vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : update llama_rope_type

Browse source
This commit is contained in:
Georgi Gerganov 2024-02-24 11:38:00 +02:00
parent 42ddf4846c
commit 31e1ec928f
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
2 changed files with 34 additions and 25 deletions
Download patch file Download diff file Expand all files Collapse all files

57
llama.cpp
View file

@ -11757,31 +11757,40 @@ enum llama_vocab_type llama_vocab_type(const struct llama_model * model) {
enum llama_rope_type llama_rope_type(const struct llama_model * model) { enum llama_rope_type llama_rope_type(const struct llama_model * model) {
switch (model->arch) { switch (model->arch) {
case LLM_ARCH_LLAMA: return LLAMA_ROPE_TYPE; // these models do not use RoPE
case LLM_ARCH_FALCON: return LLAMA_ROPE_TYPE_NEOX; case LLM_ARCH_GPT2:
case LLM_ARCH_BAICHUAN: return LLAMA_ROPE_TYPE; case LLM_ARCH_GPTJ:
case LLM_ARCH_GPT2: return LLAMA_ROPE_TYPE_NONE; case LLM_ARCH_GPTNEOX:
case LLM_ARCH_GPTJ: return LLAMA_ROPE_TYPE_NONE; case LLM_ARCH_MPT:
case LLM_ARCH_GPTNEOX: return LLAMA_ROPE_TYPE_NONE; case LLM_ARCH_REFACT:
case LLM_ARCH_MPT: return LLAMA_ROPE_TYPE_NONE; case LLM_ARCH_BLOOM:
case LLM_ARCH_STARCODER: return LLAMA_ROPE_TYPE; return LLAMA_ROPE_TYPE_NONE;
case LLM_ARCH_PERSIMMON: return LLAMA_ROPE_TYPE_NEOX;
case LLM_ARCH_REFACT: return LLAMA_ROPE_TYPE_NONE; // use what we call a normal RoPE, operating on pairs of consecutive head values
case LLM_ARCH_BERT: return LLAMA_ROPE_TYPE_NEOX; case LLM_ARCH_LLAMA:
case LLM_ARCH_NOMIC_BERT: return LLAMA_ROPE_TYPE_NEOX; case LLM_ARCH_BAICHUAN:
case LLM_ARCH_BLOOM: return LLAMA_ROPE_TYPE_NONE; case LLM_ARCH_STARCODER:
case LLM_ARCH_STABLELM: return LLAMA_ROPE_TYPE_NEOX; case LLM_ARCH_PLAMO:
case LLM_ARCH_QWEN: return LLAMA_ROPE_TYPE_NEOX; case LLM_ARCH_CODESHELL:
case LLM_ARCH_QWEN2: return LLAMA_ROPE_TYPE_NEOX; case LLM_ARCH_ORION:
case LLM_ARCH_PHI2: return LLAMA_ROPE_TYPE_NEOX; case LLM_ARCH_INTERNLM2:
case LLM_ARCH_PLAMO: return LLAMA_ROPE_TYPE; case LLM_ARCH_MINICPM:
case LLM_ARCH_CODESHELL: return LLAMA_ROPE_TYPE; case LLM_ARCH_GEMMA:
case LLM_ARCH_ORION: return LLAMA_ROPE_TYPE; return LLAMA_ROPE_TYPE_NORM;
case LLM_ARCH_INTERNLM2: return LLAMA_ROPE_TYPE;
case LLM_ARCH_MINICPM: return LLAMA_ROPE_TYPE; // the pairs of head values are offset by n_rot/2
case LLM_ARCH_GEMMA: return LLAMA_ROPE_TYPE; case LLM_ARCH_FALCON:
case LLM_ARCH_PERSIMMON:
case LLM_ARCH_BERT:
case LLM_ARCH_NOMIC_BERT:
case LLM_ARCH_STABLELM:
case LLM_ARCH_QWEN:
case LLM_ARCH_QWEN2:
case LLM_ARCH_PHI2:
return LLAMA_ROPE_TYPE_NEOX;
// all model arches should be listed explicitly here
case LLM_ARCH_UNKNOWN: case LLM_ARCH_UNKNOWN:
default:
GGML_ASSERT(false && "unknown architecture"); GGML_ASSERT(false && "unknown architecture");
return LLAMA_ROPE_TYPE_NONE; return LLAMA_ROPE_TYPE_NONE;
} }

2
llama.h
View file

@ -66,7 +66,7 @@ extern "C" {
enum llama_rope_type { enum llama_rope_type {
LLAMA_ROPE_TYPE_NONE = -1, LLAMA_ROPE_TYPE_NONE = -1,
LLAMA_ROPE_TYPE = 0, LLAMA_ROPE_TYPE_NORM = 0,
LLAMA_ROPE_TYPE_NEOX = 2, LLAMA_ROPE_TYPE_NEOX = 2,
LLAMA_ROPE_TYPE_GLM = 4, LLAMA_ROPE_TYPE_GLM = 4,
}; };