Bases: ConstrainedModel
Class for Transformers local Model.
Parameters:
| Name |
Type |
Description |
Default |
model_name_or_path
|
str
|
Name of the model on HuggingFace, or the path to a local model
|
required
|
caching
|
bool
|
Bool determining whether we access the model's cache
|
True
|
Examples:
from blendsql.models import TransformersLLM
model = TransformersLLM("Qwen/Qwen1.5-0.5B")
Source code in blendsql/models/constrained/_guidance.py
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104 | class TransformersLLM(ConstrainedModel):
"""Class for Transformers local Model.
Args:
model_name_or_path: Name of the model on HuggingFace, or the path to a local model
caching: Bool determining whether we access the model's cache
Examples:
```python
from blendsql.models import TransformersLLM
model = TransformersLLM("Qwen/Qwen1.5-0.5B")
```
"""
def __init__(
self,
model_name_or_path: str,
config: Optional[dict] = None,
caching: bool = True,
**kwargs,
):
if not _has_transformers and _has_torch:
raise ImportError(
"Please install transformers with `pip install transformers==4.47.0`!"
) from None
elif not _has_torch and _has_transformers:
raise ImportError(
"Please install pytorch with `pip install torch`!"
) from None
elif not _has_torch and not _has_transformers:
raise ImportError(
"Please install transformers and pytorch with `pip install transformers==4.47.0 torch`!"
) from None
import transformers
transformers.logging.set_verbosity_error()
if config is None:
config = {}
super().__init__(
model_name_or_path=model_name_or_path,
requires_config=False,
tokenizer=transformers.AutoTokenizer.from_pretrained(model_name_or_path),
config=DEFAULT_KWARGS | config,
caching=caching,
**kwargs,
)
@cached_property
def model_obj(self) -> ModelObj:
"""Allows for lazy loading of underlying model weights."""
return self._load_model()
def _load_model(self) -> ModelObj:
# https://huggingface.co/blog/how-to-generate
from guidance.models import Transformers
device_map = resolve_device_map(self.config.pop("device_map", None))
lm = Transformers(
self.model_name_or_path,
echo=False,
device_map=device_map,
**self.config,
)
# Try to infer if we're in chat mode
if lm.engine.tokenizer._orig_tokenizer.chat_template is None:
logger.debug(
Fore.YELLOW
+ "chat_template not found in tokenizer config.\nBlendSQL currently only works with chat models"
+ Fore.RESET
)
return lm
|