janus-llm 4.2.0__py3-none-any.whl → 4.3.1__py3-none-any.whl
Sign up to get free protection for your applications and to get access to all the features.
- janus/__init__.py +1 -1
- janus/cli.py +150 -5
- janus/converter/converter.py +1 -0
- janus/converter/evaluate.py +230 -4
- janus/language/alc/_tests/test_alc.py +1 -1
- janus/language/alc/alc.py +9 -4
- janus/language/splitter.py +31 -23
- janus/language/treesitter/treesitter.py +9 -1
- janus/llm/models_info.py +17 -12
- janus/parsers/eval_parsers/incose_parser.py +134 -0
- janus/parsers/eval_parsers/inline_comment_parser.py +112 -0
- janus/parsers/partition_parser.py +41 -9
- janus/refiners/refiner.py +30 -0
- janus/utils/enums.py +14 -0
- {janus_llm-4.2.0.dist-info → janus_llm-4.3.1.dist-info}/METADATA +1 -1
- {janus_llm-4.2.0.dist-info → janus_llm-4.3.1.dist-info}/RECORD +19 -17
- {janus_llm-4.2.0.dist-info → janus_llm-4.3.1.dist-info}/LICENSE +0 -0
- {janus_llm-4.2.0.dist-info → janus_llm-4.3.1.dist-info}/WHEEL +0 -0
- {janus_llm-4.2.0.dist-info → janus_llm-4.3.1.dist-info}/entry_points.txt +0 -0
janus/__init__.py
CHANGED
@@ -5,7 +5,7 @@ from langchain_core._api.deprecation import LangChainDeprecationWarning
|
|
5
5
|
from janus.converter.translate import Translator
|
6
6
|
from janus.metrics import * # noqa: F403
|
7
7
|
|
8
|
-
__version__ = "4.
|
8
|
+
__version__ = "4.3.1"
|
9
9
|
|
10
10
|
# Ignoring a deprecation warning from langchain_core that I can't seem to hunt down
|
11
11
|
warnings.filterwarnings("ignore", category=LangChainDeprecationWarning)
|
janus/cli.py
CHANGED
@@ -19,6 +19,7 @@ from janus.converter.aggregator import Aggregator
|
|
19
19
|
from janus.converter.converter import Converter
|
20
20
|
from janus.converter.diagram import DiagramGenerator
|
21
21
|
from janus.converter.document import Documenter, MadLibsDocumenter, MultiDocumenter
|
22
|
+
from janus.converter.evaluate import InlineCommentEvaluator, RequirementEvaluator
|
22
23
|
from janus.converter.partition import Partitioner
|
23
24
|
from janus.converter.requirements import RequirementsDocumenter
|
24
25
|
from janus.converter.translate import Translator
|
@@ -127,7 +128,7 @@ embedding = typer.Typer(
|
|
127
128
|
|
128
129
|
def version_callback(value: bool) -> None:
|
129
130
|
if value:
|
130
|
-
from
|
131
|
+
from . import __version__ as version
|
131
132
|
|
132
133
|
print(f"Janus CLI [blue]v{version}[/blue]")
|
133
134
|
raise typer.Exit()
|
@@ -655,6 +656,16 @@ def partition(
|
|
655
656
|
click_type=click.Choice(list(CUSTOM_SPLITTERS.keys())),
|
656
657
|
),
|
657
658
|
] = "file",
|
659
|
+
refiner_types: Annotated[
|
660
|
+
list[str],
|
661
|
+
typer.Option(
|
662
|
+
"-r",
|
663
|
+
"--refiner",
|
664
|
+
help="List of refiner types to use. Add -r for each refiner to use in\
|
665
|
+
refinement chain",
|
666
|
+
click_type=click.Choice(list(REFINERS.keys())),
|
667
|
+
),
|
668
|
+
] = ["JanusRefiner"],
|
658
669
|
max_tokens: Annotated[
|
659
670
|
int,
|
660
671
|
typer.Option(
|
@@ -673,6 +684,7 @@ def partition(
|
|
673
684
|
),
|
674
685
|
] = 8192,
|
675
686
|
):
|
687
|
+
refiner_types = [REFINERS[r] for r in refiner_types]
|
676
688
|
model_arguments = dict(temperature=temperature)
|
677
689
|
kwargs = dict(
|
678
690
|
model=llm_name,
|
@@ -681,6 +693,7 @@ def partition(
|
|
681
693
|
max_prompts=max_prompts,
|
682
694
|
max_tokens=max_tokens,
|
683
695
|
splitter_type=splitter_type,
|
696
|
+
refiner_types=refiner_types,
|
684
697
|
partition_token_limit=partition_token_limit,
|
685
698
|
)
|
686
699
|
partitioner = Partitioner(**kwargs)
|
@@ -815,6 +828,139 @@ def diagram(
|
|
815
828
|
diagram_generator.translate(input_dir, output_dir, overwrite, collection)
|
816
829
|
|
817
830
|
|
831
|
+
@app.command(
|
832
|
+
help="LLM self evaluation",
|
833
|
+
no_args_is_help=True,
|
834
|
+
)
|
835
|
+
def llm_self_eval(
|
836
|
+
input_dir: Annotated[
|
837
|
+
Path,
|
838
|
+
typer.Option(
|
839
|
+
"--input",
|
840
|
+
"-i",
|
841
|
+
help="The directory containing the source code to be evaluated. "
|
842
|
+
"The files should all be in one flat directory.",
|
843
|
+
),
|
844
|
+
],
|
845
|
+
language: Annotated[
|
846
|
+
str,
|
847
|
+
typer.Option(
|
848
|
+
"--language",
|
849
|
+
"-l",
|
850
|
+
help="The language of the source code.",
|
851
|
+
click_type=click.Choice(sorted(LANGUAGES)),
|
852
|
+
),
|
853
|
+
],
|
854
|
+
output_dir: Annotated[
|
855
|
+
Path,
|
856
|
+
typer.Option(
|
857
|
+
"--output-dir", "-o", help="The directory to store the evaluations in."
|
858
|
+
),
|
859
|
+
],
|
860
|
+
llm_name: Annotated[
|
861
|
+
str,
|
862
|
+
typer.Option(
|
863
|
+
"--llm",
|
864
|
+
"-L",
|
865
|
+
help="The custom name of the model set with 'janus llm add'.",
|
866
|
+
),
|
867
|
+
] = "gpt-4o",
|
868
|
+
evaluation_type: Annotated[
|
869
|
+
str,
|
870
|
+
typer.Option(
|
871
|
+
"--evaluation-type",
|
872
|
+
"-e",
|
873
|
+
help="Type of output to evaluate.",
|
874
|
+
click_type=click.Choice(["incose", "comments"]),
|
875
|
+
),
|
876
|
+
] = "incose",
|
877
|
+
max_prompts: Annotated[
|
878
|
+
int,
|
879
|
+
typer.Option(
|
880
|
+
"--max-prompts",
|
881
|
+
"-m",
|
882
|
+
help="The maximum number of times to prompt a model on one functional block "
|
883
|
+
"before exiting the application. This is to prevent wasting too much money.",
|
884
|
+
),
|
885
|
+
] = 10,
|
886
|
+
overwrite: Annotated[
|
887
|
+
bool,
|
888
|
+
typer.Option(
|
889
|
+
"--overwrite/--preserve",
|
890
|
+
help="Whether to overwrite existing files in the output directory",
|
891
|
+
),
|
892
|
+
] = False,
|
893
|
+
temperature: Annotated[
|
894
|
+
float,
|
895
|
+
typer.Option("--temperature", "-t", help="Sampling temperature.", min=0, max=2),
|
896
|
+
] = 0.7,
|
897
|
+
collection: Annotated[
|
898
|
+
str,
|
899
|
+
typer.Option(
|
900
|
+
"--collection",
|
901
|
+
"-c",
|
902
|
+
help="If set, will put the translated result into a Chroma DB "
|
903
|
+
"collection with the name provided.",
|
904
|
+
),
|
905
|
+
] = None,
|
906
|
+
splitter_type: Annotated[
|
907
|
+
str,
|
908
|
+
typer.Option(
|
909
|
+
"-S",
|
910
|
+
"--splitter",
|
911
|
+
help="Name of custom splitter to use",
|
912
|
+
click_type=click.Choice(list(CUSTOM_SPLITTERS.keys())),
|
913
|
+
),
|
914
|
+
] = "file",
|
915
|
+
refiner_types: Annotated[
|
916
|
+
list[str],
|
917
|
+
typer.Option(
|
918
|
+
"-r",
|
919
|
+
"--refiner",
|
920
|
+
help="List of refiner types to use. Add -r for each refiner to use in\
|
921
|
+
refinement chain",
|
922
|
+
click_type=click.Choice(list(REFINERS.keys())),
|
923
|
+
),
|
924
|
+
] = ["JanusRefiner"],
|
925
|
+
eval_items_per_request: Annotated[
|
926
|
+
int,
|
927
|
+
typer.Option(
|
928
|
+
"--eval-items-per-request",
|
929
|
+
"-rc",
|
930
|
+
help="The maximum number of evaluation items per request",
|
931
|
+
),
|
932
|
+
] = None,
|
933
|
+
max_tokens: Annotated[
|
934
|
+
int,
|
935
|
+
typer.Option(
|
936
|
+
"--max-tokens",
|
937
|
+
"-M",
|
938
|
+
help="The maximum number of tokens the model will take in. "
|
939
|
+
"If unspecificed, model's default max will be used.",
|
940
|
+
),
|
941
|
+
] = None,
|
942
|
+
):
|
943
|
+
model_arguments = dict(temperature=temperature)
|
944
|
+
refiner_types = [REFINERS[r] for r in refiner_types]
|
945
|
+
kwargs = dict(
|
946
|
+
eval_items_per_request=eval_items_per_request,
|
947
|
+
model=llm_name,
|
948
|
+
model_arguments=model_arguments,
|
949
|
+
source_language=language,
|
950
|
+
max_prompts=max_prompts,
|
951
|
+
max_tokens=max_tokens,
|
952
|
+
splitter_type=splitter_type,
|
953
|
+
refiner_types=refiner_types,
|
954
|
+
)
|
955
|
+
# Setting parser type here
|
956
|
+
if evaluation_type == "incose":
|
957
|
+
evaluator = RequirementEvaluator(**kwargs)
|
958
|
+
elif evaluation_type == "comments":
|
959
|
+
evaluator = InlineCommentEvaluator(**kwargs)
|
960
|
+
|
961
|
+
evaluator.translate(input_dir, output_dir, overwrite, collection)
|
962
|
+
|
963
|
+
|
818
964
|
@db.command("init", help="Connect to or create a database.")
|
819
965
|
def db_init(
|
820
966
|
path: Annotated[
|
@@ -1116,13 +1262,12 @@ def llm_add(
|
|
1116
1262
|
show_choices=False,
|
1117
1263
|
)
|
1118
1264
|
params = dict(
|
1119
|
-
|
1120
|
-
model_name=MODEL_ID_TO_LONG_ID[model_id],
|
1265
|
+
model_name=model_name,
|
1121
1266
|
temperature=0.7,
|
1122
1267
|
n=1,
|
1123
1268
|
)
|
1124
|
-
max_tokens = TOKEN_LIMITS[
|
1125
|
-
model_cost = COST_PER_1K_TOKENS[
|
1269
|
+
max_tokens = TOKEN_LIMITS[model_name]
|
1270
|
+
model_cost = COST_PER_1K_TOKENS[model_name]
|
1126
1271
|
cfg = {
|
1127
1272
|
"model_type": model_type,
|
1128
1273
|
"model_id": model_id,
|
janus/converter/converter.py
CHANGED
@@ -464,6 +464,7 @@ class Converter:
|
|
464
464
|
for in_path, out_path in in_out_pairs:
|
465
465
|
# Translate the file, skip it if there's a rate limit error
|
466
466
|
try:
|
467
|
+
log.info(f"Processing {in_path.relative_to(input_directory)}")
|
467
468
|
out_block = self.translate_file(in_path)
|
468
469
|
total_cost += out_block.total_cost
|
469
470
|
except RateLimitError:
|
janus/converter/evaluate.py
CHANGED
@@ -1,15 +1,241 @@
|
|
1
|
+
import json
|
2
|
+
import re
|
3
|
+
from copy import deepcopy
|
4
|
+
|
5
|
+
from langchain_core.runnables import Runnable, RunnableLambda, RunnableParallel
|
6
|
+
|
1
7
|
from janus.converter.converter import Converter
|
8
|
+
from janus.language.block import TranslatedCodeBlock
|
2
9
|
from janus.language.combine import JsonCombiner
|
3
|
-
from janus.parsers.
|
10
|
+
from janus.parsers.eval_parsers.incose_parser import IncoseParser
|
11
|
+
from janus.parsers.eval_parsers.inline_comment_parser import InlineCommentParser
|
4
12
|
from janus.utils.logger import create_logger
|
5
13
|
|
6
14
|
log = create_logger(__name__)
|
7
15
|
|
8
16
|
|
9
17
|
class Evaluator(Converter):
|
10
|
-
|
18
|
+
"""Evaluator
|
19
|
+
|
20
|
+
A class that performs an LLM self evaluation"
|
21
|
+
"on an input target, with an associated prompt.
|
22
|
+
|
23
|
+
Current valid evaluation types:
|
24
|
+
['incose', 'comments']
|
25
|
+
|
26
|
+
"""
|
27
|
+
|
28
|
+
def __init__(self, **kwargs) -> None:
|
29
|
+
"""Initialize the Evaluator class
|
30
|
+
|
31
|
+
Arguments:
|
32
|
+
model: The LLM to use for translation. If an OpenAI model, the
|
33
|
+
`OPENAI_API_KEY` environment variable must be set and the
|
34
|
+
`OPENAI_ORG_ID` environment variable should be set if needed.
|
35
|
+
model_arguments: Additional arguments to pass to the LLM constructor.
|
36
|
+
max_prompts: The maximum number of prompts to try before giving up.
|
37
|
+
"""
|
38
|
+
super().__init__(**kwargs)
|
39
|
+
self._combiner = JsonCombiner()
|
40
|
+
self._load_parameters()
|
41
|
+
|
42
|
+
|
43
|
+
class RequirementEvaluator(Evaluator):
|
44
|
+
"""INCOSE Requirement Evaluator
|
45
|
+
|
46
|
+
A class that performs an LLM self evaluation on an input target,
|
47
|
+
with an associated prompt.
|
48
|
+
|
49
|
+
The evaluation prompts are for Incose Evaluations
|
50
|
+
|
51
|
+
"""
|
52
|
+
|
53
|
+
def __init__(self, eval_items_per_request: int | None = None, **kwargs) -> None:
|
54
|
+
"""Initialize the Evaluator class
|
55
|
+
|
56
|
+
Arguments:
|
57
|
+
model: The LLM to use for translation. If an OpenAI model, the
|
58
|
+
`OPENAI_API_KEY` environment variable must be set and the
|
59
|
+
`OPENAI_ORG_ID` environment variable should be set if needed.
|
60
|
+
model_arguments: Additional arguments to pass to the LLM constructor.
|
61
|
+
max_prompts: The maximum number of prompts to try before giving up.
|
62
|
+
"""
|
63
|
+
super().__init__(**kwargs)
|
64
|
+
self.eval_items_per_request = eval_items_per_request
|
65
|
+
self._parser = IncoseParser()
|
66
|
+
self.set_prompt("eval_prompts/incose")
|
67
|
+
|
68
|
+
def _input_runnable(self) -> Runnable:
|
69
|
+
def _get_code(json_text: str) -> str:
|
70
|
+
return json.loads(json_text)["code"]
|
71
|
+
|
72
|
+
def _get_reqs(json_text: str) -> str:
|
73
|
+
return json.dumps(json.loads(json_text)["requirements"])
|
74
|
+
|
75
|
+
return RunnableLambda(self._parser.parse_input) | RunnableParallel(
|
76
|
+
SOURCE_CODE=_get_code,
|
77
|
+
REQUIREMENTS=_get_reqs,
|
78
|
+
context=self._retriever,
|
79
|
+
)
|
80
|
+
|
81
|
+
def _add_translation(self, block: TranslatedCodeBlock):
|
82
|
+
if block.translated:
|
83
|
+
return
|
84
|
+
|
85
|
+
if block.original.text is None:
|
86
|
+
block.translated = True
|
87
|
+
return
|
88
|
+
|
89
|
+
if self.eval_items_per_request is None:
|
90
|
+
return super()._add_translation(block)
|
91
|
+
|
92
|
+
input_obj = json.loads(block.original.text)
|
93
|
+
requirements = input_obj.get("requirements", [])
|
94
|
+
|
95
|
+
if not requirements:
|
96
|
+
log.debug(f"[{block.name}] Skipping empty block")
|
97
|
+
block.translated = True
|
98
|
+
block.text = None
|
99
|
+
block.complete = True
|
100
|
+
return
|
101
|
+
|
102
|
+
# For some reason requirements objects are in nested lists?
|
103
|
+
while isinstance(requirements[0], list):
|
104
|
+
requirements = [r for lst in requirements for r in lst]
|
105
|
+
|
106
|
+
if len(requirements) <= self.eval_items_per_request:
|
107
|
+
input_obj["requirements"] = requirements
|
108
|
+
block.original.text = json.dumps(input_obj)
|
109
|
+
return super()._add_translation(block)
|
110
|
+
|
111
|
+
block.processing_time = 0
|
112
|
+
block.cost = 0
|
113
|
+
block.retries = 0
|
114
|
+
obj = {}
|
115
|
+
for i in range(0, len(requirements), self.eval_items_per_request):
|
116
|
+
# Build a new TranslatedBlock using the new working text
|
117
|
+
working_requirements = requirements[i : i + self.eval_items_per_request]
|
118
|
+
working_copy = deepcopy(block.original)
|
119
|
+
working_obj = json.loads(working_copy.text) # type: ignore
|
120
|
+
working_obj["requirements"] = working_requirements
|
121
|
+
working_copy.text = json.dumps(working_obj)
|
122
|
+
working_block = TranslatedCodeBlock(working_copy, self._target_language)
|
123
|
+
|
124
|
+
# Run the LLM on the working text
|
125
|
+
super()._add_translation(working_block)
|
126
|
+
|
127
|
+
# Update metadata to include for all runs
|
128
|
+
block.retries += working_block.retries
|
129
|
+
block.cost += working_block.cost
|
130
|
+
block.processing_time += working_block.processing_time
|
131
|
+
|
132
|
+
# Update the output text to merge this section's output in
|
133
|
+
obj.update(json.loads(working_block.text))
|
134
|
+
|
135
|
+
block.text = json.dumps(obj)
|
136
|
+
block.tokens = self._llm.get_num_tokens(block.text)
|
137
|
+
block.translated = True
|
138
|
+
|
139
|
+
log.debug(
|
140
|
+
f"[{block.name}] Output code:\n{json.dumps(json.loads(block.text), indent=2)}"
|
141
|
+
)
|
142
|
+
|
143
|
+
|
144
|
+
class InlineCommentEvaluator(Evaluator):
|
145
|
+
"""Inline Comment Evaluator
|
146
|
+
|
147
|
+
A class that performs an LLM self evaluation on inline comments,
|
148
|
+
with an associated prompt.
|
149
|
+
"""
|
150
|
+
|
151
|
+
def __init__(self, eval_items_per_request: int | None = None, **kwargs) -> None:
|
152
|
+
"""Initialize the Evaluator class
|
153
|
+
|
154
|
+
Arguments:
|
155
|
+
model: The LLM to use for translation. If an OpenAI model, the
|
156
|
+
`OPENAI_API_KEY` environment variable must be set and the
|
157
|
+
`OPENAI_ORG_ID` environment variable should be set if needed.
|
158
|
+
model_arguments: Additional arguments to pass to the LLM constructor.
|
159
|
+
max_prompts: The maximum number of prompts to try before giving up.
|
160
|
+
"""
|
11
161
|
super().__init__(**kwargs)
|
12
|
-
self.set_prompt("evaluate")
|
13
162
|
self._combiner = JsonCombiner()
|
14
|
-
self._parser = EvaluationParser()
|
15
163
|
self._load_parameters()
|
164
|
+
self._parser = InlineCommentParser()
|
165
|
+
self.set_prompt("eval_prompts/inline_comments")
|
166
|
+
self.eval_items_per_request = eval_items_per_request
|
167
|
+
|
168
|
+
def _add_translation(self, block: TranslatedCodeBlock):
|
169
|
+
if block.translated:
|
170
|
+
return
|
171
|
+
|
172
|
+
if block.original.text is None:
|
173
|
+
block.translated = True
|
174
|
+
return
|
175
|
+
|
176
|
+
if self.eval_items_per_request is None:
|
177
|
+
return super()._add_translation(block)
|
178
|
+
|
179
|
+
comment_pattern = r"<(?:INLINE|BLOCK)_COMMENT \w{8}>.*$"
|
180
|
+
comments = list(
|
181
|
+
re.finditer(comment_pattern, block.original.text, flags=re.MULTILINE)
|
182
|
+
)
|
183
|
+
|
184
|
+
if not comments:
|
185
|
+
log.info(f"[{block.name}] Skipping commentless block")
|
186
|
+
block.translated = True
|
187
|
+
block.text = None
|
188
|
+
block.complete = True
|
189
|
+
return
|
190
|
+
|
191
|
+
if len(comments) <= self.eval_items_per_request:
|
192
|
+
return super()._add_translation(block)
|
193
|
+
|
194
|
+
comment_group_indices = list(range(0, len(comments), self.eval_items_per_request))
|
195
|
+
log.debug(
|
196
|
+
f"[{block.name}] Block contains more than {self.eval_items_per_request}"
|
197
|
+
f" comments, splitting {len(comments)} comments into"
|
198
|
+
f" {len(comment_group_indices)} groups"
|
199
|
+
)
|
200
|
+
|
201
|
+
block.processing_time = 0
|
202
|
+
block.cost = 0
|
203
|
+
block.retries = 0
|
204
|
+
obj = {}
|
205
|
+
for i in range(0, len(comments), self.eval_items_per_request):
|
206
|
+
# Split the text into the section containing comments of interest,
|
207
|
+
# all the text prior to those comments, and all the text after them
|
208
|
+
working_comments = comments[i : i + self.eval_items_per_request]
|
209
|
+
start_idx = working_comments[0].start()
|
210
|
+
end_idx = working_comments[-1].end()
|
211
|
+
prefix = block.original.text[:start_idx]
|
212
|
+
keeper = block.original.text[start_idx:end_idx]
|
213
|
+
suffix = block.original.text[end_idx:]
|
214
|
+
|
215
|
+
# Strip all comment placeholders outside of the section of interest
|
216
|
+
prefix = re.sub(comment_pattern, "", prefix, flags=re.MULTILINE)
|
217
|
+
suffix = re.sub(comment_pattern, "", suffix, flags=re.MULTILINE)
|
218
|
+
|
219
|
+
# Build a new TranslatedBlock using the new working text
|
220
|
+
working_copy = deepcopy(block.original)
|
221
|
+
working_copy.text = prefix + keeper + suffix
|
222
|
+
working_block = TranslatedCodeBlock(working_copy, self._target_language)
|
223
|
+
|
224
|
+
# Run the LLM on the working text
|
225
|
+
super()._add_translation(working_block)
|
226
|
+
|
227
|
+
# Update metadata to include for all runs
|
228
|
+
block.retries += working_block.retries
|
229
|
+
block.cost += working_block.cost
|
230
|
+
block.processing_time += working_block.processing_time
|
231
|
+
|
232
|
+
# Update the output text to merge this section's output in
|
233
|
+
obj.update(json.loads(working_block.text))
|
234
|
+
|
235
|
+
block.text = json.dumps(obj)
|
236
|
+
block.tokens = self._llm.get_num_tokens(block.text)
|
237
|
+
block.translated = True
|
238
|
+
|
239
|
+
log.debug(
|
240
|
+
f"[{block.name}] Output code:\n{json.dumps(json.loads(block.text), indent=2)}"
|
241
|
+
)
|
@@ -20,7 +20,7 @@ class TestAlcSplitter(unittest.TestCase):
|
|
20
20
|
def test_split(self):
|
21
21
|
"""Test the split method."""
|
22
22
|
tree_root = self.splitter.split(self.test_file)
|
23
|
-
self.assertAlmostEqual(tree_root.n_descendents,
|
23
|
+
self.assertAlmostEqual(tree_root.n_descendents, 16, delta=2)
|
24
24
|
self.assertLessEqual(tree_root.max_tokens, self.splitter.max_tokens)
|
25
25
|
self.assertFalse(tree_root.complete)
|
26
26
|
self.combiner.combine_children(tree_root)
|
janus/language/alc/alc.py
CHANGED
@@ -79,10 +79,15 @@ class AlcSplitter(TreeSitterSplitter):
|
|
79
79
|
if len(sects) > 1:
|
80
80
|
block.children = []
|
81
81
|
for sect in sects:
|
82
|
-
|
83
|
-
|
84
|
-
|
85
|
-
|
82
|
+
node_type = sect[0].node_type
|
83
|
+
if node_type in sect_types:
|
84
|
+
if len(sect) == 1:
|
85
|
+
# Don't make a node its own child
|
86
|
+
sect_node = sect[0]
|
87
|
+
else:
|
88
|
+
sect_node = self.merge_nodes(sect)
|
89
|
+
sect_node.children = sect
|
90
|
+
sect_node.node_type = NodeType(str(node_type)[:5])
|
86
91
|
block.children.append(sect_node)
|
87
92
|
else:
|
88
93
|
block.children.extend(sect)
|
janus/language/splitter.py
CHANGED
@@ -275,42 +275,50 @@ class Splitter(FileManager):
|
|
275
275
|
|
276
276
|
groups = [[n] for n in nodes]
|
277
277
|
while len(groups) > 1 and min(adj_sums) <= self.max_tokens and any(merge_allowed):
|
278
|
-
# Get the
|
279
|
-
#
|
278
|
+
# Get the index of the node that would result in the smallest
|
279
|
+
# merged snippet when merged with the node that follows it.
|
280
|
+
# Ignore protected nodes.
|
280
281
|
mergeable_indices = compress(range(len(adj_sums)), merge_allowed)
|
281
|
-
|
282
|
-
|
282
|
+
C = int(min(mergeable_indices, key=adj_sums.__getitem__))
|
283
|
+
|
284
|
+
# C: Central index
|
285
|
+
# L: Index to the left
|
286
|
+
# R: Index to the right (to be merged in to C)
|
287
|
+
# N: Next index (to the right of R, the "new R")
|
288
|
+
L, R, N = C - 1, C + 1, C + 2
|
283
289
|
|
284
290
|
# Recalculate the length. We can't simply use the adj_sum, because
|
285
291
|
# it is an underestimate due to the adjoining suffix/prefix.
|
286
|
-
central_node = groups[
|
287
|
-
merged_text = "".join([text_chunks[
|
292
|
+
central_node = groups[C][-1]
|
293
|
+
merged_text = "".join([text_chunks[C], central_node.suffix, text_chunks[R]])
|
288
294
|
merged_text_length = self._count_tokens(merged_text)
|
289
295
|
|
290
296
|
# If the true length of the merged pair is too long, don't merge them
|
291
297
|
# Instead, correct the estimate, since shorter pairs may yet exist
|
292
298
|
if merged_text_length > self.max_tokens:
|
293
|
-
adj_sums[
|
299
|
+
adj_sums[C] = merged_text_length
|
294
300
|
continue
|
295
301
|
|
296
302
|
# Update adjacent sum estimates
|
297
|
-
if
|
298
|
-
adj_sums[
|
299
|
-
if
|
300
|
-
adj_sums[
|
301
|
-
|
302
|
-
if i0 > 0 and i1 < len(merge_allowed) - 1:
|
303
|
-
if not (merge_allowed[i0 - 1] and merge_allowed[i1 + 1]):
|
304
|
-
merge_allowed[i0 - 1] = merge_allowed[i1 + 1] = False
|
303
|
+
if L >= 0:
|
304
|
+
adj_sums[L] = lengths[L] + merged_text_length
|
305
|
+
if N < len(adj_sums):
|
306
|
+
adj_sums[R] = lengths[N] + merged_text_length
|
305
307
|
|
306
308
|
# The potential merge length for this pair is removed
|
307
|
-
adj_sums.pop(
|
308
|
-
|
309
|
+
adj_sums.pop(C)
|
310
|
+
|
311
|
+
# The merged-in node is removed from the protected list
|
312
|
+
# The merge_allowed list need not be updated - if the node now to
|
313
|
+
# its right is protected, the merge_allowed element corresponding
|
314
|
+
# to the merged neighbor will have been True, and now corresponds
|
315
|
+
# to the merged node.
|
316
|
+
merge_allowed.pop(C)
|
309
317
|
|
310
318
|
# Merge the pair of node groups
|
311
|
-
groups[
|
312
|
-
text_chunks[
|
313
|
-
lengths[
|
319
|
+
groups[C:N] = [groups[C] + groups[R]]
|
320
|
+
text_chunks[C:N] = [merged_text]
|
321
|
+
lengths[C:N] = [merged_text_length]
|
314
322
|
|
315
323
|
return groups
|
316
324
|
|
@@ -403,13 +411,13 @@ class Splitter(FileManager):
|
|
403
411
|
self._split_into_lines(node)
|
404
412
|
|
405
413
|
def _split_into_lines(self, node: CodeBlock):
|
406
|
-
split_text = re.split(r"(\n+)", node.text)
|
414
|
+
split_text = list(re.split(r"(\n+)", node.text))
|
407
415
|
|
408
416
|
# If the string didn't start/end with newlines, make sure to include
|
409
417
|
# empty strings for the prefix/suffixes
|
410
|
-
if
|
418
|
+
if not re.match(r"^\n+$", split_text[0]):
|
411
419
|
split_text = [""] + split_text
|
412
|
-
if split_text[-1]
|
420
|
+
if not re.match(r"^\n+$", split_text[-1]):
|
413
421
|
split_text.append("")
|
414
422
|
betweens = split_text[::2]
|
415
423
|
lines = split_text[1::2]
|
@@ -154,7 +154,15 @@ class TreeSitterSplitter(Splitter):
|
|
154
154
|
The pointer to the language.
|
155
155
|
"""
|
156
156
|
lib = cdll.LoadLibrary(os.fspath(so_file))
|
157
|
-
|
157
|
+
# Added this try-except block to handle the case where the language is not
|
158
|
+
# supported in lowercase by the creator of the grammar. Ex: COBOL
|
159
|
+
# https://github.com/yutaro-sakamoto/tree-sitter-cobol/blob/main/grammar.js#L13
|
160
|
+
try:
|
161
|
+
language_function = getattr(lib, f"tree_sitter_{self.language}")
|
162
|
+
except AttributeError:
|
163
|
+
language = self.language.upper()
|
164
|
+
language_function = getattr(lib, f"tree_sitter_{language}")
|
165
|
+
|
158
166
|
language_function.restype = c_void_p
|
159
167
|
pointer = language_function()
|
160
168
|
return pointer
|
janus/llm/models_info.py
CHANGED
@@ -6,9 +6,13 @@ from typing import Callable, Protocol, TypeVar
|
|
6
6
|
from dotenv import load_dotenv
|
7
7
|
from langchain_community.llms import HuggingFaceTextGenInference
|
8
8
|
from langchain_core.runnables import Runnable
|
9
|
-
from langchain_openai import AzureChatOpenAI
|
9
|
+
from langchain_openai import AzureChatOpenAI, ChatOpenAI
|
10
10
|
|
11
|
-
from janus.llm.model_callbacks import
|
11
|
+
from janus.llm.model_callbacks import (
|
12
|
+
COST_PER_1K_TOKENS,
|
13
|
+
azure_model_reroutes,
|
14
|
+
openai_model_reroutes,
|
15
|
+
)
|
12
16
|
from janus.prompts.prompt import (
|
13
17
|
ChatGptPromptEngine,
|
14
18
|
ClaudePromptEngine,
|
@@ -127,7 +131,7 @@ bedrock_models = [
|
|
127
131
|
all_models = [*azure_models, *bedrock_models]
|
128
132
|
|
129
133
|
MODEL_TYPE_CONSTRUCTORS: dict[str, ModelType] = {
|
130
|
-
|
134
|
+
"OpenAI": ChatOpenAI,
|
131
135
|
"HuggingFace": HuggingFaceTextGenInference,
|
132
136
|
"Azure": AzureChatOpenAI,
|
133
137
|
"Bedrock": Bedrock,
|
@@ -137,7 +141,7 @@ MODEL_TYPE_CONSTRUCTORS: dict[str, ModelType] = {
|
|
137
141
|
|
138
142
|
|
139
143
|
MODEL_PROMPT_ENGINES: dict[str, Callable[..., PromptEngine]] = {
|
140
|
-
|
144
|
+
**{m: ChatGptPromptEngine for m in openai_models},
|
141
145
|
**{m: ChatGptPromptEngine for m in azure_models},
|
142
146
|
**{m: ClaudePromptEngine for m in claude_models},
|
143
147
|
**{m: Llama2PromptEngine for m in llama2_models},
|
@@ -148,7 +152,7 @@ MODEL_PROMPT_ENGINES: dict[str, Callable[..., PromptEngine]] = {
|
|
148
152
|
}
|
149
153
|
|
150
154
|
MODEL_ID_TO_LONG_ID = {
|
151
|
-
|
155
|
+
**{m: mr for m, mr in openai_model_reroutes.items()},
|
152
156
|
**{m: mr for m, mr in azure_model_reroutes.items()},
|
153
157
|
"bedrock-claude-v2": "anthropic.claude-v2",
|
154
158
|
"bedrock-claude-instant-v1": "anthropic.claude-instant-v1",
|
@@ -181,7 +185,7 @@ DEFAULT_MODELS = list(MODEL_DEFAULT_ARGUMENTS.keys())
|
|
181
185
|
MODEL_CONFIG_DIR = Path.home().expanduser() / ".janus" / "llm"
|
182
186
|
|
183
187
|
MODEL_TYPES: dict[str, PromptEngine] = {
|
184
|
-
|
188
|
+
**{m: "OpenAI" for m in openai_models},
|
185
189
|
**{m: "Azure" for m in azure_models},
|
186
190
|
**{m: "BedrockChat" for m in bedrock_models},
|
187
191
|
}
|
@@ -289,15 +293,16 @@ def load_model(model_id) -> JanusModel:
|
|
289
293
|
# log.warning("Waiting 10 seconds...")
|
290
294
|
# Give enough time for the user to read the warnings and cancel
|
291
295
|
# time.sleep(10)
|
292
|
-
raise DeprecationWarning("OpenAI models are no longer supported.")
|
296
|
+
# raise DeprecationWarning("OpenAI models are no longer supported.")
|
293
297
|
|
294
298
|
elif model_type_name == "Azure":
|
295
299
|
model_args.update(
|
296
|
-
|
297
|
-
|
298
|
-
|
299
|
-
|
300
|
-
|
300
|
+
api_key=os.getenv("AZURE_OPENAI_API_KEY"),
|
301
|
+
azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"),
|
302
|
+
api_version=os.getenv("OPENAI_API_VERSION", "2024-02-01"),
|
303
|
+
azure_deployment=model_id,
|
304
|
+
request_timeout=3600,
|
305
|
+
max_tokens=4096,
|
301
306
|
)
|
302
307
|
|
303
308
|
model_type = MODEL_TYPE_CONSTRUCTORS[model_type_name]
|
@@ -0,0 +1,134 @@
|
|
1
|
+
import json
|
2
|
+
import random
|
3
|
+
import uuid
|
4
|
+
from typing import List
|
5
|
+
|
6
|
+
from langchain.output_parsers import PydanticOutputParser
|
7
|
+
from langchain_core.exceptions import OutputParserException
|
8
|
+
from langchain_core.messages import BaseMessage
|
9
|
+
from langchain_core.pydantic_v1 import BaseModel, Field, validator
|
10
|
+
|
11
|
+
from janus.language.block import CodeBlock
|
12
|
+
from janus.parsers.parser import JanusParser
|
13
|
+
from janus.utils.logger import create_logger
|
14
|
+
|
15
|
+
log = create_logger(__name__)
|
16
|
+
RNG = random.Random()
|
17
|
+
|
18
|
+
|
19
|
+
class Criteria(BaseModel):
|
20
|
+
reasoning: str = Field(description="A short explanation for the given assessment")
|
21
|
+
score: str = Field("A simple `pass` or `fail`")
|
22
|
+
|
23
|
+
@validator("score")
|
24
|
+
def score_is_valid(cls, v: str):
|
25
|
+
v = v.lower().strip()
|
26
|
+
if v not in {"pass", "fail"}:
|
27
|
+
raise OutputParserException("Score must be either 'pass' or 'fail'")
|
28
|
+
return v
|
29
|
+
|
30
|
+
|
31
|
+
class Requirement(BaseModel):
|
32
|
+
requirement_id: str = Field(description="The 8-character comment ID")
|
33
|
+
requirement: str = Field(description="The original requirement being evaluated")
|
34
|
+
C1: Criteria
|
35
|
+
C2: Criteria
|
36
|
+
C3: Criteria
|
37
|
+
C4: Criteria
|
38
|
+
C5: Criteria
|
39
|
+
C6: Criteria
|
40
|
+
C7: Criteria
|
41
|
+
C8: Criteria
|
42
|
+
C9: Criteria
|
43
|
+
|
44
|
+
|
45
|
+
class RequirementList(BaseModel):
|
46
|
+
__root__: List[Requirement] = Field(
|
47
|
+
description=(
|
48
|
+
"A list of requirement evaluations. Each element should include"
|
49
|
+
" the requirement's 8-character ID in the `requirement_id` field,"
|
50
|
+
" the original requirement in the 'requirement' field, "
|
51
|
+
" and nine score objects corresponding to each criterion."
|
52
|
+
)
|
53
|
+
)
|
54
|
+
|
55
|
+
|
56
|
+
class IncoseParser(JanusParser, PydanticOutputParser):
|
57
|
+
requirements: dict[str, str]
|
58
|
+
|
59
|
+
def __init__(self):
|
60
|
+
PydanticOutputParser.__init__(
|
61
|
+
self,
|
62
|
+
pydantic_object=RequirementList,
|
63
|
+
requirements={},
|
64
|
+
)
|
65
|
+
|
66
|
+
def parse_input(self, block: CodeBlock) -> str:
|
67
|
+
# TODO: Perform comment stripping/placeholding here rather than in script
|
68
|
+
text = super().parse_input(block)
|
69
|
+
RNG.seed(text)
|
70
|
+
|
71
|
+
obj = json.loads(text)
|
72
|
+
|
73
|
+
# For some reason requirements objects are in a double list?
|
74
|
+
reqs = obj["requirements"]
|
75
|
+
|
76
|
+
# Generate a unique ID for each requirement (ensure they are unique)
|
77
|
+
req_ids = set()
|
78
|
+
while len(req_ids) < len(reqs):
|
79
|
+
req_ids.add(str(uuid.UUID(int=RNG.getrandbits(128), version=4))[:8])
|
80
|
+
|
81
|
+
self.requirements = dict(zip(req_ids, reqs))
|
82
|
+
reqs_str = "\n\n".join(
|
83
|
+
f"Requirement {rid} : {req}" for rid, req in self.requirements.items()
|
84
|
+
)
|
85
|
+
obj["requirements"] = reqs_str
|
86
|
+
return json.dumps(obj)
|
87
|
+
|
88
|
+
def parse(self, text: str | BaseMessage) -> str:
|
89
|
+
if isinstance(text, BaseMessage):
|
90
|
+
text = str(text.content)
|
91
|
+
|
92
|
+
# Strip everything outside the JSON object
|
93
|
+
begin, end = text.find("["), text.rfind("]")
|
94
|
+
text = text[begin : end + 1]
|
95
|
+
|
96
|
+
try:
|
97
|
+
out: RequirementList = super().parse(text)
|
98
|
+
except json.JSONDecodeError as e:
|
99
|
+
log.debug(f"Invalid JSON object. Output:\n{text}")
|
100
|
+
raise OutputParserException(f"Got invalid JSON object. Error: {e}")
|
101
|
+
|
102
|
+
evals: dict[str, dict] = {c.requirement_id: c.dict() for c in out.__root__}
|
103
|
+
|
104
|
+
seen_keys = set(evals.keys())
|
105
|
+
expected_keys = set(self.requirements.keys())
|
106
|
+
missing_keys = expected_keys.difference(seen_keys)
|
107
|
+
invalid_keys = seen_keys.difference(expected_keys)
|
108
|
+
if missing_keys:
|
109
|
+
log.debug(f"Missing keys: {missing_keys}")
|
110
|
+
if invalid_keys:
|
111
|
+
log.debug(f"Invalid keys: {invalid_keys}")
|
112
|
+
log.debug(f"Missing keys: {missing_keys}")
|
113
|
+
raise OutputParserException(
|
114
|
+
f"Got invalid return object. Missing the following expected "
|
115
|
+
f"keys: {missing_keys}"
|
116
|
+
)
|
117
|
+
|
118
|
+
for key in invalid_keys:
|
119
|
+
del evals[key]
|
120
|
+
|
121
|
+
for rid in evals.keys():
|
122
|
+
evals[rid]["requirement"] = self.requirements[rid]
|
123
|
+
evals[rid].pop("requirement_id")
|
124
|
+
|
125
|
+
return json.dumps(evals)
|
126
|
+
|
127
|
+
def parse_combined_output(self, text: str) -> str:
|
128
|
+
if not text.strip():
|
129
|
+
return str({})
|
130
|
+
objs = [json.loads(line.strip()) for line in text.split("\n") if line.strip()]
|
131
|
+
output_obj = {}
|
132
|
+
for obj in objs:
|
133
|
+
output_obj.update(obj)
|
134
|
+
return json.dumps(output_obj)
|
@@ -0,0 +1,112 @@
|
|
1
|
+
import json
|
2
|
+
import re
|
3
|
+
from typing import Any
|
4
|
+
|
5
|
+
from langchain.output_parsers import PydanticOutputParser
|
6
|
+
from langchain_core.exceptions import OutputParserException
|
7
|
+
from langchain_core.messages import BaseMessage
|
8
|
+
from langchain_core.pydantic_v1 import BaseModel, Field, conint
|
9
|
+
|
10
|
+
from janus.language.block import CodeBlock
|
11
|
+
from janus.parsers.parser import JanusParser
|
12
|
+
from janus.utils.logger import create_logger
|
13
|
+
|
14
|
+
log = create_logger(__name__)
|
15
|
+
|
16
|
+
|
17
|
+
class Criteria(BaseModel):
|
18
|
+
reasoning: str = Field(description="A short explanation for the given score")
|
19
|
+
# Constrained to an integer between 1 and 4
|
20
|
+
score: conint(ge=1, le=4) = Field( # type: ignore
|
21
|
+
description="An integer score between 1 and 4 (inclusive), 4 being the best"
|
22
|
+
)
|
23
|
+
|
24
|
+
|
25
|
+
class Comment(BaseModel):
|
26
|
+
comment_id: str = Field(description="The 8-character comment ID")
|
27
|
+
completeness: Criteria = Field(description="The completeness of the comment")
|
28
|
+
hallucination: Criteria = Field(description="The factualness of the comment")
|
29
|
+
readability: Criteria = Field(description="The readability of the comment")
|
30
|
+
usefulness: Criteria = Field(description="The usefulness of the comment")
|
31
|
+
|
32
|
+
|
33
|
+
class CommentList(BaseModel):
|
34
|
+
__root__: list[Comment] = Field(
|
35
|
+
description=(
|
36
|
+
"A list of inline comment evaluations. Each element should include"
|
37
|
+
" the comment's 8-character ID in the `comment_id` field, and four"
|
38
|
+
" score objects corresponding to each metric (`completeness`,"
|
39
|
+
" `hallucination`, `readability`, and `usefulness`)."
|
40
|
+
)
|
41
|
+
)
|
42
|
+
|
43
|
+
|
44
|
+
class InlineCommentParser(JanusParser, PydanticOutputParser):
|
45
|
+
comments: dict[str, str]
|
46
|
+
|
47
|
+
def __init__(self):
|
48
|
+
PydanticOutputParser.__init__(
|
49
|
+
self,
|
50
|
+
pydantic_object=CommentList,
|
51
|
+
comments=[],
|
52
|
+
)
|
53
|
+
|
54
|
+
def parse_input(self, block: CodeBlock) -> str:
|
55
|
+
# TODO: Perform comment stripping/placeholding here rather than in script
|
56
|
+
text = super().parse_input(block)
|
57
|
+
self.comments = dict(
|
58
|
+
re.findall(
|
59
|
+
r"<(?:BLOCK|INLINE)_COMMENT (\w{8})> (.*)$",
|
60
|
+
text,
|
61
|
+
flags=re.MULTILINE,
|
62
|
+
)
|
63
|
+
)
|
64
|
+
return text
|
65
|
+
|
66
|
+
def parse(self, text: str | BaseMessage) -> str:
|
67
|
+
if isinstance(text, BaseMessage):
|
68
|
+
text = str(text.content)
|
69
|
+
|
70
|
+
# Strip everything outside the JSON object
|
71
|
+
begin, end = text.find("["), text.rfind("]")
|
72
|
+
text = text[begin : end + 1]
|
73
|
+
|
74
|
+
try:
|
75
|
+
out: CommentList = super().parse(text)
|
76
|
+
except json.JSONDecodeError as e:
|
77
|
+
log.debug(f"Invalid JSON object. Output:\n{text}")
|
78
|
+
raise OutputParserException(f"Got invalid JSON object. Error: {e}")
|
79
|
+
|
80
|
+
evals: dict[str, Any] = {c.comment_id: c.dict() for c in out.__root__}
|
81
|
+
|
82
|
+
seen_keys = set(evals.keys())
|
83
|
+
expected_keys = set(self.comments.keys())
|
84
|
+
missing_keys = expected_keys.difference(seen_keys)
|
85
|
+
invalid_keys = seen_keys.difference(expected_keys)
|
86
|
+
if missing_keys:
|
87
|
+
log.debug(f"Missing keys: {missing_keys}")
|
88
|
+
if invalid_keys:
|
89
|
+
log.debug(f"Invalid keys: {invalid_keys}")
|
90
|
+
log.debug(f"Missing keys: {missing_keys}")
|
91
|
+
raise OutputParserException(
|
92
|
+
f"Got invalid return object. Missing the following expected "
|
93
|
+
f"keys: {missing_keys}"
|
94
|
+
)
|
95
|
+
|
96
|
+
for key in invalid_keys:
|
97
|
+
del evals[key]
|
98
|
+
|
99
|
+
for cid in evals.keys():
|
100
|
+
evals[cid]["comment"] = self.comments[cid]
|
101
|
+
evals[cid].pop("comment_id")
|
102
|
+
|
103
|
+
return json.dumps(evals)
|
104
|
+
|
105
|
+
def parse_combined_output(self, text: str) -> str:
|
106
|
+
if not text.strip():
|
107
|
+
return str({})
|
108
|
+
objs = [json.loads(line.strip()) for line in text.split("\n") if line.strip()]
|
109
|
+
output_obj = {}
|
110
|
+
for obj in objs:
|
111
|
+
output_obj.update(obj)
|
112
|
+
return json.dumps(output_obj)
|
@@ -36,6 +36,29 @@ class PartitionList(BaseModel):
|
|
36
36
|
)
|
37
37
|
|
38
38
|
|
39
|
+
# The following IDs appear in the prompt example. If the LLM produces them,
|
40
|
+
# they should be ignored
|
41
|
+
EXAMPLE_IDS = {
|
42
|
+
"0d2f4f8d",
|
43
|
+
"def2a953",
|
44
|
+
"75315253",
|
45
|
+
"e7f928da",
|
46
|
+
"1781b2a9",
|
47
|
+
"2fe21e27",
|
48
|
+
"9aef6179",
|
49
|
+
"6061bd82",
|
50
|
+
"22bd0c30",
|
51
|
+
"5d85e19e",
|
52
|
+
"06027969",
|
53
|
+
"91b722fb",
|
54
|
+
"4b3f79be",
|
55
|
+
"k57w964a",
|
56
|
+
"51638s96",
|
57
|
+
"065o6q32",
|
58
|
+
"j5q6p852",
|
59
|
+
}
|
60
|
+
|
61
|
+
|
39
62
|
class PartitionParser(JanusParser, PydanticOutputParser):
|
40
63
|
token_limit: int
|
41
64
|
model: BaseLanguageModel
|
@@ -59,7 +82,10 @@ class PartitionParser(JanusParser, PydanticOutputParser):
|
|
59
82
|
# Generate a unique ID for each line (ensure they are unique)
|
60
83
|
line_ids = set()
|
61
84
|
while len(line_ids) < len(self.lines):
|
62
|
-
|
85
|
+
line_id = str(uuid.UUID(int=RNG.getrandbits(128), version=4))[:8]
|
86
|
+
if line_id in EXAMPLE_IDS:
|
87
|
+
continue
|
88
|
+
line_ids.add(line_id)
|
63
89
|
|
64
90
|
# Prepend each line with the corresponding ID, save the mapping
|
65
91
|
self.line_id_to_index = {lid: i for i, lid in enumerate(line_ids)}
|
@@ -72,18 +98,24 @@ class PartitionParser(JanusParser, PydanticOutputParser):
|
|
72
98
|
if isinstance(text, BaseMessage):
|
73
99
|
text = str(text.content)
|
74
100
|
|
101
|
+
# Strip everything outside the JSON object
|
102
|
+
begin, end = text.find("["), text.rfind("]")
|
103
|
+
text = text[begin : end + 1]
|
104
|
+
|
75
105
|
try:
|
76
106
|
out: PartitionList = super().parse(text)
|
77
107
|
except (OutputParserException, json.JSONDecodeError):
|
78
108
|
log.debug(f"Invalid JSON object. Output:\n{text}")
|
79
109
|
raise
|
80
110
|
|
111
|
+
# Get partition locations, discard reasoning
|
112
|
+
partition_locations = {partition.location for partition in out.__root__}
|
113
|
+
|
114
|
+
# Ignore IDs from the example input
|
115
|
+
partition_locations.difference_update(EXAMPLE_IDS)
|
116
|
+
|
81
117
|
# Locate any invalid line IDs, raise exception if any found
|
82
|
-
invalid_splits =
|
83
|
-
partition.location
|
84
|
-
for partition in out.__root__
|
85
|
-
if partition.location not in self.line_id_to_index
|
86
|
-
]
|
118
|
+
invalid_splits = partition_locations.difference(self.line_id_to_index)
|
87
119
|
if invalid_splits:
|
88
120
|
err_msg = (
|
89
121
|
f"{len(invalid_splits)} line ID(s) not found in input: "
|
@@ -95,9 +127,9 @@ class PartitionParser(JanusParser, PydanticOutputParser):
|
|
95
127
|
# Map line IDs to indices (so they can be sorted and lines indexed)
|
96
128
|
index_to_line_id = {0: "START", None: "END"}
|
97
129
|
split_points = {0}
|
98
|
-
for partition in
|
99
|
-
index = self.line_id_to_index[partition
|
100
|
-
index_to_line_id[index] = partition
|
130
|
+
for partition in partition_locations:
|
131
|
+
index = self.line_id_to_index[partition]
|
132
|
+
index_to_line_id[index] = partition
|
101
133
|
split_points.add(index)
|
102
134
|
|
103
135
|
# Get partition start/ends, chunks, chunk lengths
|
janus/refiners/refiner.py
CHANGED
@@ -2,6 +2,7 @@ import re
|
|
2
2
|
from typing import Any
|
3
3
|
|
4
4
|
from langchain.output_parsers import RetryWithErrorOutputParser
|
5
|
+
from langchain_core.exceptions import OutputParserException
|
5
6
|
from langchain_core.output_parsers import StrOutputParser
|
6
7
|
from langchain_core.prompt_values import PromptValue
|
7
8
|
from langchain_core.runnables import RunnableSerializable
|
@@ -26,6 +27,35 @@ class JanusRefiner(JanusParser):
|
|
26
27
|
raise NotImplementedError
|
27
28
|
|
28
29
|
|
30
|
+
class SimpleRetry(JanusRefiner):
|
31
|
+
max_retries: int
|
32
|
+
retry_chain: RunnableSerializable
|
33
|
+
|
34
|
+
def __init__(
|
35
|
+
self,
|
36
|
+
llm: JanusModel,
|
37
|
+
parser: JanusParser,
|
38
|
+
max_retries: int,
|
39
|
+
):
|
40
|
+
retry_chain = llm | StrOutputParser()
|
41
|
+
super().__init__(
|
42
|
+
retry_chain=retry_chain,
|
43
|
+
parser=parser,
|
44
|
+
max_retries=max_retries,
|
45
|
+
)
|
46
|
+
|
47
|
+
def parse_completion(
|
48
|
+
self, completion: str, prompt_value: PromptValue, **kwargs
|
49
|
+
) -> Any:
|
50
|
+
for retry_number in range(self.max_retries):
|
51
|
+
try:
|
52
|
+
return self.parser.parse(completion)
|
53
|
+
except OutputParserException:
|
54
|
+
completion = self.retry_chain.invoke(prompt_value)
|
55
|
+
|
56
|
+
return self.parser.parse(completion)
|
57
|
+
|
58
|
+
|
29
59
|
class FixParserExceptions(JanusRefiner, RetryWithErrorOutputParser):
|
30
60
|
def __init__(self, llm: JanusModel, parser: JanusParser, max_retries: int):
|
31
61
|
retry_prompt = MODEL_PROMPT_ENGINES[llm.short_model_id](
|
janus/utils/enums.py
CHANGED
@@ -89,6 +89,20 @@ LANGUAGES: Dict[str, Dict[str, Any]] = {
|
|
89
89
|
"url": "https://github.com/stsewd/tree-sitter-comment",
|
90
90
|
"example": "# This is a comment\n",
|
91
91
|
},
|
92
|
+
"cobol": {
|
93
|
+
"comment": "*",
|
94
|
+
"suffix": "cbl",
|
95
|
+
"url": "https://github.com/yutaro-sakamoto/tree-sitter-cobol",
|
96
|
+
"example": (
|
97
|
+
" IDENTIFICATION DIVISION.\n"
|
98
|
+
" PROGRAM-ID. HelloWorld.\n"
|
99
|
+
" ENVIRONMENT DIVISION.\n"
|
100
|
+
" DATA DIVISION.\n"
|
101
|
+
" PROCEDURE DIVISION.\n"
|
102
|
+
' DISPLAY "Hello, World!".\n'
|
103
|
+
" STOP RUN.\n"
|
104
|
+
),
|
105
|
+
},
|
92
106
|
"commonlisp": {
|
93
107
|
"comment": ";;",
|
94
108
|
"suffix": "lisp",
|
@@ -1,17 +1,17 @@
|
|
1
|
-
janus/__init__.py,sha256=
|
1
|
+
janus/__init__.py,sha256=hbiNcSyVowLc5sEqV1GU1B22molrn1w3rOxtKlgrl2E,361
|
2
2
|
janus/__main__.py,sha256=lEkpNtLVPtFo8ySDZeXJ_NXDHb0GVdZFPWB4gD4RPS8,64
|
3
3
|
janus/_tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
4
4
|
janus/_tests/conftest.py,sha256=V7uW-oq3YbFiRPvrq15YoVVrA1n_83pjgiyTZ-IUGW8,963
|
5
5
|
janus/_tests/test_cli.py,sha256=6ef7h11bg4i7Q6L1-r0ZdcY7YrH4n472kvDiA03T4c8,4275
|
6
|
-
janus/cli.py,sha256=
|
6
|
+
janus/cli.py,sha256=zo8EEp0Y33jPCzMUGGRXxjr629ZPMIrVGk3FxinpyDQ,46851
|
7
7
|
janus/converter/__init__.py,sha256=Jnp3TsJ4M1LWDAzXFSyxzMpygbYOxkR-qYxU-G6Gi1k,395
|
8
8
|
janus/converter/_tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
9
9
|
janus/converter/_tests/test_translate.py,sha256=T5CzNrwHqJWfb39Izq84R9WvM3toSlJq31SeA_U7d_4,5641
|
10
10
|
janus/converter/aggregator.py,sha256=MuAXMKmq6PuUo_w6ljyiuDn81Gk2dN-Ci7FVeLc6vhs,1966
|
11
|
-
janus/converter/converter.py,sha256=
|
11
|
+
janus/converter/converter.py,sha256=citSpcCsI1bDfckK38smGNafDHsc8DC9quSoXD2J-Kc,26253
|
12
12
|
janus/converter/diagram.py,sha256=-wktVBPrSBgNIQfHIfa2bJNg6L9CYJQgrr9-xU8DFPw,1646
|
13
13
|
janus/converter/document.py,sha256=qNt2UncMheUBadXCFHGq74tqCrvZub5DCgZpd3Qa54o,4564
|
14
|
-
janus/converter/evaluate.py,sha256=
|
14
|
+
janus/converter/evaluate.py,sha256=Bdue1ESQfMVFFRK4l0CvqwLyzt5bqOKy1LB9a8Hqub0,9150
|
15
15
|
janus/converter/partition.py,sha256=ASvv4hAue44qHobO4kqr_tKr-eJsXCPPdD3NtNd9V-E,993
|
16
16
|
janus/converter/requirements.py,sha256=9tvQ40FZJtG8niIFn45gPQCgKKHVPPoFLinBv6RAqO4,2027
|
17
17
|
janus/converter/translate.py,sha256=S1DPZdmX9Vrn_sJPcobvXmhmS8U53yl5cRXjsmXPtas,4246
|
@@ -30,8 +30,8 @@ janus/language/_tests/test_combine.py,sha256=sjVVPUg4LYkAmazXGUw_S1xPrzWm67_0tCx
|
|
30
30
|
janus/language/_tests/test_splitter.py,sha256=Hqexa39LLEXlK3ZUw7Zot4PUIACvye2vkq0Jaox0T10,373
|
31
31
|
janus/language/alc/__init__.py,sha256=j7vOMGhT1Vri6p8dsjSaY-fkO5uFn0sJ0nrNGGvcizM,42
|
32
32
|
janus/language/alc/_tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
33
|
-
janus/language/alc/_tests/test_alc.py,sha256=
|
34
|
-
janus/language/alc/alc.py,sha256=
|
33
|
+
janus/language/alc/_tests/test_alc.py,sha256=8LKidOPJDlMonRBX9w8AVOKHhyR-O2srW4ntzw5rEEs,1018
|
34
|
+
janus/language/alc/alc.py,sha256=YteDO6DR5hnQULjI3j8Je-w05MH50ZARtXB66FqkZi4,7088
|
35
35
|
janus/language/binary/__init__.py,sha256=AlNAe12ZA366kcGSrQ1FJyOdbwxFqGBFkYR2K6yL818,51
|
36
36
|
janus/language/binary/_tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
37
37
|
janus/language/binary/_tests/test_binary.py,sha256=cIKIxjj6kIY3rcxLwqUPESP9bxWrHqMHx9TNuICgfeQ,1724
|
@@ -52,14 +52,14 @@ janus/language/naive/registry.py,sha256=8YQX1q0IdAm7t69-oC_00I-vfkdRnHuX-OD3KEjE
|
|
52
52
|
janus/language/naive/simple_ast.py,sha256=YzeUJomVsnttJc8tI9eDROb2Hx9Vm9XKmOnLEp3TkzI,3112
|
53
53
|
janus/language/naive/tag_splitter.py,sha256=IXWMn9tBVUGAtzvQi89GhoZ6g7fPXk5MzO0kMCr2mb0,2045
|
54
54
|
janus/language/node.py,sha256=baoYFtapwBQqBtUN6EvHFYRkbR-EcEw1b3fQvH9zIAM,204
|
55
|
-
janus/language/splitter.py,sha256=
|
55
|
+
janus/language/splitter.py,sha256=ZpNIzv0ijbcH7EMnY8DIxAf0ji7-ym1iYJXS9ei_F78,17389
|
56
56
|
janus/language/treesitter/__init__.py,sha256=mUliw7ZJLZ8NkJKyUQMSoUV82hYXE0HvLHrEdGPJF4Q,43
|
57
57
|
janus/language/treesitter/_tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
58
58
|
janus/language/treesitter/_tests/test_treesitter.py,sha256=fmr_mFSja7vaCVu0TVyLDua3A94jMjY4AqSC5NqnOdQ,2179
|
59
|
-
janus/language/treesitter/treesitter.py,sha256=
|
59
|
+
janus/language/treesitter/treesitter.py,sha256=FdsBO8CEo6l9D77aHXns5jRSoZzkvrRGZFCW3oNw15c,7928
|
60
60
|
janus/llm/__init__.py,sha256=TKLYvnsWKWfxMucy-lCLQ-4bkN9ENotJZDywDEQmrKg,45
|
61
61
|
janus/llm/model_callbacks.py,sha256=cHRZBpYgAwiYbA2k0GQ7DBwBFQZJpEGMUBV3Q_5GTpU,7940
|
62
|
-
janus/llm/models_info.py,sha256=
|
62
|
+
janus/llm/models_info.py,sha256=tHH5Hf7zWBpD5zSuhxx_Tp1fQMPTKPr9EuevacDiUTU,10711
|
63
63
|
janus/metrics/__init__.py,sha256=AsxtZJUzZiXJPr2ehPPltuYP-ddechjg6X85WZUO7mA,241
|
64
64
|
janus/metrics/_tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
65
65
|
janus/metrics/_tests/reference.py,sha256=hiaJPP9CXkvFBV_wL-gOe_BzELTw0nvB6uCxhxtIiE8,13
|
@@ -89,25 +89,27 @@ janus/parsers/_tests/test_code_parser.py,sha256=3ay5QpUPcynX_EJ-YLl3PR28poutUkT7
|
|
89
89
|
janus/parsers/code_parser.py,sha256=3l0HfzgrvJuiwk779s9ZsgUl3xbp1nE1qZxh8aDYRBI,873
|
90
90
|
janus/parsers/doc_parser.py,sha256=0pUsNZ9hKQLjIi8L8BgkOBHQZ_EGoFLHrBQ4hoDkjSw,5862
|
91
91
|
janus/parsers/eval_parser.py,sha256=Gjh6aTZgpYd2ASJUEPMo4LpCL00cBmbOqc4KM3hy8x8,2922
|
92
|
+
janus/parsers/eval_parsers/incose_parser.py,sha256=udyK-24ocfrB1SzmggcERm73dBynrCj4MFSBV8k7YDM,4478
|
93
|
+
janus/parsers/eval_parsers/inline_comment_parser.py,sha256=QzKgzeWPhyIEkLxJBpeutSocSJjjXEcWRRS635bXEO8,3973
|
92
94
|
janus/parsers/parser.py,sha256=y6VV64bgVidf-oEFla3I--_28tnJsPBc6QUD_SkbfSE,1614
|
93
|
-
janus/parsers/partition_parser.py,sha256=
|
95
|
+
janus/parsers/partition_parser.py,sha256=IW5_aNYL4g-PzB_qJ0g0NlwLiaAGGewR5iUYF19PVL4,5738
|
94
96
|
janus/parsers/reqs_parser.py,sha256=uRQC41Iqp22GjIvakb5UKv70UWHkcOTbOVl_RDnipYw,2438
|
95
97
|
janus/parsers/uml.py,sha256=SwaoG9QrHKQP8rSxlf3qu_rp7OMQqYSmLgDYBapOa9M,3379
|
96
98
|
janus/prompts/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
97
99
|
janus/prompts/prompt.py,sha256=3796YXIzzIec9b0iUzd8VZlq-AdQbzq8qUGXLy4KH-0,10586
|
98
|
-
janus/refiners/refiner.py,sha256=
|
100
|
+
janus/refiners/refiner.py,sha256=ZHP0hUIv8eLpHJSd2SP1Sex6q6SdJgH7HIPgXPBw_gI,4672
|
99
101
|
janus/refiners/uml.py,sha256=ZFvFLxOdbolYuOmZh_8K6kiHCWKuudqP71sr_TammxM,866
|
100
102
|
janus/retrievers/retriever.py,sha256=n6MzoNZs0GJCH4eqQPS3gFlVHZ3eETr7FuHYbyPzTuo,3506
|
101
103
|
janus/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
102
104
|
janus/utils/_tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
103
105
|
janus/utils/_tests/test_logger.py,sha256=jkkvrCTKwsFCsZtmyuvc-WJ0rC7LJi2Z91sIe4IiKzA,2209
|
104
106
|
janus/utils/_tests/test_progress.py,sha256=Rs_u5PiGjP-L-o6C1fhwfE1ig8jYu9Xo9s4p8yPysl8,491
|
105
|
-
janus/utils/enums.py,sha256=
|
107
|
+
janus/utils/enums.py,sha256=gmvX3MYnHAwu4ZypidENIZ27M5NI_YegY3PpCDJS34Q,28094
|
106
108
|
janus/utils/logger.py,sha256=KZeuaMAnlSZCsj4yL0P6N-JzZwpxXygzACWfdZFeuek,2337
|
107
109
|
janus/utils/pdf_docs_reader.py,sha256=beMKHdYrFwg0m_i7n0OTJrut3sf4rEWFd7P_80A76WY,5140
|
108
110
|
janus/utils/progress.py,sha256=PIpcQec7SrhsfqB25LHj2CDDkfm9umZx90d9LZnAx6k,1469
|
109
|
-
janus_llm-4.
|
110
|
-
janus_llm-4.
|
111
|
-
janus_llm-4.
|
112
|
-
janus_llm-4.
|
113
|
-
janus_llm-4.
|
111
|
+
janus_llm-4.3.1.dist-info/LICENSE,sha256=_j0st0a-HB6MRbP3_BW3PUqpS16v54luyy-1zVyl8NU,10789
|
112
|
+
janus_llm-4.3.1.dist-info/METADATA,sha256=ZeUGDDKbJjHSk2Wkzf-4zXLIwaYZqua-5_HVFbzV2yg,4574
|
113
|
+
janus_llm-4.3.1.dist-info/WHEEL,sha256=Nq82e9rUAnEjt98J6MlVmMCZb-t9cYE2Ir1kpBmnWfs,88
|
114
|
+
janus_llm-4.3.1.dist-info/entry_points.txt,sha256=OGhQwzj6pvXp79B0SaBD5apGekCu7Dwe9fZZT_TZ544,39
|
115
|
+
janus_llm-4.3.1.dist-info/RECORD,,
|
File without changes
|
File without changes
|
File without changes
|