gguf: sync type with upstream llama.cpp (huggingface#849)

Generated by `scripts/generate-llm.ts`

Author: ngxson

packages/gguf/scripts/generate-llm.ts

@@ -5,7 +5,7 @@
 
 import { writeFileSync } from "node:fs";
 
-const SOURCE_CPP_URL = "https://raw.githubusercontent.com/ggerganov/llama.cpp/master/llama.cpp";
+const SOURCE_CPP_URL = "https://raw.githubusercontent.com/ggerganov/llama.cpp/master/src/llama.cpp";
 const DEST_FILE_PATH = "./src/transformer-llm.ts";
 const DEST_COMMON_SOURCE = `
 /** This file is auto-generated by generate-llm.ts */
@@ -80,6 +80,20 @@ const KV_TYPE = {
 	LLM_KV_SSM_STATE_SIZE: "number",
 	LLM_KV_SSM_TIME_STEP_RANK: "number",
 	LLM_KV_LOGIT_SCALE: "number",
+	LLM_KV_EXPERT_FEED_FORWARD_LENGTH: "number",
+	LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH: "number",
+	LLM_KV_ATTENTION_SLIDING_WINDOW: "number",
+	LLM_KV_ATTN_LOGIT_SOFTCAPPING: "number",
+	LLM_KV_FINAL_LOGIT_SOFTCAPPING: "number",
+	LLM_KV_LEADING_DENSE_BLOCK_COUNT: "number",
+	LLM_KV_ATTENTION_KV_LORA_RANK: "number",
+	LLM_KV_EXPERT_SHARED_COUNT: "number",
+	LLM_KV_EXPERT_WEIGHTS_SCALE: "number",
+	LLM_KV_ROPE_SCALING_YARN_LOG_MUL: "number",
+	LLM_KV_ATTENTION_Q_LORA_RANK: "number",
+	LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT: "number",
+	LLM_KV_DECODER_START_TOKEN_ID: "number",
+	LLM_KV_USE_PARALLEL_RESIDUAL: "boolean",
 };
 
 interface Arch {
@@ -199,7 +213,12 @@ async function main() {
 			if (a.hparams.length) {
 				code += [
 					" & {",
-					...a.hparams.map((k) => `\t${JSON.stringify(constToKVName[k].replace("%s", a.name))}: ${KV_TYPE[k]},`),
+					...a.hparams.map((k) => {
+						if (!KV_TYPE[k]) {
+							throw new Error(`Cannot find type definition of ${k}`);
+						}
+						return `\t${JSON.stringify(constToKVName[k].replace("%s", a.name))}: ${KV_TYPE[k]},`;
+					}),
 					"};",
 				].join("\n");
 			} else {

packages/gguf/src/transformer-llm.ts

@@ -62,10 +62,10 @@ export const LLM_ARCHITECTURES = [
 	"mpt",
 	"baichuan",
 	"starcoder",
-	"persimmon",
 	"refact",
 	"bert",
 	"nomic-bert",
+	"jina-bert-v2",
 	"bloom",
 	"stablelm",
 	"qwen",
@@ -79,12 +79,21 @@ export const LLM_ARCHITECTURES = [
 	"internlm2",
 	"minicpm",
 	"gemma",
+	"gemma2",
 	"starcoder2",
 	"mamba",
 	"xverse",
 	"command-r",
 	"dbrx",
 	"olmo",
+	"openelm",
+	"arctic",
+	"deepseek2",
+	"chatglm",
+	"bitnet",
+	"t5",
+	"t5encoder",
+	"jais",
 ] as const;
 type LLMArchitecture = (typeof LLM_ARCHITECTURES)[number];
 export type ArchLlama = TransformerLLMBase<"llama"> & {
@@ -100,7 +109,10 @@ export type ArchGpt2 = TransformerLLMBase<"gpt2"> & {
 	"gpt2.attention.layer_norm_epsilon": number;
 };
 export type ArchGptj = TransformerLLMBase<"gptj">;
-export type ArchGptneox = TransformerLLMBase<"gptneox">;
+export type ArchGptneox = TransformerLLMBase<"gptneox"> & {
+	"gptneox.attention.layer_norm_epsilon": number;
+	"gptneox.use_parallel_residual": boolean;
+};
 export type ArchMpt = TransformerLLMBase<"mpt"> & {
 	"mpt.attention.layer_norm_epsilon": number;
 	"mpt.attention.clamp_kqv": number;
@@ -112,9 +124,6 @@ export type ArchBaichuan = TransformerLLMBase<"baichuan"> & {
 export type ArchStarcoder = TransformerLLMBase<"starcoder"> & {
 	"starcoder.attention.layer_norm_epsilon": number;
 };
-export type ArchPersimmon = TransformerLLMBase<"persimmon"> & {
-	"persimmon.attention.layer_norm_epsilon": number;
-};
 export type ArchRefact = TransformerLLMBase<"refact"> & {
 	"refact.attention.layer_norm_rms_epsilon": number;
 };
@@ -130,6 +139,12 @@ export type ArchNomicBert = TransformerLLMBase<"nomic-bert"> & {
 	"tokenizer.ggml.token_type_count": number;
 	"nomic-bert.pooling_type": TransformerLLMPoolingType;
 };
+export type ArchJinaBertV2 = TransformerLLMBase<"jina-bert-v2"> & {
+	"jina-bert-v2.attention.layer_norm_epsilon": number;
+	"jina-bert-v2.attention.causal": boolean;
+	"tokenizer.ggml.token_type_count": number;
+	"jina-bert-v2.pooling_type": TransformerLLMPoolingType;
+};
 export type ArchBloom = TransformerLLMBase<"bloom"> & {
 	"bloom.attention.layer_norm_epsilon": number;
 };
@@ -143,13 +158,16 @@ export type ArchQwen2 = TransformerLLMBase<"qwen2"> & {
 	"qwen2.attention.layer_norm_rms_epsilon": number;
 };
 export type ArchQwen2moe = TransformerLLMBase<"qwen2moe"> & {
+	"qwen2moe.expert_feed_forward_length": number;
+	"qwen2moe.expert_shared_feed_forward_length": number;
 	"qwen2moe.attention.layer_norm_rms_epsilon": number;
 };
 export type ArchPhi2 = TransformerLLMBase<"phi2"> & {
 	"phi2.attention.layer_norm_epsilon": number;
 };
 export type ArchPhi3 = TransformerLLMBase<"phi3"> & {
 	"phi3.attention.layer_norm_rms_epsilon": number;
+	"phi3.attention.sliding_window": number;
 };
 export type ArchPlamo = TransformerLLMBase<"plamo"> & {
 	"plamo.attention.layer_norm_rms_epsilon": number;
@@ -169,6 +187,12 @@ export type ArchMinicpm = TransformerLLMBase<"minicpm"> & {
 export type ArchGemma = TransformerLLMBase<"gemma"> & {
 	"gemma.attention.layer_norm_rms_epsilon": number;
 };
+export type ArchGemma2 = TransformerLLMBase<"gemma2"> & {
+	"gemma2.attention.sliding_window": number;
+	"gemma2.attention.layer_norm_rms_epsilon": number;
+	"gemma2.attn_logit_softcapping": number;
+	"gemma2.final_logit_softcapping": number;
+};
 export type ArchStarcoder2 = TransformerLLMBase<"starcoder2"> & {
 	"starcoder2.attention.layer_norm_epsilon": number;
 };
@@ -194,6 +218,41 @@ export type ArchOlmo = TransformerLLMBase<"olmo"> & {
 	"olmo.attention.layer_norm_epsilon": number;
 	"olmo.attention.clamp_kqv": number;
 };
+export type ArchOpenelm = TransformerLLMBase<"openelm"> & {
+	"openelm.attention.layer_norm_rms_epsilon": number;
+};
+export type ArchArctic = TransformerLLMBase<"arctic"> & {
+	"arctic.attention.layer_norm_rms_epsilon": number;
+};
+export type ArchDeepseek2 = TransformerLLMBase<"deepseek2"> & {
+	"deepseek2.attention.layer_norm_rms_epsilon": number;
+	"deepseek2.leading_dense_block_count": number;
+	"deepseek2.attention.q_lora_rank": number;
+	"deepseek2.attention.kv_lora_rank": number;
+	"deepseek2.expert_feed_forward_length": number;
+	"deepseek2.expert_shared_count": number;
+	"deepseek2.expert_weights_scale": number;
+	"deepseek2.rope.scaling.yarn_log_multiplier": number;
+};
+export type ArchChatglm = TransformerLLMBase<"chatglm"> & {
+	"chatglm.attention.layer_norm_rms_epsilon": number;
+};
+export type ArchBitnet = TransformerLLMBase<"bitnet"> & {
+	"bitnet.attention.layer_norm_rms_epsilon": number;
+};
+export type ArchT5 = TransformerLLMBase<"t5"> & {
+	"t5.attention.layer_norm_rms_epsilon": number;
+	"t5.attention.relative_buckets_count": number;
+	"t5.decoder_start_token_id": number;
+};
+export type ArchT5encoder = TransformerLLMBase<"t5encoder"> & {
+	"t5encoder.attention.layer_norm_rms_epsilon": number;
+	"t5encoder.attention.relative_buckets_count": number;
+};
+export type ArchJais = TransformerLLMBase<"jais"> & {
+	"jais.attention.layer_norm_epsilon": number;
+	"jais.attention.max_alibi_bias": number;
+};
 
 export type TransformerLLM =
 	| ArchLlama
@@ -205,10 +264,10 @@ export type TransformerLLM =
 	| ArchMpt
 	| ArchBaichuan
 	| ArchStarcoder
-	| ArchPersimmon
 	| ArchRefact
 	| ArchBert
 	| ArchNomicBert
+	| ArchJinaBertV2
 	| ArchBloom
 	| ArchStablelm
 	| ArchQwen
@@ -222,9 +281,18 @@ export type TransformerLLM =
 	| ArchInternlm2
 	| ArchMinicpm
 	| ArchGemma
+	| ArchGemma2
 	| ArchStarcoder2
 	| ArchMamba
 	| ArchXverse
 	| ArchCommandR
 	| ArchDbrx
-	| ArchOlmo;
+	| ArchOlmo
+	| ArchOpenelm
+	| ArchArctic
+	| ArchDeepseek2
+	| ArchChatglm
+	| ArchBitnet
+	| ArchT5
+	| ArchT5encoder
+	| ArchJais;