13481713164

Committed 23 Feb 2025 09:27AM UTC coverage: 80.826% (-0.09%) from 80.918%

Build # 13481713164

Build Type

push

github

Committed by

web-flow

Commit Message

Enable offline mode for hugginface by using local pre-downloaded metrics, datasets and models (#1603)

* Enable local caching with UNITXT_LOCAL_CACHE

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Add cache_dir parameter to model and tokenizer loading for improved performance

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Update line numbers and generated timestamp in secrets baseline

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Update line number and generated timestamp in secrets baseline

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Fix metrics.py

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Use huggingface prefix

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Clarify documentation for offline Hugging Face model, metrics, and datasets paths

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Rename hf_offline_loaders_path to hf_offline_models_path for clarity

Signed-off-by: elronbandel <elronbandel@gmail.com>

---------

Signed-off-by: elronbandel <elronbandel@gmail.com>

Run Details

1540 of 1899 branches covered (81.1%)

Branch coverage included in aggregate %.

9677 of 11979 relevant lines covered (80.78%)

0.81 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

86.59

src/unitxt/string_operators.py

import os
import re
from typing import (
    Any,
    Dict,
    List,
    Optional,
)

from .operators import FieldOperator, InstanceOperator
from .settings_utils import get_settings

settings = get_settings()

class Split(FieldOperator):
    by: str

    def process_value(self, value: str) -> List[str]:
        return value.split(self.by)


class RegexSplit(FieldOperator):
    by: str

    def process_value(self, value: str) -> List[str]:
        return re.split(self.by, value)


class TokensSplit(FieldOperator):
    model: str
    _requirements_list = ["transformers"]

    def prepare(self):
        super().prepare()
        from transformers import AutoTokenizer
        path = self.model
        if settings.hf_offline_models_path is not None:
            path = os.path.join(settings.hf_offline_models_path, path)
        self.tokenizer = AutoTokenizer.from_pretrained(path)

    def process_value(self, value: str) -> List[str]:
        return self.tokenizer.tokenize(value)


class TokensSlice(FieldOperator):
    model: str
    start: Optional[int] = None
    stop: Optional[int] = None
    step: Optional[int] = None

    _requirements_list = ["transformers"]

    def prepare(self):
        super().prepare()
        from transformers import AutoTokenizer
        path = self.model
        if settings.hf_offline_models_path is not None:
            path = os.path.join(settings.hf_offline_models_path, path)
        self.tokenizer = AutoTokenizer.from_pretrained(path)

    def process_value(self, value: str) -> str:
        encoded = self.tokenizer.encode(value)
        slicer = slice(self.start, self.stop, self.step)
        sliced = encoded[slicer]
        return self.tokenizer.decode(sliced)


class Join(FieldOperator):
    by: str

    def process_value(self, value: List[str]) -> str:
        return self.by.join(value)


class FormatText(InstanceOperator):
    to_field: str
    text: str

    def process(
        self, instance: Dict[str, Any], stream_name: Optional[str] = None
    ) -> Dict[str, Any]:
        instance[self.to_field] = self.text.format(**instance)
        return instance


class Strip(FieldOperator):
    def process_value(self, value: str) -> str:
        return value.strip()


class Replace(FieldOperator):
    old: str
    new: str

    def process_value(self, value: str) -> str:
        return value.replace(self.old, self.new)


class MapReplace(FieldOperator):
    mapping: Dict[str, str]

    def process_value(self, value: Any) -> Any:
        for key, val in self.mapping.items():
            value = value.replace(key, val)
        return value


class RegexReplace(FieldOperator):
    pattern: str  # A regex pattern
    replacement: str  # The replacement string or template

    def prepare(self):
        super().prepare()
        self.pattern = re.compile(self.pattern)

    def process_value(self, value: Any) -> Any:
        if isinstance(value, str):
            return re.sub(self.pattern, self.replacement, value)
        return value  # If not a string, return the value as is

1	import os	1✔
2	import re	1✔
3	from typing import (	1✔
4	Any,
5	Dict,
6	List,
7	Optional,
8	)
9
10	from .operators import FieldOperator, InstanceOperator	1✔
11	from .settings_utils import get_settings	1✔
12
13	settings = get_settings()	1✔
14
15	class Split(FieldOperator):	1✔
16	by: str	1✔
17
18	def process_value(self, value: str) -> List[str]:	1✔
19	return value.split(self.by)	1✔
20
21
22	class RegexSplit(FieldOperator):	1✔
23	by: str	1✔
24
25	def process_value(self, value: str) -> List[str]:	1✔
26	return re.split(self.by, value)	1✔
27
28
29	class TokensSplit(FieldOperator):	1✔
30	model: str	1✔
31	_requirements_list = ["transformers"]	1✔
32
33	def prepare(self):	1✔
34	super().prepare()	1✔
35	from transformers import AutoTokenizer	1✔
36	path = self.model	1✔
37	if settings.hf_offline_models_path is not None:	1✔
38	path = os.path.join(settings.hf_offline_models_path, path)	×
39	self.tokenizer = AutoTokenizer.from_pretrained(path)	1✔
40
41	def process_value(self, value: str) -> List[str]:	1✔
42	return self.tokenizer.tokenize(value)	1✔
43
44
45	class TokensSlice(FieldOperator):	1✔
46	model: str	1✔
47	start: Optional[int] = None	1✔
48	stop: Optional[int] = None	1✔
49	step: Optional[int] = None	1✔
50
51	_requirements_list = ["transformers"]	1✔
52
53	def prepare(self):	1✔
54	super().prepare()	1✔
55	from transformers import AutoTokenizer	1✔
56	path = self.model	1✔
57	if settings.hf_offline_models_path is not None:	1✔
58	path = os.path.join(settings.hf_offline_models_path, path)	×
59	self.tokenizer = AutoTokenizer.from_pretrained(path)	1✔
60
61	def process_value(self, value: str) -> str:	1✔
62	encoded = self.tokenizer.encode(value)	1✔
63	slicer = slice(self.start, self.stop, self.step)	1✔
64	sliced = encoded[slicer]	1✔
65	return self.tokenizer.decode(sliced)	1✔
66
67
68	class Join(FieldOperator):	1✔
69	by: str	1✔
70
71	def process_value(self, value: List[str]) -> str:	1✔
72	return self.by.join(value)	1✔
73
74
75	class FormatText(InstanceOperator):	1✔
76	to_field: str	1✔
77	text: str	1✔
78
79	def process(	1✔
80	self, instance: Dict[str, Any], stream_name: Optional[str] = None
81	) -> Dict[str, Any]:
82	instance[self.to_field] = self.text.format(**instance)	×
83	return instance	×
84
85
86	class Strip(FieldOperator):	1✔
87	def process_value(self, value: str) -> str:	1✔
88	return value.strip()	×
89
90
91	class Replace(FieldOperator):	1✔
92	old: str	1✔
93	new: str	1✔
94
95	def process_value(self, value: str) -> str:	1✔
96	return value.replace(self.old, self.new)	×
97
98
99	class MapReplace(FieldOperator):	1✔
100	mapping: Dict[str, str]	1✔
101
102	def process_value(self, value: Any) -> Any:	1✔
103	for key, val in self.mapping.items():	×
104	value = value.replace(key, val)	×
105	return value	×
106
107
108	class RegexReplace(FieldOperator):	1✔
109	pattern: str # A regex pattern	1✔
110	replacement: str # The replacement string or template	1✔
111
112	def prepare(self):	1✔
113	super().prepare()	1✔
114	self.pattern = re.compile(self.pattern)	1✔
115
116	def process_value(self, value: Any) -> Any:	1✔
117	if isinstance(value, str):	1✔
118	return re.sub(self.pattern, self.replacement, value)	1✔
119	return value # If not a string, return the value as is	×

IBM / unitxt / 13481713164

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous