github/DB-GPT
1
0
Fork 0
mirror of https://github.com/csunny/DB-GPT.git synced 2025-09-16 22:51:24 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(model): Support Qwen2 models (#1612)

Browse Source
This commit is contained in:
Fangyin Cheng
2024-06-07 00:56:20 +08:00
committed by GitHub
parent 20e7ccc831
commit e11087aa7e
5 changed files with 67 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
dbgpt/configs/model_config.py
View File

@@ -54,6 +54,7 @@ LLM_MODEL_CONFIG = {
"chatglm3-6b": os.path.join(MODEL_PATH, "chatglm3-6b"),
# https://huggingface.co/THUDM/glm-4-9b-chat
"glm-4-9b-chat": os.path.join(MODEL_PATH, "glm-4-9b-chat"),
"glm-4-9b-chat-1m": os.path.join(MODEL_PATH, "glm-4-9b-chat-1m"),
"guanaco-33b-merged": os.path.join(MODEL_PATH, "guanaco-33b-merged"),
"falcon-40b": os.path.join(MODEL_PATH, "falcon-40b"),
"gorilla-7b": os.path.join(MODEL_PATH, "gorilla-7b"),
@@ -124,6 +125,42 @@ LLM_MODEL_CONFIG = {
"codeqwen1.5-7b-chat": os.path.join(MODEL_PATH, "CodeQwen1.5-7B-Chat"),
# https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B-Chat
"qwen1.5-moe-a2.7b-chat": os.path.join(MODEL_PATH, "Qwen1.5-MoE-A2.7B-Chat"),
"qwen2-57b-a14b-instruct": os.path.join(MODEL_PATH, "Qwen2-57B-A14B-Instruct"),
"qwen2-57b-a14b-instruct-gptq-int4": os.path.join(
MODEL_PATH, "Qwen2-57B-A14B-Instruct-GPTQ-Int4"
),
"qwen2-72b-instruct": os.path.join(MODEL_PATH, "Qwen2-72B-Instruct"),
"qwen2-72b-instruct-awq": os.path.join(MODEL_PATH, "Qwen2-72B-Instruct-AWQ"),
"qwen2-72b-instruct-gptq-int8": os.path.join(
MODEL_PATH, "Qwen2-72B-Instruct-GPTQ-Int8"
),
"qwen2-72b-instruct-gptq-int4": os.path.join(
MODEL_PATH, "Qwen2-72B-Instruct-GPTQ-Int4"
),
"qwen2-7b-instruct": os.path.join(MODEL_PATH, "Qwen2-7B-Instruct"),
"qwen2-7b-instruct-awq": os.path.join(MODEL_PATH, "Qwen2-7B-Instruct-AWQ"),
"qwen2-7b-instruct-gptq-int8": os.path.join(
MODEL_PATH, "Qwen2-7B-Instruct-GPTQ-Int8"
),
"qwen2-7b-instruct-gptq-int4": os.path.join(
MODEL_PATH, "Qwen2-7B-Instruct-GPTQ-Int4"
),
"qwen2-1.5b-instruct": os.path.join(MODEL_PATH, "Qwen2-1.5B-Instruct"),
"qwen2-1.5b-instruct-awq": os.path.join(MODEL_PATH, "Qwen2-1.5B-Instruct-AWQ"),
"qwen2-1.5b-instruct-gptq-int8": os.path.join(
MODEL_PATH, "Qwen2-1.5B-Instruct-GPTQ-Int8"
),
"qwen2-1.5b-instruct-gptq-int4": os.path.join(
MODEL_PATH, "Qwen2-1.5B-Instruct-GPTQ-Int4"
),
"qwen2-0.5b-instruct": os.path.join(MODEL_PATH, "Qwen2-0.5B-Instruct"),
"qwen2-0.5b-instruct-awq": os.path.join(MODEL_PATH, "Qwen2-0.5B-Instruct-AWQ"),
"qwen2-0.5b-instruct-gptq-int8": os.path.join(
MODEL_PATH, "Qwen2-0.5B-Instruct-GPTQ-Int8"
),
"qwen2-0.5b-instruct-gptq-int4": os.path.join(
MODEL_PATH, "Qwen2-0.5B-Instruct-GPTQ-Int4"
),
# (Llama2 based) We only support WizardLM-13B-V1.2 for now, which is trained from Llama-2 13b, see https://huggingface.co/WizardLM/WizardLM-13B-V1.2
"wizardlm-13b": os.path.join(MODEL_PATH, "WizardLM-13B-V1.2"),
# wget https://huggingface.co/TheBloke/vicuna-13B-v1.5-GGUF/resolve/main/vicuna-13b-v1.5.Q4_K_M.gguf -O models/ggml-model-q4_0.gguf

15
dbgpt/model/adapter/hf_adapter.py
View File

@@ -310,6 +310,20 @@ class QwenAdapter(NewHFChatModelAdapter):
and "qwen" in lower_model_name_or_path
and "1.5" in lower_model_name_or_path
and "moe" not in lower_model_name_or_path
and "qwen2" not in lower_model_name_or_path
)
class Qwen2Adapter(QwenAdapter):
support_4bit: bool = True
support_8bit: bool = True
def do_match(self, lower_model_name_or_path: Optional[str] = None):
return (
lower_model_name_or_path
and "qwen2" in lower_model_name_or_path
and "instruct" in lower_model_name_or_path
)
@@ -517,3 +531,4 @@ register_model_adapter(PhiAdapter)
register_model_adapter(SQLCoderAdapter)
register_model_adapter(OpenChatAdapter)
register_model_adapter(GLM4Aapter)
register_model_adapter(Qwen2Adapter)