Spaces:
Sleeping
Sleeping
Tuchuanhuhuhu
commited on
Commit
•
75dddd5
1
Parent(s):
2342c7b
加入MOSS支持
Browse files- modules/models/MOSS.py +33 -0
- modules/models/base_model.py +3 -0
- modules/presets.py +2 -1
modules/models/MOSS.py
ADDED
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer
|
3 |
+
import time
|
4 |
+
import numpy as np
|
5 |
+
from torch.nn import functional as F
|
6 |
+
import os
|
7 |
+
from accelerate import init_empty_weights, load_checkpoint_and_dispatch
|
8 |
+
from transformers import MossForCausalLM, MossConfig
|
9 |
+
|
10 |
+
from .base_model import BaseLLMModel
|
11 |
+
|
12 |
+
MOSS_MODEL = None
|
13 |
+
MOSS_TOKENIZER = None
|
14 |
+
|
15 |
+
class MOSS_Client(BaseLLMModel):
|
16 |
+
def __init__(self, model_name) -> None:
|
17 |
+
super().__init__(model_name=model_name)
|
18 |
+
global MOSS_MODEL, MOSS_TOKENIZER
|
19 |
+
config = MossConfig.from_pretrained("fnlp/moss-16B-sft")
|
20 |
+
print("MOSS Model Parallelism Devices: ", torch.cuda.device_count())
|
21 |
+
with init_empty_weights():
|
22 |
+
raw_model = MossForCausalLM._from_config(config, torch_dtype=torch.float16)
|
23 |
+
raw_model.tie_weights()
|
24 |
+
MOSS_MODEL = load_checkpoint_and_dispatch(
|
25 |
+
raw_model,
|
26 |
+
"fnlp/moss-16B-sft",
|
27 |
+
device_map="auto",
|
28 |
+
no_split_module_classes=["MossBlock"],
|
29 |
+
dtype=torch.float16
|
30 |
+
)
|
31 |
+
|
32 |
+
if __name__ == "__main__":
|
33 |
+
model = MOSS_Client("MOSS")
|
modules/models/base_model.py
CHANGED
@@ -31,6 +31,7 @@ class ModelType(Enum):
|
|
31 |
LLaMA = 2
|
32 |
XMChat = 3
|
33 |
StableLM = 4
|
|
|
34 |
|
35 |
@classmethod
|
36 |
def get_type(cls, model_name: str):
|
@@ -46,6 +47,8 @@ class ModelType(Enum):
|
|
46 |
model_type = ModelType.XMChat
|
47 |
elif "stablelm" in model_name_lower:
|
48 |
model_type = ModelType.StableLM
|
|
|
|
|
49 |
else:
|
50 |
model_type = ModelType.Unknown
|
51 |
return model_type
|
|
|
31 |
LLaMA = 2
|
32 |
XMChat = 3
|
33 |
StableLM = 4
|
34 |
+
MOSS = 5
|
35 |
|
36 |
@classmethod
|
37 |
def get_type(cls, model_name: str):
|
|
|
47 |
model_type = ModelType.XMChat
|
48 |
elif "stablelm" in model_name_lower:
|
49 |
model_type = ModelType.StableLM
|
50 |
+
elif "moss" in model_name_lower:
|
51 |
+
model_type = ModelType.MOSS
|
52 |
else:
|
53 |
model_type = ModelType.Unknown
|
54 |
return model_type
|
modules/presets.py
CHANGED
@@ -74,11 +74,12 @@ LOCAL_MODELS = [
|
|
74 |
"chatglm-6b",
|
75 |
"chatglm-6b-int4",
|
76 |
"chatglm-6b-int4-qe",
|
|
|
|
|
77 |
"llama-7b-hf",
|
78 |
"llama-13b-hf",
|
79 |
"llama-30b-hf",
|
80 |
"llama-65b-hf",
|
81 |
-
"StableLM"
|
82 |
]
|
83 |
|
84 |
if os.environ.get('HIDE_LOCAL_MODELS', 'false') == 'true':
|
|
|
74 |
"chatglm-6b",
|
75 |
"chatglm-6b-int4",
|
76 |
"chatglm-6b-int4-qe",
|
77 |
+
"StableLM",
|
78 |
+
"MOSS",
|
79 |
"llama-7b-hf",
|
80 |
"llama-13b-hf",
|
81 |
"llama-30b-hf",
|
82 |
"llama-65b-hf",
|
|
|
83 |
]
|
84 |
|
85 |
if os.environ.get('HIDE_LOCAL_MODELS', 'false') == 'true':
|