rapidata 2.35.0__py3-none-any.whl → 2.35.1__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Potentially problematic release.
This version of rapidata might be problematic. Click here for more details.
- rapidata/__init__.py +1 -1
- rapidata/rapidata_client/benchmark/_detail_mapper.py +0 -2
- rapidata/rapidata_client/benchmark/leaderboard/rapidata_leaderboard.py +33 -1
- rapidata/rapidata_client/benchmark/rapidata_benchmark.py +11 -1
- {rapidata-2.35.0.dist-info → rapidata-2.35.1.dist-info}/METADATA +1 -1
- {rapidata-2.35.0.dist-info → rapidata-2.35.1.dist-info}/RECORD +8 -8
- {rapidata-2.35.0.dist-info → rapidata-2.35.1.dist-info}/LICENSE +0 -0
- {rapidata-2.35.0.dist-info → rapidata-2.35.1.dist-info}/WHEEL +0 -0
rapidata/__init__.py
CHANGED
|
@@ -31,6 +31,7 @@ class RapidataLeaderboard:
|
|
|
31
31
|
show_prompt_asset: bool,
|
|
32
32
|
inverse_ranking: bool,
|
|
33
33
|
response_budget: int,
|
|
34
|
+
min_responses_per_matchup: int,
|
|
34
35
|
id: str,
|
|
35
36
|
openapi_service: OpenAPIService,
|
|
36
37
|
):
|
|
@@ -41,6 +42,7 @@ class RapidataLeaderboard:
|
|
|
41
42
|
self.__show_prompt_asset = show_prompt_asset
|
|
42
43
|
self.__inverse_ranking = inverse_ranking
|
|
43
44
|
self.__response_budget = response_budget
|
|
45
|
+
self.__min_responses_per_matchup = min_responses_per_matchup
|
|
44
46
|
self.id = id
|
|
45
47
|
|
|
46
48
|
@property
|
|
@@ -62,11 +64,41 @@ class RapidataLeaderboard:
|
|
|
62
64
|
leaderboard_id=self.id,
|
|
63
65
|
update_leaderboard_response_config_model=UpdateLeaderboardResponseConfigModel(
|
|
64
66
|
responseBudget=DetailMapper.get_budget(level_of_detail),
|
|
65
|
-
minResponses=
|
|
67
|
+
minResponses=self.__min_responses_per_matchup,
|
|
66
68
|
),
|
|
67
69
|
)
|
|
68
70
|
self.__response_budget = DetailMapper.get_budget(level_of_detail)
|
|
69
71
|
|
|
72
|
+
@property
|
|
73
|
+
def min_responses_per_matchup(self) -> int:
|
|
74
|
+
"""
|
|
75
|
+
Returns the minimum number of responses required to be considered for the leaderboard.
|
|
76
|
+
"""
|
|
77
|
+
return self.__min_responses_per_matchup
|
|
78
|
+
|
|
79
|
+
@min_responses_per_matchup.setter
|
|
80
|
+
def min_responses_per_matchup(self, min_responses: int):
|
|
81
|
+
"""
|
|
82
|
+
Sets the minimum number of responses required to be considered for the leaderboard.
|
|
83
|
+
"""
|
|
84
|
+
if not isinstance(min_responses, int):
|
|
85
|
+
raise ValueError("Min responses per matchup must be an integer")
|
|
86
|
+
|
|
87
|
+
if min_responses < 3:
|
|
88
|
+
raise ValueError("Min responses per matchup must be at least 3")
|
|
89
|
+
|
|
90
|
+
logger.debug(
|
|
91
|
+
f"Setting min responses per matchup to {min_responses} for leaderboard {self.name}"
|
|
92
|
+
)
|
|
93
|
+
self.__openapi_service.leaderboard_api.leaderboard_leaderboard_id_response_config_put(
|
|
94
|
+
leaderboard_id=self.id,
|
|
95
|
+
update_leaderboard_response_config_model=UpdateLeaderboardResponseConfigModel(
|
|
96
|
+
responseBudget=self.__response_budget,
|
|
97
|
+
minResponses=min_responses,
|
|
98
|
+
),
|
|
99
|
+
)
|
|
100
|
+
self.__min_responses_per_matchup = min_responses
|
|
101
|
+
|
|
70
102
|
@property
|
|
71
103
|
def show_prompt_asset(self) -> bool:
|
|
72
104
|
"""
|
|
@@ -172,6 +172,7 @@ class RapidataBenchmark:
|
|
|
172
172
|
leaderboard.show_prompt_asset,
|
|
173
173
|
leaderboard.is_inversed,
|
|
174
174
|
leaderboard.response_budget,
|
|
175
|
+
leaderboard.min_responses,
|
|
175
176
|
leaderboard.id,
|
|
176
177
|
self.__openapi_service,
|
|
177
178
|
)
|
|
@@ -258,6 +259,7 @@ class RapidataBenchmark:
|
|
|
258
259
|
show_prompt_asset: bool = False,
|
|
259
260
|
inverse_ranking: bool = False,
|
|
260
261
|
level_of_detail: Literal["low", "medium", "high", "very high"] = "low",
|
|
262
|
+
min_responses_per_matchup: int = 3,
|
|
261
263
|
) -> RapidataLeaderboard:
|
|
262
264
|
"""
|
|
263
265
|
Creates a new leaderboard for the benchmark.
|
|
@@ -269,7 +271,14 @@ class RapidataBenchmark:
|
|
|
269
271
|
show_prompt_asset: Whether to show the prompt asset to the users. (only works if the prompt asset is a URL) (default: False)
|
|
270
272
|
inverse_ranking: Whether to inverse the ranking of the leaderboard. (if the question is inversed, e.g. "Which video is worse?")
|
|
271
273
|
level_of_detail: The level of detail of the leaderboard. This will effect how many comparisons are done per model evaluation. (default: "low")
|
|
274
|
+
min_responses_per_matchup: The minimum number of responses required to be considered for the leaderboard. (default: 3)
|
|
272
275
|
"""
|
|
276
|
+
if not isinstance(min_responses_per_matchup, int):
|
|
277
|
+
raise ValueError("Min responses per matchup must be an integer")
|
|
278
|
+
|
|
279
|
+
if min_responses_per_matchup < 3:
|
|
280
|
+
raise ValueError("Min responses per matchup must be at least 3")
|
|
281
|
+
|
|
273
282
|
leaderboard_result = self.__openapi_service.leaderboard_api.leaderboard_post(
|
|
274
283
|
create_leaderboard_model=CreateLeaderboardModel(
|
|
275
284
|
benchmarkId=self.id,
|
|
@@ -278,7 +287,7 @@ class RapidataBenchmark:
|
|
|
278
287
|
showPrompt=show_prompt,
|
|
279
288
|
showPromptAsset=show_prompt_asset,
|
|
280
289
|
isInversed=inverse_ranking,
|
|
281
|
-
minResponses=
|
|
290
|
+
minResponses=min_responses_per_matchup,
|
|
282
291
|
responseBudget=DetailMapper.get_budget(level_of_detail),
|
|
283
292
|
)
|
|
284
293
|
)
|
|
@@ -294,6 +303,7 @@ class RapidataBenchmark:
|
|
|
294
303
|
show_prompt_asset,
|
|
295
304
|
inverse_ranking,
|
|
296
305
|
leaderboard_result.response_budget,
|
|
306
|
+
min_responses_per_matchup,
|
|
297
307
|
leaderboard_result.id,
|
|
298
308
|
self.__openapi_service,
|
|
299
309
|
)
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
rapidata/__init__.py,sha256=
|
|
1
|
+
rapidata/__init__.py,sha256=7ZoBaYqvNh-HjfP3kpiQCpwmvJYhFGdAVtpwwzyhLsQ,897
|
|
2
2
|
rapidata/api_client/__init__.py,sha256=fb2lqv3sj48wAgarp3g6hvtTgd7bfI01DJJLuBZQnFI,34850
|
|
3
3
|
rapidata/api_client/api/__init__.py,sha256=dGnSE9oPO_ahGh-E1jtw4_VuM_vQueQFuv0IVMQo6uo,1546
|
|
4
4
|
rapidata/api_client/api/benchmark_api.py,sha256=bC8hAPgHIDU5u1e0loWPWnZX33BW6gsAR8oc5199q2k,129777
|
|
@@ -542,12 +542,12 @@ rapidata/rapidata_client/__init__.py,sha256=CfkQxCdURXzJsVP6sxKmufze2u-IE_snG_G8
|
|
|
542
542
|
rapidata/rapidata_client/api/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
543
543
|
rapidata/rapidata_client/api/rapidata_exception.py,sha256=BIdmHRrJUGW-Mqhp1H_suemZaR6w9TgjWq-ZW5iUPdQ,3878
|
|
544
544
|
rapidata/rapidata_client/benchmark/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
545
|
-
rapidata/rapidata_client/benchmark/_detail_mapper.py,sha256=
|
|
545
|
+
rapidata/rapidata_client/benchmark/_detail_mapper.py,sha256=HmzJwR2dojs0c2PaEJ5lwSPtjaqeoPXI7RQk5kwSB9A,944
|
|
546
546
|
rapidata/rapidata_client/benchmark/leaderboard/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
547
|
-
rapidata/rapidata_client/benchmark/leaderboard/rapidata_leaderboard.py,sha256=
|
|
547
|
+
rapidata/rapidata_client/benchmark/leaderboard/rapidata_leaderboard.py,sha256=xRNXw__K4-4wb4UI-EgLyJRwGO5OkFdoN3A1S3MkdF4,6092
|
|
548
548
|
rapidata/rapidata_client/benchmark/participant/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
549
549
|
rapidata/rapidata_client/benchmark/participant/_participant.py,sha256=yN82EWrZXYszsM8Ns0HRMXCTivltkyxcpGRK-cdT01Y,3683
|
|
550
|
-
rapidata/rapidata_client/benchmark/rapidata_benchmark.py,sha256=
|
|
550
|
+
rapidata/rapidata_client/benchmark/rapidata_benchmark.py,sha256=Z0jT9hiWyS3km0BwtGXbdmtopdnecf1z-ucDdwg06Y0,14793
|
|
551
551
|
rapidata/rapidata_client/benchmark/rapidata_benchmark_manager.py,sha256=NnXCsU2_8SlGxdI51MeIUWK7Ku0pEylDSPhnzOPzTV0,5291
|
|
552
552
|
rapidata/rapidata_client/country_codes/__init__.py,sha256=FB9Dcks44J6C6YBSYmTmNZ71tE130x6NO_3aLJ8fKzQ,40
|
|
553
553
|
rapidata/rapidata_client/country_codes/country_codes.py,sha256=ePHqeb7y9DWQZAnddBzPx1puYBcrgUjdR2sbFijuFD8,283
|
|
@@ -651,7 +651,7 @@ rapidata/service/__init__.py,sha256=s9bS1AJZaWIhLtJX_ZA40_CK39rAAkwdAmymTMbeWl4,
|
|
|
651
651
|
rapidata/service/credential_manager.py,sha256=pUEEtp6VrFWYhfUUtyqmS0AlRqe2Y0kFkY6o22IT4KM,8682
|
|
652
652
|
rapidata/service/local_file_service.py,sha256=pgorvlWcx52Uh3cEG6VrdMK_t__7dacQ_5AnfY14BW8,877
|
|
653
653
|
rapidata/service/openapi_service.py,sha256=v2fhPbHmD0J11ZRZY6f80PdIdGwpRFlbfMH9t8Ypg5A,5403
|
|
654
|
-
rapidata-2.35.
|
|
655
|
-
rapidata-2.35.
|
|
656
|
-
rapidata-2.35.
|
|
657
|
-
rapidata-2.35.
|
|
654
|
+
rapidata-2.35.1.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
|
|
655
|
+
rapidata-2.35.1.dist-info/METADATA,sha256=mQGouVcYnBiJg-noTAW2JpQPxqSId68ZqIznzJDyJWE,1264
|
|
656
|
+
rapidata-2.35.1.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
|
|
657
|
+
rapidata-2.35.1.dist-info/RECORD,,
|
|
File without changes
|
|
File without changes
|