CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-03 09:22:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : support InternLM2 (#5184)

Browse Source 
* support InternLM2 inference
  * add add_space_prefix KV pair
This commit is contained in:
Guoteng
2024-02-01 17:19:51 +08:00
committed by GitHub
parent 1cfb5372cf
commit ce32060198
5 changed files with 387 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
gguf-py/gguf/constants.py
View File

@@ -72,6 +72,7 @@ class Keys:
PAD_ID = "tokenizer.ggml.padding_token_id"
ADD_BOS = "tokenizer.ggml.add_bos_token"
ADD_EOS = "tokenizer.ggml.add_eos_token"
ADD_PREFIX = "tokenizer.ggml.add_space_prefix"
HF_JSON = "tokenizer.huggingface.json"
RWKV = "tokenizer.rwkv.world"
CHAT_TEMPLATE = "tokenizer.chat_template"
@@ -102,6 +103,7 @@ class MODEL_ARCH(IntEnum):
PLAMO = auto()
CODESHELL = auto()
ORION = auto()
INTERNLM2 = auto()
class MODEL_TENSOR(IntEnum):
@@ -153,6 +155,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.PLAMO: "plamo",
MODEL_ARCH.CODESHELL: "codeshell",
MODEL_ARCH.ORION: "orion",
MODEL_ARCH.INTERNLM2: "internlm2",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@@ -446,6 +449,21 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.INTERNLM2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_ROT_EMBD,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
# TODO
}

3
gguf-py/gguf/gguf_writer.py
View File

@@ -411,6 +411,9 @@ class GGUFWriter:
def add_add_eos_token(self, value: bool) -> None:
self.add_bool(Keys.Tokenizer.ADD_EOS, value)
def add_add_space_prefix(self, value: bool) -> None:
self.add_bool(Keys.Tokenizer.ADD_PREFIX, value)
def add_chat_template(self, value: str) -> None:
self.add_string(Keys.Tokenizer.CHAT_TEMPLATE, value)

14
gguf-py/gguf/tensor_mapping.py
View File

@@ -19,6 +19,7 @@ class TensorNameMap:
"language_model.embedding.word_embeddings", # persimmon
"wte", # gpt2
"transformer.embd.wte", # phi2
"model.tok_embeddings", # internlm2
),
# Token type embeddings
@@ -42,7 +43,7 @@ class TensorNameMap:
MODEL_TENSOR.OUTPUT: (
"embed_out", # gptneox
"lm_head", # gpt2 mpt falcon llama-hf baichuan qwen
"output", # llama-pth bloom
"output", # llama-pth bloom internlm2
"word_embeddings_for_head", # persimmon
"lm_head.linear", # phi2
),
@@ -51,7 +52,7 @@ class TensorNameMap:
MODEL_TENSOR.OUTPUT_NORM: (
"gpt_neox.final_layer_norm", # gptneox
"transformer.ln_f", # gpt2 gpt-j falcon
"model.norm", # llama-hf baichuan
"model.norm", # llama-hf baichuan internlm2
"norm", # llama-pth
"embeddings.LayerNorm", # bert
"transformer.norm_f", # mpt
@@ -84,6 +85,7 @@ class TensorNameMap:
"h.{bid}.ln_1", # gpt2
"transformer.h.{bid}.ln", # phi2
"model.layers.layers.{bid}.norm", # plamo
"model.layers.{bid}.attention_norm", # internlm2
),
# Attention norm 2
@@ -111,6 +113,7 @@ class TensorNameMap:
"encoder.layer.{bid}.attention.self.query", # bert
"transformer.h.{bid}.attn.q_proj", # gpt-j
"model.layers.layers.{bid}.self_attn.q_proj", # plamo
"model.layers.{bid}.attention.wq" # internlm2
),
# Attention key
@@ -120,6 +123,7 @@ class TensorNameMap:
"encoder.layer.{bid}.attention.self.key", # bert
"transformer.h.{bid}.attn.k_proj", # gpt-j
"model.layers.layers.{bid}.self_attn.k_proj", # plamo
"model.layers.{bid}.attention.wk" # internlm2
),
# Attention value
@@ -129,6 +133,7 @@ class TensorNameMap:
"encoder.layer.{bid}.attention.self.value", # bert
"transformer.h.{bid}.attn.v_proj", # gpt-j
"model.layers.layers.{bid}.self_attn.v_proj", # plamo
"model.layers.{bid}.attention.wv" # internlm2
),
# Attention output
@@ -147,6 +152,7 @@ class TensorNameMap:
"h.{bid}.attn.c_proj", # gpt2
"transformer.h.{bid}.mixer.out_proj", # phi2
"model.layers.layers.{bid}.self_attn.o_proj", # plamo
"model.layers.{bid}.attention.wo", # internlm2
),
# Rotary embeddings
@@ -169,6 +175,7 @@ class TensorNameMap:
"language_model.encoder.layers.{bid}.post_attention_layernorm", # persimmon
"model.layers.{bid}.ln2", # yi
"h.{bid}.ln_2", # gpt2
"model.layers.{bid}.ffn_norm", # internlm2
),
MODEL_TENSOR.FFN_GATE_INP: (
@@ -194,6 +201,7 @@ class TensorNameMap:
"transformer.h.{bid}.mlp.fc1", # phi2
"model.layers.{bid}.mlp.fc1", # phi2
"model.layers.layers.{bid}.mlp.up_proj", # plamo
"model.layers.{bid}.feed_forward.w3", # internlm2
),
MODEL_TENSOR.FFN_UP_EXP: (
@@ -212,6 +220,7 @@ class TensorNameMap:
"layers.{bid}.feed_forward.w1", # llama-pth
"transformer.h.{bid}.mlp.w2", # qwen
"model.layers.layers.{bid}.mlp.gate_proj", # plamo
"model.layers.{bid}.feed_forward.w1", # internlm2
),
MODEL_TENSOR.FFN_GATE_EXP: (
@@ -236,6 +245,7 @@ class TensorNameMap:
"transformer.h.{bid}.mlp.fc2", # phi2
"model.layers.{bid}.mlp.fc2", # phi2
"model.layers.layers.{bid}.mlp.down_proj", # plamo
"model.layers.{bid}.feed_forward.w2", # internlm2
),
MODEL_TENSOR.FFN_DOWN_EXP: (