@huggingface/gguf - npm Package Compare versions

Comparing version 0.1.2 to 0.1.3

dist/browser/index.d.ts

@@ -0,1 +1,150 @@
+/** This file is auto-generated by generate-llm.ts */
+
+declare const LLM_ARCHITECTURES: readonly ["llama", "falcon", "grok", "gpt2", "gptj", "gptneox", "mpt", "baichuan", "starcoder", "persimmon", "refact", "bert", "nomic-bert", "bloom", "stablelm", "qwen", "qwen2", "qwen2moe", "phi2", "phi3", "plamo", "codeshell", "orion", "internlm2", "minicpm", "gemma", "starcoder2", "mamba", "xverse", "command-r", "dbrx", "olmo"];
+type LLMArchitecture = (typeof LLM_ARCHITECTURES)[number];
+type Attention<TArchitecture extends string> = {
+    [K in `${TArchitecture}.attention.head_count`]: number;
+} & {
+    [K in `${TArchitecture}.attention.head_count_kv`]: number;
+} & {
+    [K in `${TArchitecture}.attention.layer_norm_epsilon`]: number;
+} & {
+    [K in `${TArchitecture}.attention.layer_norm_rms_epsilon`]: number;
+} & {
+    [K in `${TArchitecture}.attention.alibi_bias_max`]: number;
+} & {
+    [K in `${TArchitecture}.attention.clip_kqv`]: number;
+} & {
+    [K in `${TArchitecture}.attention.use_norm`]: number;
+};
+type Rope<TArchitecture extends LLMArchitecture> = {
+    [K in `${TArchitecture}.rope.dimension_count`]: number;
+} & {
+    [K in `${TArchitecture}.rope.freq_base`]: number;
+} & {
+    [K in `${TArchitecture}.rope.scale`]: number;
+} & {
+    [K in `${TArchitecture}.rope.scale_linear`]: number;
+};
+type MOE<TArchitecture extends LLMArchitecture> = {
+    [K in `${TArchitecture}.expert_count`]: number;
+} & {
+    [K in `${TArchitecture}.expert_used_count`]: number;
+};
+type TransformerLLMBase<TArchitecture extends LLMArchitecture> = ModelBase<TArchitecture> & MOE<TArchitecture> & Attention<TArchitecture> & Rope<TArchitecture>;
+declare enum TransformerLLMPoolingType {
+    UNSPECIFIED = -1,
+    NONE = 0,
+    MEAN = 1,
+    CLS = 2
+}
+type ArchLlama = TransformerLLMBase<"llama"> & {
+    "llama.attention.layer_norm_rms_epsilon": number;
+};
+type ArchFalcon = TransformerLLMBase<"falcon"> & {
+    "falcon.attention.layer_norm_epsilon": number;
+};
+type ArchGrok = TransformerLLMBase<"grok"> & {
+    "grok.attention.layer_norm_rms_epsilon": number;
+};
+type ArchGpt2 = TransformerLLMBase<"gpt2"> & {
+    "gpt2.attention.layer_norm_epsilon": number;
+};
+type ArchGptj = TransformerLLMBase<"gptj">;
+type ArchGptneox = TransformerLLMBase<"gptneox">;
+type ArchMpt = TransformerLLMBase<"mpt"> & {
+    "mpt.attention.layer_norm_epsilon": number;
+    "mpt.attention.clamp_kqv": number;
+    "mpt.attention.max_alibi_bias": number;
+};
+type ArchBaichuan = TransformerLLMBase<"baichuan"> & {
+    "baichuan.attention.layer_norm_rms_epsilon": number;
+};
+type ArchStarcoder = TransformerLLMBase<"starcoder"> & {
+    "starcoder.attention.layer_norm_epsilon": number;
+};
+type ArchPersimmon = TransformerLLMBase<"persimmon"> & {
+    "persimmon.attention.layer_norm_epsilon": number;
+};
+type ArchRefact = TransformerLLMBase<"refact"> & {
+    "refact.attention.layer_norm_rms_epsilon": number;
+};
+type ArchBert = TransformerLLMBase<"bert"> & {
+    "bert.attention.layer_norm_epsilon": number;
+    "bert.attention.causal": boolean;
+    "tokenizer.ggml.token_type_count": number;
+    "bert.pooling_type": TransformerLLMPoolingType;
+};
+type ArchNomicBert = TransformerLLMBase<"nomic-bert"> & {
+    "nomic-bert.attention.layer_norm_epsilon": number;
+    "nomic-bert.attention.causal": boolean;
+    "tokenizer.ggml.token_type_count": number;
+    "nomic-bert.pooling_type": TransformerLLMPoolingType;
+};
+type ArchBloom = TransformerLLMBase<"bloom"> & {
+    "bloom.attention.layer_norm_epsilon": number;
+};
+type ArchStablelm = TransformerLLMBase<"stablelm"> & {
+    "stablelm.attention.layer_norm_epsilon": number;
+};
+type ArchQwen = TransformerLLMBase<"qwen"> & {
+    "qwen.attention.layer_norm_rms_epsilon": number;
+};
+type ArchQwen2 = TransformerLLMBase<"qwen2"> & {
+    "qwen2.attention.layer_norm_rms_epsilon": number;
+};
+type ArchQwen2moe = TransformerLLMBase<"qwen2moe"> & {
+    "qwen2moe.attention.layer_norm_rms_epsilon": number;
+};
+type ArchPhi2 = TransformerLLMBase<"phi2"> & {
+    "phi2.attention.layer_norm_epsilon": number;
+};
+type ArchPhi3 = TransformerLLMBase<"phi3"> & {
+    "phi3.attention.layer_norm_rms_epsilon": number;
+};
+type ArchPlamo = TransformerLLMBase<"plamo"> & {
+    "plamo.attention.layer_norm_rms_epsilon": number;
+};
+type ArchCodeshell = TransformerLLMBase<"codeshell"> & {
+    "codeshell.attention.layer_norm_epsilon": number;
+};
+type ArchOrion = TransformerLLMBase<"orion"> & {
+    "orion.attention.layer_norm_epsilon": number;
+};
+type ArchInternlm2 = TransformerLLMBase<"internlm2"> & {
+    "internlm2.attention.layer_norm_rms_epsilon": number;
+};
+type ArchMinicpm = TransformerLLMBase<"minicpm"> & {
+    "minicpm.attention.layer_norm_rms_epsilon": number;
+};
+type ArchGemma = TransformerLLMBase<"gemma"> & {
+    "gemma.attention.layer_norm_rms_epsilon": number;
+};
+type ArchStarcoder2 = TransformerLLMBase<"starcoder2"> & {
+    "starcoder2.attention.layer_norm_epsilon": number;
+};
+type ArchMamba = TransformerLLMBase<"mamba"> & {
+    "mamba.ssm.conv_kernel": number;
+    "mamba.ssm.inner_size": number;
+    "mamba.ssm.state_size": number;
+    "mamba.ssm.time_step_rank": number;
+    "mamba.attention.layer_norm_rms_epsilon": number;
+};
+type ArchXverse = TransformerLLMBase<"xverse"> & {
+    "xverse.attention.layer_norm_rms_epsilon": number;
+};
+type ArchCommandR = TransformerLLMBase<"command-r"> & {
+    "command-r.logit_scale": number;
+    "command-r.attention.layer_norm_epsilon": number;
+};
+type ArchDbrx = TransformerLLMBase<"dbrx"> & {
+    "dbrx.attention.layer_norm_epsilon": number;
+    "dbrx.attention.clamp_kqv": number;
+};
+type ArchOlmo = TransformerLLMBase<"olmo"> & {
+    "olmo.attention.layer_norm_epsilon": number;
+    "olmo.attention.clamp_kqv": number;
+};
+type TransformerLLM = ArchLlama | ArchFalcon | ArchGrok | ArchGpt2 | ArchGptj | ArchGptneox | ArchMpt | ArchBaichuan | ArchStarcoder | ArchPersimmon | ArchRefact | ArchBert | ArchNomicBert | ArchBloom | ArchStablelm | ArchQwen | ArchQwen2 | ArchQwen2moe | ArchPhi2 | ArchPhi3 | ArchPlamo | ArchCodeshell | ArchOrion | ArchInternlm2 | ArchMinicpm | ArchGemma | ArchStarcoder2 | ArchMamba | ArchXverse | ArchCommandR | ArchDbrx | ArchOlmo;
+
 type MetadataBaseValue = string | number | bigint | boolean;
@@ -43,3 +192,3 @@ type MetadataValue = MetadataBaseValue | MetadataBaseValue[] | MetadataValue[];
 }
-declare const ARCHITECTURES: readonly ["llama", "mpt", "gptneox", "gptj", "gpt2", "bloom", "falcon", "gemma", "rwkv", "whisper"];
+declare const ARCHITECTURES: readonly ["llama", "falcon", "grok", "gpt2", "gptj", "gptneox", "mpt", "baichuan", "starcoder", "persimmon", "refact", "bert", "nomic-bert", "bloom", "stablelm", "qwen", "qwen2", "qwen2moe", "phi2", "phi3", "plamo", "codeshell", "orion", "internlm2", "minicpm", "gemma", "starcoder2", "mamba", "xverse", "command-r", "dbrx", "olmo", "rwkv", "whisper"];
 type Architecture = (typeof ARCHITECTURES)[number];
@@ -52,44 +201,16 @@ interface General {
 }
-type Attention<TArchitecture extends Architecture> = {
-    [K in `${TArchitecture}.attention.head_count`]: number;
-} | {
-    [K in `${TArchitecture}.attention.head_count_kv`]: number;
-} | {
-    [K in `${TArchitecture}.attention.layer_norm_epsilon`]: number;
-} | {
-    [K in `${TArchitecture}.attention.layer_norm_rms_epsilon`]: number;
-} | {
-    [K in `${TArchitecture}.attention.alibi_bias_max`]: number;
-} | {
-    [K in `${TArchitecture}.attention.clip_kqv`]: number;
-} | {
-    [K in `${TArchitecture}.attention.use_norm`]: number;
-};
-type Rope<TArchitecture extends Architecture> = {
-    [K in `${TArchitecture}.rope.dimension_count`]: number;
-} | {
-    [K in `${TArchitecture}.rope.freq_base`]: number;
-} | {
-    [K in `${TArchitecture}.rope.scale`]: number;
-} | {
-    [K in `${TArchitecture}.rope.scale_linear`]: number;
-};
 type ModelBase<TArchitecture extends Architecture | `encoder.${Extract<Architecture, "whisper">}` | `decoder.${Extract<Architecture, "whisper">}`> = {
     [K in `${TArchitecture}.layer_count`]: number;
-} | {
+} & {
     [K in `${TArchitecture}.feed_forward_length`]: number;
-} | {
+} & {
     [K in `${TArchitecture}.context_length`]: number;
-} | {
+} & {
     [K in `${TArchitecture}.embedding_length`]: number;
-} | {
+} & {
     [K in `${TArchitecture}.block_count`]: number;
 };
-type MOE<TArchitecture extends Architecture> = {
-    [K in `${TArchitecture}.expert_count`]: number;
-} | {
-    [K in `${TArchitecture}.expert_used_count`]: number;
-};
+type TokenizerModel = "no_vocab" | "llama" | "gpt2" | "bert";
 interface Tokenizer {
-    "tokenizer.ggml.model": Architecture;
+    "tokenizer.ggml.model": TokenizerModel;
     "tokenizer.ggml.tokens": string[];
@@ -103,4 +224,2 @@ "tokenizer.ggml.scores": number[];
 }
-type TransformerLLMArchitecture = Exclude<Architecture, "rwkv" | "whisper">;
-type TransformerLLM = ModelBase<TransformerLLMArchitecture> & MOE<TransformerLLMArchitecture> & Attention<TransformerLLMArchitecture> & Rope<TransformerLLMArchitecture>;
 type RWKV = ModelBase<"rwkv"> & {
@@ -147,2 +266,3 @@ "rwkv.architecture_version": number;
     fetch?: typeof fetch;
+    additionalFetchHeaders?: Record<string, string>;
     computeParametersCount: true;
@@ -157,2 +277,3 @@ }): Promise<GGUFParseOutput & {
     fetch?: typeof fetch;
+    additionalFetchHeaders?: Record<string, string>;
 }): Promise<GGUFParseOutput>;
@@ -169,2 +290,2 @@ declare function ggufAllShards(url: string, params?: {
 
-export { GGMLQuantizationType, GGUFMetadata, GGUFParseOutput, GGUFTensorInfo, GGUFValueType, GGUF_QUANT_DESCRIPTIONS, GgufShardFileInfo, MetadataBaseValue, MetadataValue, RE_GGUF_FILE, RE_GGUF_SHARD_FILE, Version, gguf, ggufAllShards, parseGgufShardFilename };
+export { Architecture, GGMLQuantizationType, GGUFMetadata, GGUFParseOutput, GGUFTensorInfo, GGUFValueType, GGUF_QUANT_DESCRIPTIONS, GgufShardFileInfo, MetadataBaseValue, MetadataValue, RE_GGUF_FILE, RE_GGUF_SHARD_FILE, Version, gguf, ggufAllShards, parseGgufShardFilename };

dist/browser/index.js

@@ -8,2 +8,38 @@ "use strict";Object.defineProperty(exports, "__esModule", {value: true});var __defProp = Object.defineProperty;
 
+// src/transformer-llm.ts
+var LLM_ARCHITECTURES = [
+  "llama",
+  "falcon",
+  "grok",
+  "gpt2",
+  "gptj",
+  "gptneox",
+  "mpt",
+  "baichuan",
+  "starcoder",
+  "persimmon",
+  "refact",
+  "bert",
+  "nomic-bert",
+  "bloom",
+  "stablelm",
+  "qwen",
+  "qwen2",
+  "qwen2moe",
+  "phi2",
+  "phi3",
+  "plamo",
+  "codeshell",
+  "orion",
+  "internlm2",
+  "minicpm",
+  "gemma",
+  "starcoder2",
+  "mamba",
+  "xverse",
+  "command-r",
+  "dbrx",
+  "olmo"
+];
+
 // src/types.ts
@@ -51,2 +87,3 @@ var GGMLQuantizationType = /* @__PURE__ */ ((GGMLQuantizationType2) => {
 })(GGUFValueType || {});
+var ARCHITECTURES = [...LLM_ARCHITECTURES, "rwkv", "whisper"];
 
@@ -205,3 +242,3 @@ // src/utils/promisesQueue.ts
   async fetchChunk() {
-    var _a, _b;
+    var _a, _b, _c, _d;
     const range = [this.chunk * HTTP_CHUNK_SIZE, (this.chunk + 1) * HTTP_CHUNK_SIZE - 1];
@@ -211,2 +248,3 @@ const buf = new Uint8Array(
         headers: {
+          ...(_d = (_c = this.params) == null ? void 0 : _c.additionalFetchHeaders) != null ? _d : {},
           Range: `bytes=${range[0]}-${range[1]}`
@@ -213,0 +251,0 @@ }

dist/index.d.ts

@@ -0,1 +1,150 @@
+/** This file is auto-generated by generate-llm.ts */
+
+declare const LLM_ARCHITECTURES: readonly ["llama", "falcon", "grok", "gpt2", "gptj", "gptneox", "mpt", "baichuan", "starcoder", "persimmon", "refact", "bert", "nomic-bert", "bloom", "stablelm", "qwen", "qwen2", "qwen2moe", "phi2", "phi3", "plamo", "codeshell", "orion", "internlm2", "minicpm", "gemma", "starcoder2", "mamba", "xverse", "command-r", "dbrx", "olmo"];
+type LLMArchitecture = (typeof LLM_ARCHITECTURES)[number];
+type Attention<TArchitecture extends string> = {
+    [K in `${TArchitecture}.attention.head_count`]: number;
+} & {
+    [K in `${TArchitecture}.attention.head_count_kv`]: number;
+} & {
+    [K in `${TArchitecture}.attention.layer_norm_epsilon`]: number;
+} & {
+    [K in `${TArchitecture}.attention.layer_norm_rms_epsilon`]: number;
+} & {
+    [K in `${TArchitecture}.attention.alibi_bias_max`]: number;
+} & {
+    [K in `${TArchitecture}.attention.clip_kqv`]: number;
+} & {
+    [K in `${TArchitecture}.attention.use_norm`]: number;
+};
+type Rope<TArchitecture extends LLMArchitecture> = {
+    [K in `${TArchitecture}.rope.dimension_count`]: number;
+} & {
+    [K in `${TArchitecture}.rope.freq_base`]: number;
+} & {
+    [K in `${TArchitecture}.rope.scale`]: number;
+} & {
+    [K in `${TArchitecture}.rope.scale_linear`]: number;
+};
+type MOE<TArchitecture extends LLMArchitecture> = {
+    [K in `${TArchitecture}.expert_count`]: number;
+} & {
+    [K in `${TArchitecture}.expert_used_count`]: number;
+};
+type TransformerLLMBase<TArchitecture extends LLMArchitecture> = ModelBase<TArchitecture> & MOE<TArchitecture> & Attention<TArchitecture> & Rope<TArchitecture>;
+declare enum TransformerLLMPoolingType {
+    UNSPECIFIED = -1,
+    NONE = 0,
+    MEAN = 1,
+    CLS = 2
+}
+type ArchLlama = TransformerLLMBase<"llama"> & {
+    "llama.attention.layer_norm_rms_epsilon": number;
+};
+type ArchFalcon = TransformerLLMBase<"falcon"> & {
+    "falcon.attention.layer_norm_epsilon": number;
+};
+type ArchGrok = TransformerLLMBase<"grok"> & {
+    "grok.attention.layer_norm_rms_epsilon": number;
+};
+type ArchGpt2 = TransformerLLMBase<"gpt2"> & {
+    "gpt2.attention.layer_norm_epsilon": number;
+};
+type ArchGptj = TransformerLLMBase<"gptj">;
+type ArchGptneox = TransformerLLMBase<"gptneox">;
+type ArchMpt = TransformerLLMBase<"mpt"> & {
+    "mpt.attention.layer_norm_epsilon": number;
+    "mpt.attention.clamp_kqv": number;
+    "mpt.attention.max_alibi_bias": number;
+};
+type ArchBaichuan = TransformerLLMBase<"baichuan"> & {
+    "baichuan.attention.layer_norm_rms_epsilon": number;
+};
+type ArchStarcoder = TransformerLLMBase<"starcoder"> & {
+    "starcoder.attention.layer_norm_epsilon": number;
+};
+type ArchPersimmon = TransformerLLMBase<"persimmon"> & {
+    "persimmon.attention.layer_norm_epsilon": number;
+};
+type ArchRefact = TransformerLLMBase<"refact"> & {
+    "refact.attention.layer_norm_rms_epsilon": number;
+};
+type ArchBert = TransformerLLMBase<"bert"> & {
+    "bert.attention.layer_norm_epsilon": number;
+    "bert.attention.causal": boolean;
+    "tokenizer.ggml.token_type_count": number;
+    "bert.pooling_type": TransformerLLMPoolingType;
+};
+type ArchNomicBert = TransformerLLMBase<"nomic-bert"> & {
+    "nomic-bert.attention.layer_norm_epsilon": number;
+    "nomic-bert.attention.causal": boolean;
+    "tokenizer.ggml.token_type_count": number;
+    "nomic-bert.pooling_type": TransformerLLMPoolingType;
+};
+type ArchBloom = TransformerLLMBase<"bloom"> & {
+    "bloom.attention.layer_norm_epsilon": number;
+};
+type ArchStablelm = TransformerLLMBase<"stablelm"> & {
+    "stablelm.attention.layer_norm_epsilon": number;
+};
+type ArchQwen = TransformerLLMBase<"qwen"> & {
+    "qwen.attention.layer_norm_rms_epsilon": number;
+};
+type ArchQwen2 = TransformerLLMBase<"qwen2"> & {
+    "qwen2.attention.layer_norm_rms_epsilon": number;
+};
+type ArchQwen2moe = TransformerLLMBase<"qwen2moe"> & {
+    "qwen2moe.attention.layer_norm_rms_epsilon": number;
+};
+type ArchPhi2 = TransformerLLMBase<"phi2"> & {
+    "phi2.attention.layer_norm_epsilon": number;
+};
+type ArchPhi3 = TransformerLLMBase<"phi3"> & {
+    "phi3.attention.layer_norm_rms_epsilon": number;
+};
+type ArchPlamo = TransformerLLMBase<"plamo"> & {
+    "plamo.attention.layer_norm_rms_epsilon": number;
+};
+type ArchCodeshell = TransformerLLMBase<"codeshell"> & {
+    "codeshell.attention.layer_norm_epsilon": number;
+};
+type ArchOrion = TransformerLLMBase<"orion"> & {
+    "orion.attention.layer_norm_epsilon": number;
+};
+type ArchInternlm2 = TransformerLLMBase<"internlm2"> & {
+    "internlm2.attention.layer_norm_rms_epsilon": number;
+};
+type ArchMinicpm = TransformerLLMBase<"minicpm"> & {
+    "minicpm.attention.layer_norm_rms_epsilon": number;
+};
+type ArchGemma = TransformerLLMBase<"gemma"> & {
+    "gemma.attention.layer_norm_rms_epsilon": number;
+};
+type ArchStarcoder2 = TransformerLLMBase<"starcoder2"> & {
+    "starcoder2.attention.layer_norm_epsilon": number;
+};
+type ArchMamba = TransformerLLMBase<"mamba"> & {
+    "mamba.ssm.conv_kernel": number;
+    "mamba.ssm.inner_size": number;
+    "mamba.ssm.state_size": number;
+    "mamba.ssm.time_step_rank": number;
+    "mamba.attention.layer_norm_rms_epsilon": number;
+};
+type ArchXverse = TransformerLLMBase<"xverse"> & {
+    "xverse.attention.layer_norm_rms_epsilon": number;
+};
+type ArchCommandR = TransformerLLMBase<"command-r"> & {
+    "command-r.logit_scale": number;
+    "command-r.attention.layer_norm_epsilon": number;
+};
+type ArchDbrx = TransformerLLMBase<"dbrx"> & {
+    "dbrx.attention.layer_norm_epsilon": number;
+    "dbrx.attention.clamp_kqv": number;
+};
+type ArchOlmo = TransformerLLMBase<"olmo"> & {
+    "olmo.attention.layer_norm_epsilon": number;
+    "olmo.attention.clamp_kqv": number;
+};
+type TransformerLLM = ArchLlama | ArchFalcon | ArchGrok | ArchGpt2 | ArchGptj | ArchGptneox | ArchMpt | ArchBaichuan | ArchStarcoder | ArchPersimmon | ArchRefact | ArchBert | ArchNomicBert | ArchBloom | ArchStablelm | ArchQwen | ArchQwen2 | ArchQwen2moe | ArchPhi2 | ArchPhi3 | ArchPlamo | ArchCodeshell | ArchOrion | ArchInternlm2 | ArchMinicpm | ArchGemma | ArchStarcoder2 | ArchMamba | ArchXverse | ArchCommandR | ArchDbrx | ArchOlmo;
+
 type MetadataBaseValue = string | number | bigint | boolean;
@@ -43,3 +192,3 @@ type MetadataValue = MetadataBaseValue | MetadataBaseValue[] | MetadataValue[];
 }
-declare const ARCHITECTURES: readonly ["llama", "mpt", "gptneox", "gptj", "gpt2", "bloom", "falcon", "gemma", "rwkv", "whisper"];
+declare const ARCHITECTURES: readonly ["llama", "falcon", "grok", "gpt2", "gptj", "gptneox", "mpt", "baichuan", "starcoder", "persimmon", "refact", "bert", "nomic-bert", "bloom", "stablelm", "qwen", "qwen2", "qwen2moe", "phi2", "phi3", "plamo", "codeshell", "orion", "internlm2", "minicpm", "gemma", "starcoder2", "mamba", "xverse", "command-r", "dbrx", "olmo", "rwkv", "whisper"];
 type Architecture = (typeof ARCHITECTURES)[number];
@@ -52,44 +201,16 @@ interface General {
 }
-type Attention<TArchitecture extends Architecture> = {
-    [K in `${TArchitecture}.attention.head_count`]: number;
-} | {
-    [K in `${TArchitecture}.attention.head_count_kv`]: number;
-} | {
-    [K in `${TArchitecture}.attention.layer_norm_epsilon`]: number;
-} | {
-    [K in `${TArchitecture}.attention.layer_norm_rms_epsilon`]: number;
-} | {
-    [K in `${TArchitecture}.attention.alibi_bias_max`]: number;
-} | {
-    [K in `${TArchitecture}.attention.clip_kqv`]: number;
-} | {
-    [K in `${TArchitecture}.attention.use_norm`]: number;
-};
-type Rope<TArchitecture extends Architecture> = {
-    [K in `${TArchitecture}.rope.dimension_count`]: number;
-} | {
-    [K in `${TArchitecture}.rope.freq_base`]: number;
-} | {
-    [K in `${TArchitecture}.rope.scale`]: number;
-} | {
-    [K in `${TArchitecture}.rope.scale_linear`]: number;
-};
 type ModelBase<TArchitecture extends Architecture | `encoder.${Extract<Architecture, "whisper">}` | `decoder.${Extract<Architecture, "whisper">}`> = {
     [K in `${TArchitecture}.layer_count`]: number;
-} | {
+} & {
     [K in `${TArchitecture}.feed_forward_length`]: number;
-} | {
+} & {
     [K in `${TArchitecture}.context_length`]: number;
-} | {
+} & {
     [K in `${TArchitecture}.embedding_length`]: number;
-} | {
+} & {
     [K in `${TArchitecture}.block_count`]: number;
 };
-type MOE<TArchitecture extends Architecture> = {
-    [K in `${TArchitecture}.expert_count`]: number;
-} | {
-    [K in `${TArchitecture}.expert_used_count`]: number;
-};
+type TokenizerModel = "no_vocab" | "llama" | "gpt2" | "bert";
 interface Tokenizer {
-    "tokenizer.ggml.model": Architecture;
+    "tokenizer.ggml.model": TokenizerModel;
     "tokenizer.ggml.tokens": string[];
@@ -103,4 +224,2 @@ "tokenizer.ggml.scores": number[];
 }
-type TransformerLLMArchitecture = Exclude<Architecture, "rwkv" | "whisper">;
-type TransformerLLM = ModelBase<TransformerLLMArchitecture> & MOE<TransformerLLMArchitecture> & Attention<TransformerLLMArchitecture> & Rope<TransformerLLMArchitecture>;
 type RWKV = ModelBase<"rwkv"> & {
@@ -147,2 +266,3 @@ "rwkv.architecture_version": number;
     fetch?: typeof fetch;
+    additionalFetchHeaders?: Record<string, string>;
     computeParametersCount: true;
@@ -157,2 +277,3 @@ }): Promise<GGUFParseOutput & {
     fetch?: typeof fetch;
+    additionalFetchHeaders?: Record<string, string>;
 }): Promise<GGUFParseOutput>;
@@ -169,2 +290,2 @@ declare function ggufAllShards(url: string, params?: {
 
-export { GGMLQuantizationType, GGUFMetadata, GGUFParseOutput, GGUFTensorInfo, GGUFValueType, GGUF_QUANT_DESCRIPTIONS, GgufShardFileInfo, MetadataBaseValue, MetadataValue, RE_GGUF_FILE, RE_GGUF_SHARD_FILE, Version, gguf, ggufAllShards, parseGgufShardFilename };
+export { Architecture, GGMLQuantizationType, GGUFMetadata, GGUFParseOutput, GGUFTensorInfo, GGUFValueType, GGUF_QUANT_DESCRIPTIONS, GgufShardFileInfo, MetadataBaseValue, MetadataValue, RE_GGUF_FILE, RE_GGUF_SHARD_FILE, Version, gguf, ggufAllShards, parseGgufShardFilename };

dist/index.js

@@ -34,2 +34,38 @@ "use strict";
 
+// src/transformer-llm.ts
+var LLM_ARCHITECTURES = [
+  "llama",
+  "falcon",
+  "grok",
+  "gpt2",
+  "gptj",
+  "gptneox",
+  "mpt",
+  "baichuan",
+  "starcoder",
+  "persimmon",
+  "refact",
+  "bert",
+  "nomic-bert",
+  "bloom",
+  "stablelm",
+  "qwen",
+  "qwen2",
+  "qwen2moe",
+  "phi2",
+  "phi3",
+  "plamo",
+  "codeshell",
+  "orion",
+  "internlm2",
+  "minicpm",
+  "gemma",
+  "starcoder2",
+  "mamba",
+  "xverse",
+  "command-r",
+  "dbrx",
+  "olmo"
+];
+
 // src/types.ts
@@ -77,2 +113,3 @@ var GGMLQuantizationType = /* @__PURE__ */ ((GGMLQuantizationType2) => {
 })(GGUFValueType || {});
+var ARCHITECTURES = [...LLM_ARCHITECTURES, "rwkv", "whisper"];
 
@@ -235,2 +272,3 @@ // src/utils/promisesQueue.ts
         headers: {
+          ...this.params?.additionalFetchHeaders ?? {},
           Range: `bytes=${range[0]}-${range[1]}`
@@ -237,0 +275,0 @@ }

package.json

 {
   "name": "@huggingface/gguf",
   "packageManager": "pnpm@8.10.5",
-  "version": "0.1.2",
+  "version": "0.1.3",
   "description": "a GGUF parser that works on remotely hosted files",
@@ -49,2 +49,3 @@ "repository": "https://github.com/huggingface/huggingface.js.git",
     "build": "tsup src/index.ts --format cjs,esm --clean --dts",
+    "build:llm": "tsx scripts/generate-llm.ts && pnpm run format",
     "test": "vitest run",
@@ -51,0 +52,0 @@ "check": "tsc"

src/gguf.ts

@@ -6,3 +6,3 @@ import type { MetadataValue, Version, GGUFMetadata, GGUFTensorInfo, GGUFParseOutput } from "./types";
 export type { MetadataBaseValue, MetadataValue, Version, GGUFMetadata, GGUFTensorInfo, GGUFParseOutput } from "./types";
-export { GGUFValueType, GGMLQuantizationType } from "./types";
+export { GGUFValueType, GGMLQuantizationType, Architecture } from "./types";
 export { GGUF_QUANT_DESCRIPTIONS } from "./quant-descriptions";
@@ -68,2 +68,3 @@
 			fetch?: typeof fetch;
+			additionalFetchHeaders?: Record<string, string>;
 		}
@@ -87,2 +88,3 @@ ) {
 					headers: {
+						...(this.params?.additionalFetchHeaders ?? {}),
 						Range: `bytes=${range[0]}-${range[1]}`,
@@ -215,2 +217,3 @@ },
 		fetch?: typeof fetch;
+		additionalFetchHeaders?: Record<string, string>;
 		computeParametersCount: true;
@@ -226,2 +229,3 @@ }
 		fetch?: typeof fetch;
+		additionalFetchHeaders?: Record<string, string>;
 	}
@@ -236,2 +240,3 @@ ): Promise<GGUFParseOutput>;
 		fetch?: typeof fetch;
+		additionalFetchHeaders?: Record<string, string>;
 		computeParametersCount?: boolean;
@@ -238,0 +243,0 @@ }

src/types.ts

@@ -0,1 +1,4 @@
+import type { TransformerLLM } from "./transformer-llm";
+import { LLM_ARCHITECTURES } from "./transformer-llm";
+
 export type MetadataBaseValue = string | number | bigint | boolean;
@@ -47,15 +50,3 @@ export type MetadataValue = MetadataBaseValue | MetadataBaseValue[] | MetadataValue[]; /// recursive as arrays can be nested.
 
-export const ARCHITECTURES = [
-	"llama",
-	"mpt",
-	"gptneox",
-	"gptj",
-	"gpt2",
-	"bloom",
-	"falcon",
-	"gemma",
-	"rwkv",
-	"whisper",
-] as const;
-
+const ARCHITECTURES = [...LLM_ARCHITECTURES, "rwkv", "whisper"] as const;
 export type Architecture = (typeof ARCHITECTURES)[number];
@@ -70,18 +61,3 @@
 
-type Attention<TArchitecture extends Architecture> =
-	| { [K in `${TArchitecture}.attention.head_count`]: number }
-	| { [K in `${TArchitecture}.attention.head_count_kv`]: number }
-	| { [K in `${TArchitecture}.attention.layer_norm_epsilon`]: number }
-	| { [K in `${TArchitecture}.attention.layer_norm_rms_epsilon`]: number }
-	| { [K in `${TArchitecture}.attention.alibi_bias_max`]: number }
-	| { [K in `${TArchitecture}.attention.clip_kqv`]: number }
-	| { [K in `${TArchitecture}.attention.use_norm`]: number };
-
-type Rope<TArchitecture extends Architecture> =
-	| { [K in `${TArchitecture}.rope.dimension_count`]: number }
-	| { [K in `${TArchitecture}.rope.freq_base`]: number }
-	| { [K in `${TArchitecture}.rope.scale`]: number }
-	| { [K in `${TArchitecture}.rope.scale_linear`]: number };
-
-type ModelBase<
+export type ModelBase<
 	TArchitecture extends
@@ -91,15 +67,9 @@ | Architecture
 		| `decoder.${Extract<Architecture, "whisper">}`,
-> =
-	| { [K in `${TArchitecture}.layer_count`]: number }
-	| { [K in `${TArchitecture}.feed_forward_length`]: number }
-	| { [K in `${TArchitecture}.context_length`]: number }
-	| { [K in `${TArchitecture}.embedding_length`]: number }
-	| { [K in `${TArchitecture}.block_count`]: number };
+> = { [K in `${TArchitecture}.layer_count`]: number } & { [K in `${TArchitecture}.feed_forward_length`]: number } & {
+	[K in `${TArchitecture}.context_length`]: number;
+} & { [K in `${TArchitecture}.embedding_length`]: number } & { [K in `${TArchitecture}.block_count`]: number };
 
-type MOE<TArchitecture extends Architecture> =
-	| { [K in `${TArchitecture}.expert_count`]: number }
-	| { [K in `${TArchitecture}.expert_used_count`]: number };
-
+type TokenizerModel = "no_vocab" | "llama" | "gpt2" | "bert";
 interface Tokenizer {
-	"tokenizer.ggml.model": Architecture;
+	"tokenizer.ggml.model": TokenizerModel;
 	"tokenizer.ggml.tokens": string[];
@@ -114,8 +84,2 @@ "tokenizer.ggml.scores": number[];
 
-type TransformerLLMArchitecture = Exclude<Architecture, "rwkv" | "whisper">;
-type TransformerLLM = ModelBase<TransformerLLMArchitecture> &
-	MOE<TransformerLLMArchitecture> &
-	Attention<TransformerLLMArchitecture> &
-	Rope<TransformerLLMArchitecture>;
-
 export type RWKV = ModelBase<"rwkv"> & { "rwkv.architecture_version": number };
@@ -122,0 +86,0 @@ export type LLM = TransformerLLM | RWKV;

New alerts

License Policy Violation

License

This package is not allowed per your license policy. Review the package's license to ensure compliance.

Found 1 instance in 1 package

Fixed alerts

License Policy Violation

License

This package is not allowed per your license policy. Review the package's license to ensure compliance.

Found 1 instance in 1 package

Improved metrics

Total package byte prevSize

140269

increased by15.28%

Number of package files

17

increased by6.25%

Lines of code

3391

increased by20.04%

No dependency changes