Update handler.py
Browse files- handler.py +22 -28
handler.py
CHANGED
@@ -1,4 +1,3 @@
|
|
1 |
-
# handler.py —— 放在模型仓库根目录
|
2 |
from typing import Dict, Any
|
3 |
import torch
|
4 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
@@ -6,58 +5,53 @@ from accelerate import init_empty_weights, load_checkpoint_and_dispatch
|
|
6 |
|
7 |
|
8 |
class EndpointHandler:
|
9 |
-
"""
|
10 |
-
Hugging Face Inference Endpoints 约定的自定义入口:
|
11 |
-
• __init__(model_dir, **kwargs) —— 加载模型
|
12 |
-
• __call__(inputs: Dict) -> Dict —— 处理一次请求
|
13 |
-
"""
|
14 |
-
|
15 |
def __init__(self, model_dir: str, **kwargs):
|
16 |
-
# 1️⃣ Tokenizer
|
17 |
self.tokenizer = AutoTokenizer.from_pretrained(
|
18 |
model_dir, trust_remote_code=True
|
19 |
)
|
20 |
|
21 |
-
#
|
22 |
with init_empty_weights():
|
23 |
-
|
24 |
model_dir,
|
25 |
torch_dtype=torch.float16,
|
26 |
trust_remote_code=True,
|
27 |
)
|
28 |
|
29 |
-
#
|
30 |
self.model = load_checkpoint_and_dispatch(
|
31 |
-
|
32 |
checkpoint=model_dir,
|
33 |
-
device_map="auto",
|
34 |
dtype=torch.float16,
|
35 |
-
)
|
36 |
|
37 |
-
#
|
|
|
|
|
|
|
|
|
38 |
self.generation_kwargs = dict(
|
39 |
-
max_new_tokens=
|
40 |
do_sample=True,
|
41 |
temperature=0.7,
|
42 |
top_p=0.9,
|
43 |
)
|
44 |
|
|
|
|
|
|
|
45 |
def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
|
46 |
prompt = data["inputs"]
|
47 |
-
|
48 |
-
#
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
# ② 生成(其余逻辑不变)
|
53 |
with torch.inference_mode():
|
54 |
-
output_ids = self.model.generate(
|
55 |
-
|
56 |
-
**self.generation_kwargs,
|
57 |
-
)
|
58 |
-
|
59 |
return {
|
60 |
"generated_text": self.tokenizer.decode(
|
61 |
output_ids[0], skip_special_tokens=True
|
62 |
)
|
63 |
-
}
|
|
|
|
|
1 |
from typing import Dict, Any
|
2 |
import torch
|
3 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
|
|
5 |
|
6 |
|
7 |
class EndpointHandler:
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
def __init__(self, model_dir: str, **kwargs):
|
|
|
9 |
self.tokenizer = AutoTokenizer.from_pretrained(
|
10 |
model_dir, trust_remote_code=True
|
11 |
)
|
12 |
|
13 |
+
# ① 构空壳模型
|
14 |
with init_empty_weights():
|
15 |
+
base = AutoModelForCausalLM.from_pretrained(
|
16 |
model_dir,
|
17 |
torch_dtype=torch.float16,
|
18 |
trust_remote_code=True,
|
19 |
)
|
20 |
|
21 |
+
# ② 分片加载到多 GPU
|
22 |
self.model = load_checkpoint_and_dispatch(
|
23 |
+
base,
|
24 |
checkpoint=model_dir,
|
25 |
+
device_map="auto",
|
26 |
dtype=torch.float16,
|
27 |
+
).eval()
|
28 |
|
29 |
+
# ③ 记录 embedding 所在 GPU,并把 **默认 GPU** 也切过去
|
30 |
+
self.first_device = next(self.model.parameters()).device
|
31 |
+
torch.cuda.set_device(self.first_device) # ← 关键一行
|
32 |
+
|
33 |
+
# ④ 生成参数
|
34 |
self.generation_kwargs = dict(
|
35 |
+
max_new_tokens=512, # 🛈 2 k token 占显存极高,先压到 512 再逐步调
|
36 |
do_sample=True,
|
37 |
temperature=0.7,
|
38 |
top_p=0.9,
|
39 |
)
|
40 |
|
41 |
+
# (可选)在日志中打印设备映射,方便后续排查
|
42 |
+
print(">>> device_map =", self.model.hf_device_map)
|
43 |
+
|
44 |
def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
|
45 |
prompt = data["inputs"]
|
46 |
+
|
47 |
+
# 把 *所有* 输入张量放到 first_device
|
48 |
+
inputs = self.tokenizer(prompt, return_tensors="pt").to(self.first_device)
|
49 |
+
|
|
|
|
|
50 |
with torch.inference_mode():
|
51 |
+
output_ids = self.model.generate(**inputs, **self.generation_kwargs)
|
52 |
+
|
|
|
|
|
|
|
53 |
return {
|
54 |
"generated_text": self.tokenizer.decode(
|
55 |
output_ids[0], skip_special_tokens=True
|
56 |
)
|
57 |
+
}
|