAiB,EAAE,CAAC,CAAC;QACxG,cAAc,EAAE,KAAK,CAAC,cAAc,IAAI,IAAI;YACxC,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,CAAC,KAAK,EAAE,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,cAAc,IAAI,EAAE,CAAC,CAAC;QACnE,cAAc,EAAE,KAAK,CAAC,cAAc;KAChB,CAAC;AAC7B,CAAC"}

package/dist/gguf/types/GgufFileInfoTypes.d.ts

@@ -0,0 +1,84 @@
+import type { GgufReadOffset } from "../utils/GgufReadOffset.js";
+import type { GgufFileReader } from "../fileReaders/GgufFileReader.js";
+import type { MergeOptionalUnionTypes } from "../../utils/mergeUnionTypes.js";
+import type { GgufArchitectureType, GgufMetadata } from "./GgufMetadataTypes.js";
+import type { GgufTensorInfo } from "./GgufTensorInfoTypes.js";
+export type MetadataValue = string | number | bigint | boolean | MetadataValue[];
+export type MetadataKeyValueRecord = Record<string, MetadataValue>;
+export type MetadataNestedObject = {
+    [key: string]: MetadataValue | MetadataNestedObject;
+};
+export type GgufFileInfo = {
+    readonly version: 2 | 3 | number;
+    readonly tensorCount: number | bigint;
+    readonly metadata: GgufMetadata;
+    readonly metadataSize: number;
+    /** Same value as `metadata[metadata.general.architecture]`, but with merged types for convenience */
+    readonly architectureMetadata: MergeOptionalUnionTypes<Exclude<GgufMetadata[GgufArchitectureType], undefined>>;
+    /** can be null if `readTensorInfo` is set to `false` */
+    readonly tensorInfo?: GgufTensorInfo[];
+    /** can be null if `readTensorInfo` is set to `false` */
+    readonly tensorInfoSize?: number;
+    /**
+     * For spliced metadata of multiple file parts,
+     * this will be the number of files parts read and spliced into this metadata.
+     *
+     * Whe no splicing is done, this will be `1`.
+     */
+    readonly splicedParts: number;
+    /**
+     * For spliced metadata of multiple file parts, this will be the total tensor count from all the parts
+     *
+     * When no splicing is done, this will be the same as `tensorCount`.
+     */
+    readonly totalTensorCount: number | bigint;
+    /**
+     * For spliced metadata of multiple file parts, this will be the total metadata size from all the parts
+     *
+     * When no splicing is done, this will be the same as `metadataSize`.
+     */
+    readonly totalMetadataSize: number;
+    /**
+     * For spliced metadata of multiple file parts, this will be the spliced tensorInfo from all the parts.
+     * Can be null if `readTensorInfo` is set to `false`
+     *
+     * When no splicing is done, this will be the same as `tensorInfo`.
+     */
+    readonly fullTensorInfo?: GgufTensorInfo[];
+    /**
+     * For spliced metadata of multiple file parts, this will be the total tensor info size from all the parts
+     *
+     * When no splicing is done, this will be the same as `tensorInfoSize`.
+     */
+    readonly totalTensorInfoSize?: number;
+};
+export declare const enum GgufValueType {
+    Uint8 = 0,
+    Int8 = 1,
+    Uint16 = 2,
+    Int16 = 3,
+    Uint32 = 4,
+    Int32 = 5,
+    Float32 = 6,
+    Bool = 7,
+    String = 8,
+    Array = 9,
+    Uint64 = 10,
+    Int64 = 11,
+    Float64 = 12
+}
+export type GgufVersionParserOptions = {
+    fileReader: GgufFileReader;
+    readTensorInfo?: boolean;
+    ignoreKeys?: string[];
+    version: number;
+    readOffset: GgufReadOffset;
+    logWarnings: boolean;
+};
+export type GgufVersionParserResult = {
+    tensorCount: number | bigint;
+    metadata: GgufMetadata;
+    tensorInfo?: GgufTensorInfo[];
+    metadataSize: number;
+    tensorInfoSize?: number;
+};

package/dist/gguf/types/GgufFileInfoTypes.js

@@ -0,0 +1,18 @@
+// source: `enum gguf_type` in `ggml.h` in the `llama.cpp` source code
+export var GgufValueType;
+(function (GgufValueType) {
+    GgufValueType[GgufValueType["Uint8"] = 0] = "Uint8";
+    GgufValueType[GgufValueType["Int8"] = 1] = "Int8";
+    GgufValueType[GgufValueType["Uint16"] = 2] = "Uint16";
+    GgufValueType[GgufValueType["Int16"] = 3] = "Int16";
+    GgufValueType[GgufValueType["Uint32"] = 4] = "Uint32";
+    GgufValueType[GgufValueType["Int32"] = 5] = "Int32";
+    GgufValueType[GgufValueType["Float32"] = 6] = "Float32";
+    GgufValueType[GgufValueType["Bool"] = 7] = "Bool";
+    GgufValueType[GgufValueType["String"] = 8] = "String";
+    GgufValueType[GgufValueType["Array"] = 9] = "Array";
+    GgufValueType[GgufValueType["Uint64"] = 10] = "Uint64";
+    GgufValueType[GgufValueType["Int64"] = 11] = "Int64";
+    GgufValueType[GgufValueType["Float64"] = 12] = "Float64";
+})(GgufValueType || (GgufValueType = {}));
+//# sourceMappingURL=GgufFileInfoTypes.js.map

package/dist/gguf/types/GgufFileInfoTypes.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"GgufFileInfoTypes.js","sourceRoot":"","sources":["../../../src/gguf/types/GgufFileInfoTypes.ts"],"names":[],"mappings":"AAkEA,sEAAsE;AACtE,MAAM,CAAN,IAAkB,aAcjB;AAdD,WAAkB,aAAa;IAC3B,mDAAS,CAAA;IACT,iDAAQ,CAAA;IACR,qDAAU,CAAA;IACV,mDAAS,CAAA;IACT,qDAAU,CAAA;IACV,mDAAS,CAAA;IACT,uDAAW,CAAA;IACX,iDAAQ,CAAA;IACR,qDAAU,CAAA;IACV,mDAAS,CAAA;IACT,sDAAW,CAAA;IACX,oDAAU,CAAA;IACV,wDAAY,CAAA;AAChB,CAAC,EAdiB,aAAa,KAAb,aAAa,QAc9B"}

package/dist/gguf/types/GgufMetadataTypes.d.ts

@@ -0,0 +1,334 @@
+export declare const enum GgufArchitectureType {
+    llama = "llama",
+    falcon = "falcon",
+    gpt2 = "gpt2",
+    gptj = "gptj",
+    gptneox = "gptneox",
+    mpt = "mpt",
+    baichuan = "baichuan",
+    starcoder = "starcoder",
+    persimmon = "persimmon",
+    refact = "refact",
+    bert = "bert",
+    nomicBert = "nomic-bert",
+    bloom = "bloom",
+    stablelm = "stablelm",
+    qwen = "qwen",
+    qwen2 = "qwen2",
+    phi2 = "phi2",
+    plamo = "plamo",
+    codeshell = "codeshell",
+    orion = "orion",
+    internlm2 = "internlm2",
+    minicpm = "minicpm",
+    gemma = "gemma",
+    starcoder2 = "starcoder2",
+    mamba = "mamba",
+    commandR = "command-r",
+    rwkv = "rwkv"
+}
+export type GgufMetadata<A extends GgufArchitectureType = GgufArchitectureType> = {
+    readonly general: GgufMetadataGeneral<A>;
+    readonly tokenizer: GgufMetadataTokenizer;
+} & (GgufArchitectureType extends A ? {
+    readonly [key in GgufArchitectureType]?: key extends keyof GgufMetadataLlmToType ? GgufMetadataLlmToType[key] : GgufMetadataDefaultArchitectureType;
+} : {
+    readonly [key in A]: key extends keyof GgufMetadataLlmToType ? GgufMetadataLlmToType[key] : GgufMetadataDefaultArchitectureType;
+});
+export type GgufMetadataLlmToType = {
+    [GgufArchitectureType.llama]: GgufMetadataLlmLLaMA;
+    [GgufArchitectureType.mpt]: GgufMetadataMPT;
+    [GgufArchitectureType.gptneox]: GgufMetadataGPTNeoX;
+    [GgufArchitectureType.gptj]: GgufMetadataGPTJ;
+    [GgufArchitectureType.gpt2]: GgufMetadataGPT2;
+    [GgufArchitectureType.bloom]: GgufMetadataBloom;
+    [GgufArchitectureType.falcon]: GgufMetadataFalcon;
+    [GgufArchitectureType.mamba]: GgufMetadataMamba;
+    [GgufArchitectureType.rwkv]: GgufMetadataRWKV;
+};
+export declare enum GgufFileType {
+    ALL_F32 = 0,
+    MOSTLY_F16 = 1,
+    MOSTLY_Q4_0 = 2,
+    MOSTLY_Q4_1 = 3,
+    MOSTLY_Q4_1_SOME_F16 = 4,
+    MOSTLY_Q4_2 = 5,
+    MOSTLY_Q4_3 = 6,
+    MOSTLY_Q8_0 = 7,
+    MOSTLY_Q5_0 = 8,
+    MOSTLY_Q5_1 = 9,
+    MOSTLY_Q2_K = 10,
+    MOSTLY_Q3_K_S = 11,
+    MOSTLY_Q3_K_M = 12,
+    MOSTLY_Q3_K_L = 13,
+    MOSTLY_Q4_K_S = 14,
+    MOSTLY_Q4_K_M = 15,
+    MOSTLY_Q5_K_S = 16,
+    MOSTLY_Q5_K_M = 17,
+    MOSTLY_Q6_K = 18,
+    MOSTLY_IQ2_XXS = 19,
+    MOSTLY_IQ2_XS = 20,
+    MOSTLY_Q2_K_S = 21,
+    MOSTLY_IQ3_XS = 22,
+    MOSTLY_IQ3_XXS = 23,
+    MOSTLY_IQ1_S = 24,
+    MOSTLY_IQ4_NL = 25,
+    MOSTLY_IQ3_S = 26,
+    MOSTLY_IQ3_M = 27,
+    MOSTLY_IQ2_S = 28,
+    MOSTLY_IQ2_M = 29,
+    MOSTLY_IQ4_XS = 30
+}
+export type GgufMetadataGeneral<A extends GgufArchitectureType = GgufArchitectureType> = {
+    readonly architecture: A;
+    /**
+     * The version of the quantization format. Not required if the model is not
+     * quantized (i.e. no tensors are quantized). If any tensors are quantized,
+     * this must be present. This is separate to the quantization scheme of the
+     * tensors itself; the quantization version may change without changing the
+     * scheme's name (e.g. the quantization scheme is Q5_K, and the quantization
+     * version is 4).
+     */
+    readonly quantization_version: string;
+    /**
+     * the global alignment to use, as described above. This can vary to allow
+     * for different alignment schemes, but it must be a multiple of 8. Some
+     * writers may not write the alignment. If the alignment is not specified,
+     * assume it is `32`.
+     */
+    readonly alignment?: string;
+    /**
+     * The name of the model. This should be a human-readable name that can be
+     * used to identify the model. It should be unique within the community
+     * that the model is defined in.
+     */
+    readonly name?: string;
+    readonly author?: string;
+    /**
+     * URL to the model's homepage. This can be a GitHub repo, a paper, etc.
+     */
+    readonly url?: string;
+    /**
+     * free-form description of the model including anything that isn't
+     * covered by the other fields
+     */
+    readonly description?: string;
+    /**
+     * License of the model, expressed as a SPDX license expression
+     * (e.g. `MIT OR Apache-2.0`). *Should not* include any other information,
+     * such as the license text or the URL to the license.
+     */
+    readonly license?: string;
+    /**
+     * Information about where this model came from. This is useful for tracking
+     * the provenance of the model, and for finding the original source if the
+     * model is modified. For a model that was converted from GGML, for
+     * example, these keys would point to the model that was converted from.
+     */
+    readonly source?: {
+        /**
+         * URL to the source of the model. Can be a GitHub repo, a paper, etc.
+         */
+        readonly url?: string;
+        readonly huggingface?: {
+            readonly repository?: string;
+        };
+    };
+    /**
+     * An enumerated value describing the type of the majority of the tensors
+     * in the file. Optional; can be inferred from the tensor types.
+     */
+    readonly file_type?: GgufFileType | undefined;
+};
+export declare const enum GgufMetadataTokenizerTokenType {
+    undefined = 0,
+    normal = 1,
+    unknown = 2,
+    control = 3,
+    userDefined = 4,
+    unused = 5,
+    byte = 6
+}
+export type GgufMetadataTokenizer = {
+    readonly ggml: {
+        readonly model: "no_vocab" | "llama" | "gpt2" | "bert" | "replit" | "rwkv" | string;
+        readonly tokens: readonly string[];
+        readonly token_type: GgufMetadataTokenizerTokenType[];
+        readonly token_type_count?: number;
+        readonly scores?: readonly number[];
+        readonly merges?: readonly string[];
+        readonly bos_token_id?: number;
+        readonly eos_token_id?: number;
+        readonly unknown_token_id?: number;
+        readonly separator_token_id?: number;
+        readonly padding_token_id?: number;
+        readonly add_bos_token?: boolean;
+        readonly add_eos_token?: boolean;
+        readonly add_space_prefix?: boolean;
+        readonly added_tokens?: readonly string[];
+        readonly prefix_token_id?: number;
+        readonly suffix_token_id?: number;
+        readonly middle_token_id?: number;
+        readonly eot_token_id?: number;
+    };
+    readonly huggingface?: {
+        readonly json?: string;
+    };
+    readonly chat_template?: string;
+};
+export declare const enum GgufMetadataArchitecturePoolingType {
+    unspecified = -1,
+    none = 0,
+    mean = 1,
+    max = 2
+}
+export type GgufMetadataDefaultArchitectureType = {
+    readonly vocab_size?: number;
+    readonly context_length?: number;
+    readonly embedding_length?: number;
+    readonly block_count?: number;
+    readonly feed_forward_length?: number;
+    readonly use_parallel_residual?: boolean;
+    readonly tensor_data_layout?: string;
+    readonly expert_count?: number;
+    readonly expert_used_count?: number;
+    readonly pooling_type?: GgufMetadataArchitecturePoolingType;
+    readonly logit_scale?: number;
+    readonly attention?: {
+        readonly head_count?: number;
+        readonly head_count_kv?: number;
+        readonly max_alibi_bias?: number;
+        readonly clamp_kqv?: number;
+        readonly layer_norm_epsilon?: number;
+        readonly layer_norm_rms_epsilon?: number;
+        readonly key_length?: number;
+        readonly value_length?: number;
+        readonly causal?: boolean;
+    };
+    readonly rope?: {
+        readonly dimension_count?: number;
+        readonly freq_base?: number;
+        readonly scale_linear?: number;
+        readonly scaling?: {
+            readonly type?: "none" | "linear" | "yarn" | string;
+            readonly factor?: number;
+            readonly original_context_length?: number;
+            readonly finetuned?: boolean;
+        };
+    };
+    readonly ssm?: {
+        readonly conv_kernel?: number;
+        readonly inner_size?: number;
+        readonly state_size?: number;
+        readonly time_step_rank?: number;
+    };
+};
+export type GgufMetadataLlmLLaMA = {
+    readonly context_length: number;
+    readonly embedding_length: number;
+    readonly block_count: number;
+    readonly feed_forward_length: number;
+    readonly attention: {
+        readonly head_count: number;
+        readonly layer_norm_rms_epsilon: number;
+        readonly head_count_kv?: number;
+    };
+    readonly rope: {
+        readonly dimension_count: number;
+        readonly scale?: number;
+    };
+    readonly expert_count?: number;
+    readonly expert_used_count?: number;
+    readonly tensor_data_layout?: string;
+};
+export type GgufMetadataMPT = {
+    readonly context_length: number;
+    readonly embedding_length: number;
+    readonly block_count: number;
+    readonly attention: {
+        readonly head_count: number;
+        readonly alibi_bias_max: number;
+        readonly clip_kqv: number;
+        readonly layer_norm_epsilon: number;
+    };
+};
+export type GgufMetadataGPTNeoX = {
+    readonly context_length: number;
+    readonly embedding_length: number;
+    readonly block_count: number;
+    readonly use_parallel_residual: boolean;
+    readonly rope: {
+        readonly dimension_count: number;
+        readonly scale?: number;
+    };
+    readonly attention: {
+        readonly head_count: number;
+        readonly layer_norm_epsilon: number;
+    };
+};
+export type GgufMetadataGPTJ = {
+    readonly context_length: number;
+    readonly embedding_length: number;
+    readonly block_count: number;
+    readonly rope: {
+        readonly dimension_count: number;
+        readonly scale?: number;
+    };
+    readonly attention: {
+        readonly head_count: number;
+        readonly layer_norm_epsilon: number;
+    };
+};
+export type GgufMetadataGPT2 = {
+    readonly context_length: number;
+    readonly embedding_length: number;
+    readonly block_count: number;
+    readonly attention: {
+        readonly head_count: number;
+        readonly layer_norm_epsilon: number;
+    };
+};
+export type GgufMetadataBloom = {
+    readonly context_length: number;
+    readonly embedding_length: number;
+    readonly block_count: number;
+    readonly feed_forward_length: number;
+    readonly attention: {
+        readonly head_count: number;
+        readonly layer_norm_epsilon: number;
+    };
+};
+export type GgufMetadataFalcon = {
+    readonly context_length: number;
+    readonly embedding_length: number;
+    readonly block_count: number;
+    readonly attention: {
+        readonly head_count: number;
+        readonly head_count_kv: number;
+        readonly use_norm: boolean;
+        readonly layer_norm_epsilon: number;
+    };
+    readonly tensor_data_layout?: string;
+};
+export type GgufMetadataMamba = {
+    readonly context_length: number;
+    readonly embedding_length: number;
+    readonly block_count: number;
+    readonly ssm: {
+        readonly conv_kernel: number;
+        readonly inner_size: number;
+        readonly state_size: number;
+        readonly time_step_rank: number;
+    };
+    readonly attention: {
+        readonly layer_norm_rms_epsilon: number;
+    };
+};
+export type GgufMetadataRWKV = {
+    readonly architecture_version: 4 | number;
+    readonly context_length: number;
+    readonly block_count: number;
+    readonly embedding_length: number;
+    readonly feed_forward_length: number;
+};
+export declare function isGgufMetadataOfArchitectureType<A extends GgufArchitectureType>(metadata: GgufMetadata, type: A): metadata is GgufMetadata<A>;

package/dist/gguf/types/GgufMetadataTypes.js

@@ -0,0 +1,86 @@
+export var GgufArchitectureType;
+(function (GgufArchitectureType) {
+    GgufArchitectureType["llama"] = "llama";
+    GgufArchitectureType["falcon"] = "falcon";
+    GgufArchitectureType["gpt2"] = "gpt2";
+    GgufArchitectureType["gptj"] = "gptj";
+    GgufArchitectureType["gptneox"] = "gptneox";
+    GgufArchitectureType["mpt"] = "mpt";
+    GgufArchitectureType["baichuan"] = "baichuan";
+    GgufArchitectureType["starcoder"] = "starcoder";
+    GgufArchitectureType["persimmon"] = "persimmon";
+    GgufArchitectureType["refact"] = "refact";
+    GgufArchitectureType["bert"] = "bert";
+    GgufArchitectureType["nomicBert"] = "nomic-bert";
+    GgufArchitectureType["bloom"] = "bloom";
+    GgufArchitectureType["stablelm"] = "stablelm";
+    GgufArchitectureType["qwen"] = "qwen";
+    GgufArchitectureType["qwen2"] = "qwen2";
+    GgufArchitectureType["phi2"] = "phi2";
+    GgufArchitectureType["plamo"] = "plamo";
+    GgufArchitectureType["codeshell"] = "codeshell";
+    GgufArchitectureType["orion"] = "orion";
+    GgufArchitectureType["internlm2"] = "internlm2";
+    GgufArchitectureType["minicpm"] = "minicpm";
+    GgufArchitectureType["gemma"] = "gemma";
+    GgufArchitectureType["starcoder2"] = "starcoder2";
+    GgufArchitectureType["mamba"] = "mamba";
+    GgufArchitectureType["commandR"] = "command-r";
+    GgufArchitectureType["rwkv"] = "rwkv";
+})(GgufArchitectureType || (GgufArchitectureType = {}));
+// source: `enum llama_ftype` in `llama.h` in the `llama.cpp` source code
+export var GgufFileType;
+(function (GgufFileType) {
+    GgufFileType[GgufFileType["ALL_F32"] = 0] = "ALL_F32";
+    GgufFileType[GgufFileType["MOSTLY_F16"] = 1] = "MOSTLY_F16";
+    GgufFileType[GgufFileType["MOSTLY_Q4_0"] = 2] = "MOSTLY_Q4_0";
+    GgufFileType[GgufFileType["MOSTLY_Q4_1"] = 3] = "MOSTLY_Q4_1";
+    GgufFileType[GgufFileType["MOSTLY_Q4_1_SOME_F16"] = 4] = "MOSTLY_Q4_1_SOME_F16";
+    GgufFileType[GgufFileType["MOSTLY_Q4_2"] = 5] = "MOSTLY_Q4_2";
+    GgufFileType[GgufFileType["MOSTLY_Q4_3"] = 6] = "MOSTLY_Q4_3";
+    GgufFileType[GgufFileType["MOSTLY_Q8_0"] = 7] = "MOSTLY_Q8_0";
+    GgufFileType[GgufFileType["MOSTLY_Q5_0"] = 8] = "MOSTLY_Q5_0";
+    GgufFileType[GgufFileType["MOSTLY_Q5_1"] = 9] = "MOSTLY_Q5_1";
+    GgufFileType[GgufFileType["MOSTLY_Q2_K"] = 10] = "MOSTLY_Q2_K";
+    GgufFileType[GgufFileType["MOSTLY_Q3_K_S"] = 11] = "MOSTLY_Q3_K_S";
+    GgufFileType[GgufFileType["MOSTLY_Q3_K_M"] = 12] = "MOSTLY_Q3_K_M";
+    GgufFileType[GgufFileType["MOSTLY_Q3_K_L"] = 13] = "MOSTLY_Q3_K_L";
+    GgufFileType[GgufFileType["MOSTLY_Q4_K_S"] = 14] = "MOSTLY_Q4_K_S";
+    GgufFileType[GgufFileType["MOSTLY_Q4_K_M"] = 15] = "MOSTLY_Q4_K_M";
+    GgufFileType[GgufFileType["MOSTLY_Q5_K_S"] = 16] = "MOSTLY_Q5_K_S";
+    GgufFileType[GgufFileType["MOSTLY_Q5_K_M"] = 17] = "MOSTLY_Q5_K_M";
+    GgufFileType[GgufFileType["MOSTLY_Q6_K"] = 18] = "MOSTLY_Q6_K";
+    GgufFileType[GgufFileType["MOSTLY_IQ2_XXS"] = 19] = "MOSTLY_IQ2_XXS";
+    GgufFileType[GgufFileType["MOSTLY_IQ2_XS"] = 20] = "MOSTLY_IQ2_XS";
+    GgufFileType[GgufFileType["MOSTLY_Q2_K_S"] = 21] = "MOSTLY_Q2_K_S";
+    GgufFileType[GgufFileType["MOSTLY_IQ3_XS"] = 22] = "MOSTLY_IQ3_XS";
+    GgufFileType[GgufFileType["MOSTLY_IQ3_XXS"] = 23] = "MOSTLY_IQ3_XXS";
+    GgufFileType[GgufFileType["MOSTLY_IQ1_S"] = 24] = "MOSTLY_IQ1_S";
+    GgufFileType[GgufFileType["MOSTLY_IQ4_NL"] = 25] = "MOSTLY_IQ4_NL";
+    GgufFileType[GgufFileType["MOSTLY_IQ3_S"] = 26] = "MOSTLY_IQ3_S";
+    GgufFileType[GgufFileType["MOSTLY_IQ3_M"] = 27] = "MOSTLY_IQ3_M";
+    GgufFileType[GgufFileType["MOSTLY_IQ2_S"] = 28] = "MOSTLY_IQ2_S";
+    GgufFileType[GgufFileType["MOSTLY_IQ2_M"] = 29] = "MOSTLY_IQ2_M";
+    GgufFileType[GgufFileType["MOSTLY_IQ4_XS"] = 30] = "MOSTLY_IQ4_XS";
+})(GgufFileType || (GgufFileType = {}));
+export var GgufMetadataTokenizerTokenType;
+(function (GgufMetadataTokenizerTokenType) {
+    GgufMetadataTokenizerTokenType[GgufMetadataTokenizerTokenType["undefined"] = 0] = "undefined";
+    GgufMetadataTokenizerTokenType[GgufMetadataTokenizerTokenType["normal"] = 1] = "normal";
+    GgufMetadataTokenizerTokenType[GgufMetadataTokenizerTokenType["unknown"] = 2] = "unknown";
+    GgufMetadataTokenizerTokenType[GgufMetadataTokenizerTokenType["control"] = 3] = "control";
+    GgufMetadataTokenizerTokenType[GgufMetadataTokenizerTokenType["userDefined"] = 4] = "userDefined";
+    GgufMetadataTokenizerTokenType[GgufMetadataTokenizerTokenType["unused"] = 5] = "unused";
+    GgufMetadataTokenizerTokenType[GgufMetadataTokenizerTokenType["byte"] = 6] = "byte";
+})(GgufMetadataTokenizerTokenType || (GgufMetadataTokenizerTokenType = {}));
+export var GgufMetadataArchitecturePoolingType;
+(function (GgufMetadataArchitecturePoolingType) {
+    GgufMetadataArchitecturePoolingType[GgufMetadataArchitecturePoolingType["unspecified"] = -1] = "unspecified";
+    GgufMetadataArchitecturePoolingType[GgufMetadataArchitecturePoolingType["none"] = 0] = "none";
+    GgufMetadataArchitecturePoolingType[GgufMetadataArchitecturePoolingType["mean"] = 1] = "mean";
+    GgufMetadataArchitecturePoolingType[GgufMetadataArchitecturePoolingType["max"] = 2] = "max";
+})(GgufMetadataArchitecturePoolingType || (GgufMetadataArchitecturePoolingType = {}));
+export function isGgufMetadataOfArchitectureType(metadata, type) {
+    return metadata?.general?.architecture === type;
+}
+//# sourceMappingURL=GgufMetadataTypes.js.map

package/dist/gguf/types/GgufMetadataTypes.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"GgufMetadataTypes.js","sourceRoot":"","sources":["../../../src/gguf/types/GgufMetadataTypes.ts"],"names":[],"mappings":"AAAA,MAAM,CAAN,IAAkB,oBA4BjB;AA5BD,WAAkB,oBAAoB;IAClC,uCAAe,CAAA;IACf,yCAAiB,CAAA;IACjB,qCAAa,CAAA;IACb,qCAAa,CAAA;IACb,2CAAmB,CAAA;IACnB,mCAAW,CAAA;IACX,6CAAqB,CAAA;IACrB,+CAAuB,CAAA;IACvB,+CAAuB,CAAA;IACvB,yCAAiB,CAAA;IACjB,qCAAa,CAAA;IACb,gDAAwB,CAAA;IACxB,uCAAe,CAAA;IACf,6CAAqB,CAAA;IACrB,qCAAa,CAAA;IACb,uCAAe,CAAA;IACf,qCAAa,CAAA;IACb,uCAAe,CAAA;IACf,+CAAuB,CAAA;IACvB,uCAAe,CAAA;IACf,+CAAuB,CAAA;IACvB,2CAAmB,CAAA;IACnB,uCAAe,CAAA;IACf,iDAAyB,CAAA;IACzB,uCAAe,CAAA;IACf,8CAAsB,CAAA;IACtB,qCAAa,CAAA;AACjB,CAAC,EA5BiB,oBAAoB,KAApB,oBAAoB,QA4BrC;AA+BD,yEAAyE;AACzE,MAAM,CAAN,IAAY,YAgCX;AAhCD,WAAY,YAAY;IACpB,qDAAW,CAAA;IACX,2DAAc,CAAA;IACd,6DAAe,CAAA;IACf,6DAAe,CAAA;IACf,+EAAwB,CAAA;IACxB,6DAAe,CAAA;IACf,6DAAe,CAAA;IACf,6DAAe,CAAA;IACf,6DAAe,CAAA;IACf,6DAAe,CAAA;IACf,8DAAgB,CAAA;IAChB,kEAAkB,CAAA;IAClB,kEAAkB,CAAA;IAClB,kEAAkB,CAAA;IAClB,kEAAkB,CAAA;IAClB,kEAAkB,CAAA;IAClB,kEAAkB,CAAA;IAClB,kEAAkB,CAAA;IAClB,8DAAgB,CAAA;IAChB,oEAAmB,CAAA;IACnB,kEAAkB,CAAA;IAClB,kEAAkB,CAAA;IAClB,kEAAkB,CAAA;IAClB,oEAAmB,CAAA;IACnB,gEAAiB,CAAA;IACjB,kEAAkB,CAAA;IAClB,gEAAiB,CAAA;IACjB,gEAAiB,CAAA;IACjB,gEAAiB,CAAA;IACjB,gEAAiB,CAAA;IACjB,kEAAkB,CAAA;AACtB,CAAC,EAhCW,YAAY,KAAZ,YAAY,QAgCvB;AAyED,MAAM,CAAN,IAAkB,8BAQjB;AARD,WAAkB,8BAA8B;IAC5C,6FAAa,CAAA;IACb,uFAAU,CAAA;IACV,yFAAW,CAAA;IACX,yFAAW,CAAA;IACX,iGAAe,CAAA;IACf,uFAAU,CAAA;IACV,mFAAQ,CAAA;AACZ,CAAC,EARiB,8BAA8B,KAA9B,8BAA8B,QAQ/C;AA8BD,MAAM,CAAN,IAAkB,mCAKjB;AALD,WAAkB,mCAAmC;IACjD,4GAAgB,CAAA;IAChB,6FAAQ,CAAA;IACR,6FAAQ,CAAA;IACR,2FAAO,CAAA;AACX,CAAC,EALiB,mCAAmC,KAAnC,mCAAmC,QAKpD;AAuND,MAAM,UAAU,gCAAgC,CAC5C,QAAsB,EAAE,IAAO;IAE/B,OAAO,QAAQ,EAAE,OAAO,EAAE,YAAY,KAAK,IAAI,CAAC;AACpD,CAAC"}

package/dist/gguf/types/GgufTensorInfoTypes.d.ts

@@ -0,0 +1,37 @@
+export type GgufTensorInfo = {
+    readonly name: string;
+    readonly dimensions: readonly (number | bigint)[];
+    readonly ggmlType: GgmlType;
+    readonly offset: number | bigint;
+};
+export declare const enum GgmlType {
+    F32 = 0,
+    F16 = 1,
+    Q4_0 = 2,
+    Q4_1 = 3,
+    Q4_2 = 4,
+    Q4_3 = 5,
+    Q5_0 = 6,
+    Q5_1 = 7,
+    Q8_0 = 8,
+    Q8_1 = 9,
+    Q2_K = 10,
+    Q3_K = 11,
+    Q4_K = 12,
+    Q5_K = 13,
+    Q6_K = 14,
+    Q8_K = 15,
+    IQ2_XXS = 16,
+    IQ2_XS = 17,
+    IQ3_XXS = 18,
+    IQ1_S = 19,
+    IQ4_NL = 20,
+    IQ3_S = 21,
+    IQ2_S = 22,
+    IQ4_XS = 23,
+    I8 = 24,
+    I16 = 25,
+    I32 = 26,
+    I64 = 27,
+    F64 = 28
+}

package/dist/gguf/types/GgufTensorInfoTypes.js

@@ -0,0 +1,33 @@
+export var GgmlType;
+(function (GgmlType) {
+    GgmlType[GgmlType["F32"] = 0] = "F32";
+    GgmlType[GgmlType["F16"] = 1] = "F16";
+    GgmlType[GgmlType["Q4_0"] = 2] = "Q4_0";
+    GgmlType[GgmlType["Q4_1"] = 3] = "Q4_1";
+    GgmlType[GgmlType["Q4_2"] = 4] = "Q4_2";
+    GgmlType[GgmlType["Q4_3"] = 5] = "Q4_3";
+    GgmlType[GgmlType["Q5_0"] = 6] = "Q5_0";
+    GgmlType[GgmlType["Q5_1"] = 7] = "Q5_1";
+    GgmlType[GgmlType["Q8_0"] = 8] = "Q8_0";
+    GgmlType[GgmlType["Q8_1"] = 9] = "Q8_1";
+    GgmlType[GgmlType["Q2_K"] = 10] = "Q2_K";
+    GgmlType[GgmlType["Q3_K"] = 11] = "Q3_K";
+    GgmlType[GgmlType["Q4_K"] = 12] = "Q4_K";
+    GgmlType[GgmlType["Q5_K"] = 13] = "Q5_K";
+    GgmlType[GgmlType["Q6_K"] = 14] = "Q6_K";
+    GgmlType[GgmlType["Q8_K"] = 15] = "Q8_K";
+    GgmlType[GgmlType["IQ2_XXS"] = 16] = "IQ2_XXS";
+    GgmlType[GgmlType["IQ2_XS"] = 17] = "IQ2_XS";
+    GgmlType[GgmlType["IQ3_XXS"] = 18] = "IQ3_XXS";
+    GgmlType[GgmlType["IQ1_S"] = 19] = "IQ1_S";
+    GgmlType[GgmlType["IQ4_NL"] = 20] = "IQ4_NL";
+    GgmlType[GgmlType["IQ3_S"] = 21] = "IQ3_S";
+    GgmlType[GgmlType["IQ2_S"] = 22] = "IQ2_S";
+    GgmlType[GgmlType["IQ4_XS"] = 23] = "IQ4_XS";
+    GgmlType[GgmlType["I8"] = 24] = "I8";
+    GgmlType[GgmlType["I16"] = 25] = "I16";
+    GgmlType[GgmlType["I32"] = 26] = "I32";
+    GgmlType[GgmlType["I64"] = 27] = "I64";
+    GgmlType[GgmlType["F64"] = 28] = "F64";
+})(GgmlType || (GgmlType = {}));
+//# sourceMappingURL=GgufTensorInfoTypes.js.map