Spaces:
Sleeping
Sleeping
Update pyproject.toml and apply ruff
Browse files- pyproject.toml +1 -1
- uv.lock +50 -50
- yourbench_space/app.py +2 -2
- yourbench_space/evaluation.py +2 -2
- yourbench_space/leaderboard_space/utils.py +11 -9
pyproject.toml
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
[project]
|
2 |
-
name = "yourbench-
|
3 |
version = "0.1.0"
|
4 |
authors = [
|
5 |
{ name = "Sumuk Shashidhar", email = "[email protected]" },
|
|
|
1 |
[project]
|
2 |
+
name = "yourbench-advanced"
|
3 |
version = "0.1.0"
|
4 |
authors = [
|
5 |
{ name = "Sumuk Shashidhar", email = "[email protected]" },
|
uv.lock
CHANGED
@@ -12,7 +12,7 @@ wheels = [
|
|
12 |
|
13 |
[[package]]
|
14 |
name = "accelerate"
|
15 |
-
version = "1.
|
16 |
source = { registry = "https://pypi.org/simple" }
|
17 |
dependencies = [
|
18 |
{ name = "huggingface-hub" },
|
@@ -23,9 +23,9 @@ dependencies = [
|
|
23 |
{ name = "safetensors" },
|
24 |
{ name = "torch" },
|
25 |
]
|
26 |
-
sdist = { url = "https://files.pythonhosted.org/packages/
|
27 |
wheels = [
|
28 |
-
{ url = "https://files.pythonhosted.org/packages/
|
29 |
]
|
30 |
|
31 |
[[package]]
|
@@ -57,7 +57,7 @@ wheels = [
|
|
57 |
|
58 |
[[package]]
|
59 |
name = "aiohttp"
|
60 |
-
version = "3.11.
|
61 |
source = { registry = "https://pypi.org/simple" }
|
62 |
dependencies = [
|
63 |
{ name = "aiohappyeyeballs" },
|
@@ -68,24 +68,24 @@ dependencies = [
|
|
68 |
{ name = "propcache" },
|
69 |
{ name = "yarl" },
|
70 |
]
|
71 |
-
sdist = { url = "https://files.pythonhosted.org/packages/
|
72 |
wheels = [
|
73 |
-
{ url = "https://files.pythonhosted.org/packages/
|
74 |
-
{ url = "https://files.pythonhosted.org/packages/
|
75 |
-
{ url = "https://files.pythonhosted.org/packages/
|
76 |
-
{ url = "https://files.pythonhosted.org/packages/
|
77 |
-
{ url = "https://files.pythonhosted.org/packages/
|
78 |
-
{ url = "https://files.pythonhosted.org/packages/
|
79 |
-
{ url = "https://files.pythonhosted.org/packages/
|
80 |
-
{ url = "https://files.pythonhosted.org/packages/
|
81 |
-
{ url = "https://files.pythonhosted.org/packages/
|
82 |
-
{ url = "https://files.pythonhosted.org/packages/
|
83 |
-
{ url = "https://files.pythonhosted.org/packages/
|
84 |
-
{ url = "https://files.pythonhosted.org/packages/
|
85 |
-
{ url = "https://files.pythonhosted.org/packages/
|
86 |
-
{ url = "https://files.pythonhosted.org/packages/
|
87 |
-
{ url = "https://files.pythonhosted.org/packages/
|
88 |
-
{ url = "https://files.pythonhosted.org/packages/
|
89 |
]
|
90 |
|
91 |
[[package]]
|
@@ -152,14 +152,14 @@ wheels = [
|
|
152 |
|
153 |
[[package]]
|
154 |
name = "authlib"
|
155 |
-
version = "1.5.
|
156 |
source = { registry = "https://pypi.org/simple" }
|
157 |
dependencies = [
|
158 |
{ name = "cryptography" },
|
159 |
]
|
160 |
-
sdist = { url = "https://files.pythonhosted.org/packages/
|
161 |
wheels = [
|
162 |
-
{ url = "https://files.pythonhosted.org/packages/
|
163 |
]
|
164 |
|
165 |
[[package]]
|
@@ -573,7 +573,7 @@ wheels = [
|
|
573 |
|
574 |
[[package]]
|
575 |
name = "gradio"
|
576 |
-
version = "5.23.
|
577 |
source = { registry = "https://pypi.org/simple" }
|
578 |
dependencies = [
|
579 |
{ name = "aiofiles" },
|
@@ -605,9 +605,9 @@ dependencies = [
|
|
605 |
{ name = "urllib3", marker = "sys_platform == 'emscripten'" },
|
606 |
{ name = "uvicorn", marker = "sys_platform != 'emscripten'" },
|
607 |
]
|
608 |
-
sdist = { url = "https://files.pythonhosted.org/packages/
|
609 |
wheels = [
|
610 |
-
{ url = "https://files.pythonhosted.org/packages/
|
611 |
]
|
612 |
|
613 |
[package.optional-dependencies]
|
@@ -703,7 +703,7 @@ wheels = [
|
|
703 |
|
704 |
[[package]]
|
705 |
name = "huggingface-hub"
|
706 |
-
version = "0.
|
707 |
source = { registry = "https://pypi.org/simple" }
|
708 |
dependencies = [
|
709 |
{ name = "filelock" },
|
@@ -714,9 +714,9 @@ dependencies = [
|
|
714 |
{ name = "tqdm" },
|
715 |
{ name = "typing-extensions" },
|
716 |
]
|
717 |
-
sdist = { url = "https://files.pythonhosted.org/packages/
|
718 |
wheels = [
|
719 |
-
{ url = "https://files.pythonhosted.org/packages/
|
720 |
]
|
721 |
|
722 |
[package.optional-dependencies]
|
@@ -1074,26 +1074,26 @@ wheels = [
|
|
1074 |
|
1075 |
[[package]]
|
1076 |
name = "multidict"
|
1077 |
-
version = "6.3.
|
1078 |
source = { registry = "https://pypi.org/simple" }
|
1079 |
-
sdist = { url = "https://files.pythonhosted.org/packages/
|
1080 |
wheels = [
|
1081 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1082 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1083 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1084 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1085 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1086 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1087 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1088 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1089 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1090 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1091 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1092 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1093 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1094 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1095 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1096 |
-
{ url = "https://files.pythonhosted.org/packages/
|
1097 |
]
|
1098 |
|
1099 |
[[package]]
|
@@ -2348,7 +2348,7 @@ wheels = [
|
|
2348 |
[[package]]
|
2349 |
name = "yourbench"
|
2350 |
version = "0.2.0"
|
2351 |
-
source = { git = "https://github.com/huggingface/yourbench.git?branch=main#
|
2352 |
dependencies = [
|
2353 |
{ name = "asyncio" },
|
2354 |
{ name = "bert-score" },
|
@@ -2373,7 +2373,7 @@ dependencies = [
|
|
2373 |
]
|
2374 |
|
2375 |
[[package]]
|
2376 |
-
name = "yourbench-
|
2377 |
version = "0.1.0"
|
2378 |
source = { editable = "." }
|
2379 |
dependencies = [
|
|
|
12 |
|
13 |
[[package]]
|
14 |
name = "accelerate"
|
15 |
+
version = "1.6.0"
|
16 |
source = { registry = "https://pypi.org/simple" }
|
17 |
dependencies = [
|
18 |
{ name = "huggingface-hub" },
|
|
|
23 |
{ name = "safetensors" },
|
24 |
{ name = "torch" },
|
25 |
]
|
26 |
+
sdist = { url = "https://files.pythonhosted.org/packages/8a/6e/c29a1dcde7db07f47870ed63e5124086b11874ad52ccd533dc1ca2c799da/accelerate-1.6.0.tar.gz", hash = "sha256:28c1ef1846e690944f98b68dc7b8bb6c51d032d45e85dcbb3adb0c8b99dffb32", size = 363804 }
|
27 |
wheels = [
|
28 |
+
{ url = "https://files.pythonhosted.org/packages/63/b1/8198e3cdd11a426b1df2912e3381018c4a4a55368f6d0857ba3ca418ef93/accelerate-1.6.0-py3-none-any.whl", hash = "sha256:1aee717d3d3735ad6d09710a7c26990ee4652b79b4e93df46551551b5227c2aa", size = 354748 },
|
29 |
]
|
30 |
|
31 |
[[package]]
|
|
|
57 |
|
58 |
[[package]]
|
59 |
name = "aiohttp"
|
60 |
+
version = "3.11.16"
|
61 |
source = { registry = "https://pypi.org/simple" }
|
62 |
dependencies = [
|
63 |
{ name = "aiohappyeyeballs" },
|
|
|
68 |
{ name = "propcache" },
|
69 |
{ name = "yarl" },
|
70 |
]
|
71 |
+
sdist = { url = "https://files.pythonhosted.org/packages/f1/d9/1c4721d143e14af753f2bf5e3b681883e1f24b592c0482df6fa6e33597fa/aiohttp-3.11.16.tar.gz", hash = "sha256:16f8a2c9538c14a557b4d309ed4d0a7c60f0253e8ed7b6c9a2859a7582f8b1b8", size = 7676826 }
|
72 |
wheels = [
|
73 |
+
{ url = "https://files.pythonhosted.org/packages/db/38/100d01cbc60553743baf0fba658cb125f8ad674a8a771f765cdc155a890d/aiohttp-3.11.16-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:911a6e91d08bb2c72938bc17f0a2d97864c531536b7832abee6429d5296e5b27", size = 704881 },
|
74 |
+
{ url = "https://files.pythonhosted.org/packages/21/ed/b4102bb6245e36591209e29f03fe87e7956e54cb604ee12e20f7eb47f994/aiohttp-3.11.16-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:6ac13b71761e49d5f9e4d05d33683bbafef753e876e8e5a7ef26e937dd766713", size = 464564 },
|
75 |
+
{ url = "https://files.pythonhosted.org/packages/3b/e1/a9ab6c47b62ecee080eeb33acd5352b40ecad08fb2d0779bcc6739271745/aiohttp-3.11.16-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:fd36c119c5d6551bce374fcb5c19269638f8d09862445f85a5a48596fd59f4bb", size = 456548 },
|
76 |
+
{ url = "https://files.pythonhosted.org/packages/80/ad/216c6f71bdff2becce6c8776f0aa32cb0fa5d83008d13b49c3208d2e4016/aiohttp-3.11.16-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d489d9778522fbd0f8d6a5c6e48e3514f11be81cb0a5954bdda06f7e1594b321", size = 1691749 },
|
77 |
+
{ url = "https://files.pythonhosted.org/packages/bd/ea/7df7bcd3f4e734301605f686ffc87993f2d51b7acb6bcc9b980af223f297/aiohttp-3.11.16-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:69a2cbd61788d26f8f1e626e188044834f37f6ae3f937bd9f08b65fc9d7e514e", size = 1736874 },
|
78 |
+
{ url = "https://files.pythonhosted.org/packages/51/41/c7724b9c87a29b7cfd1202ec6446bae8524a751473d25e2ff438bc9a02bf/aiohttp-3.11.16-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:cd464ba806e27ee24a91362ba3621bfc39dbbb8b79f2e1340201615197370f7c", size = 1786885 },
|
79 |
+
{ url = "https://files.pythonhosted.org/packages/86/b3/f61f8492fa6569fa87927ad35a40c159408862f7e8e70deaaead349e2fba/aiohttp-3.11.16-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:1ce63ae04719513dd2651202352a2beb9f67f55cb8490c40f056cea3c5c355ce", size = 1698059 },
|
80 |
+
{ url = "https://files.pythonhosted.org/packages/ce/be/7097cf860a9ce8bbb0e8960704e12869e111abcd3fbd245153373079ccec/aiohttp-3.11.16-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:09b00dd520d88eac9d1768439a59ab3d145065c91a8fab97f900d1b5f802895e", size = 1626527 },
|
81 |
+
{ url = "https://files.pythonhosted.org/packages/1d/1d/aaa841c340e8c143a8d53a1f644c2a2961c58cfa26e7b398d6bf75cf5d23/aiohttp-3.11.16-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:7f6428fee52d2bcf96a8aa7b62095b190ee341ab0e6b1bcf50c615d7966fd45b", size = 1644036 },
|
82 |
+
{ url = "https://files.pythonhosted.org/packages/2c/88/59d870f76e9345e2b149f158074e78db457985c2b4da713038d9da3020a8/aiohttp-3.11.16-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:13ceac2c5cdcc3f64b9015710221ddf81c900c5febc505dbd8f810e770011540", size = 1685270 },
|
83 |
+
{ url = "https://files.pythonhosted.org/packages/2b/b1/c6686948d4c79c3745595efc469a9f8a43cab3c7efc0b5991be65d9e8cb8/aiohttp-3.11.16-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:fadbb8f1d4140825069db3fedbbb843290fd5f5bc0a5dbd7eaf81d91bf1b003b", size = 1650852 },
|
84 |
+
{ url = "https://files.pythonhosted.org/packages/fe/94/3e42a6916fd3441721941e0f1b8438e1ce2a4c49af0e28e0d3c950c9b3c9/aiohttp-3.11.16-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:6a792ce34b999fbe04a7a71a90c74f10c57ae4c51f65461a411faa70e154154e", size = 1704481 },
|
85 |
+
{ url = "https://files.pythonhosted.org/packages/b1/6d/6ab5854ff59b27075c7a8c610597d2b6c38945f9a1284ee8758bc3720ff6/aiohttp-3.11.16-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:f4065145bf69de124accdd17ea5f4dc770da0a6a6e440c53f6e0a8c27b3e635c", size = 1735370 },
|
86 |
+
{ url = "https://files.pythonhosted.org/packages/73/2a/08a68eec3c99a6659067d271d7553e4d490a0828d588e1daa3970dc2b771/aiohttp-3.11.16-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:fa73e8c2656a3653ae6c307b3f4e878a21f87859a9afab228280ddccd7369d71", size = 1697619 },
|
87 |
+
{ url = "https://files.pythonhosted.org/packages/61/d5/fea8dbbfb0cd68fbb56f0ae913270a79422d9a41da442a624febf72d2aaf/aiohttp-3.11.16-cp312-cp312-win32.whl", hash = "sha256:f244b8e541f414664889e2c87cac11a07b918cb4b540c36f7ada7bfa76571ea2", size = 411710 },
|
88 |
+
{ url = "https://files.pythonhosted.org/packages/33/fb/41cde15fbe51365024550bf77b95a4fc84ef41365705c946da0421f0e1e0/aiohttp-3.11.16-cp312-cp312-win_amd64.whl", hash = "sha256:23a15727fbfccab973343b6d1b7181bfb0b4aa7ae280f36fd2f90f5476805682", size = 438012 },
|
89 |
]
|
90 |
|
91 |
[[package]]
|
|
|
152 |
|
153 |
[[package]]
|
154 |
name = "authlib"
|
155 |
+
version = "1.5.2"
|
156 |
source = { registry = "https://pypi.org/simple" }
|
157 |
dependencies = [
|
158 |
{ name = "cryptography" },
|
159 |
]
|
160 |
+
sdist = { url = "https://files.pythonhosted.org/packages/2a/b3/5f5bc73c6558a21f951ffd267f41c6340d15f5fe0ff4b6bf37694f3558b8/authlib-1.5.2.tar.gz", hash = "sha256:fe85ec7e50c5f86f1e2603518bb3b4f632985eb4a355e52256530790e326c512", size = 153000 }
|
161 |
wheels = [
|
162 |
+
{ url = "https://files.pythonhosted.org/packages/e3/71/8dcec996ea8cc882cec9cace91ae1b630a226b88b0f04ab2ffa778f565ad/authlib-1.5.2-py2.py3-none-any.whl", hash = "sha256:8804dd4402ac5e4a0435ac49e0b6e19e395357cfa632a3f624dcb4f6df13b4b1", size = 232055 },
|
163 |
]
|
164 |
|
165 |
[[package]]
|
|
|
573 |
|
574 |
[[package]]
|
575 |
name = "gradio"
|
576 |
+
version = "5.23.3"
|
577 |
source = { registry = "https://pypi.org/simple" }
|
578 |
dependencies = [
|
579 |
{ name = "aiofiles" },
|
|
|
605 |
{ name = "urllib3", marker = "sys_platform == 'emscripten'" },
|
606 |
{ name = "uvicorn", marker = "sys_platform != 'emscripten'" },
|
607 |
]
|
608 |
+
sdist = { url = "https://files.pythonhosted.org/packages/4a/55/a9c52db0f88b217058abe23ed0bf80e4c4101edbed5c1a0e17cf8208e849/gradio-5.23.3.tar.gz", hash = "sha256:15e1844bd488f2b1f62ab8ac510231ee6d552190aa3fbc0eaaf1c384e7b08e12", size = 55906880 }
|
609 |
wheels = [
|
610 |
+
{ url = "https://files.pythonhosted.org/packages/5f/e3/f1e6569558d212f747820b2c976b4f57c034413a5747acb373f2a04bcd9a/gradio-5.23.3-py3-none-any.whl", hash = "sha256:4176c4f8e49e5338baea0c844e884c28711d4d673c9ea2b4e74f51f68a2d6f2d", size = 46485678 },
|
611 |
]
|
612 |
|
613 |
[package.optional-dependencies]
|
|
|
703 |
|
704 |
[[package]]
|
705 |
name = "huggingface-hub"
|
706 |
+
version = "0.30.1"
|
707 |
source = { registry = "https://pypi.org/simple" }
|
708 |
dependencies = [
|
709 |
{ name = "filelock" },
|
|
|
714 |
{ name = "tqdm" },
|
715 |
{ name = "typing-extensions" },
|
716 |
]
|
717 |
+
sdist = { url = "https://files.pythonhosted.org/packages/78/be/049689a7197630e75c4bb53021cb209a56617c9bf39b3a0950650d1f96e1/huggingface_hub-0.30.1.tar.gz", hash = "sha256:f379e8b8d0791295602538856638460ae3cf679c7f304201eb80fb98c771950e", size = 400784 }
|
718 |
wheels = [
|
719 |
+
{ url = "https://files.pythonhosted.org/packages/99/e3/2232d0e726d4d6ea69643b9593d97d0e7e6ea69c2fe9ed5de34d476c1c47/huggingface_hub-0.30.1-py3-none-any.whl", hash = "sha256:0f6aa5ec5a4e68e5b9e45d556b4e5ea180c58f5a5ffa734e7f38c9d573028959", size = 481170 },
|
720 |
]
|
721 |
|
722 |
[package.optional-dependencies]
|
|
|
1074 |
|
1075 |
[[package]]
|
1076 |
name = "multidict"
|
1077 |
+
version = "6.3.1"
|
1078 |
source = { registry = "https://pypi.org/simple" }
|
1079 |
+
sdist = { url = "https://files.pythonhosted.org/packages/a1/d4/4e0e25aabd94f289b7d47da8293a3563e73ac1a4f7e9caddf11b6eeaf52d/multidict-6.3.1.tar.gz", hash = "sha256:3e18d6afe3f855736022748606def2000af18e90253fb8b4d698b51f61e21283", size = 86832 }
|
1080 |
wheels = [
|
1081 |
+
{ url = "https://files.pythonhosted.org/packages/62/74/acf7ecc3eb1195b3930008cfaae9ffbc54c71a5582a0c86db291307dd7cf/multidict-6.3.1-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:9679106633695b132ebc9191ec6230bfb1415d37c483833fcef2b35a2e8665ec", size = 62598 },
|
1082 |
+
{ url = "https://files.pythonhosted.org/packages/d3/78/544442c1bb61986e5cbd529bfb7038b38f88fc39fe17962bc63bb0013199/multidict-6.3.1-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:73a43b3b2409aa395cce91b7471cff6b45814548063b18412162ba2222084201", size = 37323 },
|
1083 |
+
{ url = "https://files.pythonhosted.org/packages/f5/d1/8260a5da38ac65e885a0adbd40b50113e6f9e8eeb94b91aeeac08e4ad36e/multidict-6.3.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:1ce924e24c4f1c014f2ed8782e82a5232d5f61293fc5c204d8569f451191ffa8", size = 36053 },
|
1084 |
+
{ url = "https://files.pythonhosted.org/packages/e7/51/cceca6f30954620b9b03f96e06a54f908367cf86ae0c61a3472aa3de9363/multidict-6.3.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:123b1d48eeed2ac1126be078deb88006f871559787cefc8759a884442a6f2cdc", size = 244641 },
|
1085 |
+
{ url = "https://files.pythonhosted.org/packages/08/df/fb512545fa3cbf20f870175a4698c6ba58abf261ab19faf28d293e39621d/multidict-6.3.1-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:6d98447906885e7f0f90456cde1d14ff41f30d9d7e127ab7140a45e784a0ff1b", size = 255743 },
|
1086 |
+
{ url = "https://files.pythonhosted.org/packages/65/fb/a7648f5764e25e0a18e7d3bdda9fd67e86bb7e0c70a6ffee0348e1fb493f/multidict-6.3.1-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:5072a9efe7f7f79d3dff1f26ac41e4893478f85ce55fe5318625f7eb703d76f8", size = 252217 },
|
1087 |
+
{ url = "https://files.pythonhosted.org/packages/c0/98/32ef5e26956ee8cd7af8d7367b9da11f5a211438498e854bd5385e8eb812/multidict-6.3.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:bbc825b34595fe43966242e30b54d29617013e51b4310134aa2c16c3b3d00c91", size = 245195 },
|
1088 |
+
{ url = "https://files.pythonhosted.org/packages/4d/e1/a8cf44bef56bee949ec68d993ecbc4b713338b3137fa42416f0e34a46c48/multidict-6.3.1-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:baec41c191855f92507f9e0bb182eea7eea5992d649f9c712c96a38076e59d00", size = 232457 },
|
1089 |
+
{ url = "https://files.pythonhosted.org/packages/08/7c/81e91ef84b5df88d4780bcd03b08df423668b61ada7b387e0482ac19690a/multidict-6.3.1-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:eacd4036bb3d632828702a076458d660b53d12e049155eaeb7d11a91242d67b8", size = 252077 },
|
1090 |
+
{ url = "https://files.pythonhosted.org/packages/6c/00/32b94a1b060f602aa9056189febefddfaf6cece4c6f4c5873668011fd67b/multidict-6.3.1-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:284737db826cc00fbd5292225717492f037afa404a2ddfea812cfbef7a3f0e93", size = 247211 },
|
1091 |
+
{ url = "https://files.pythonhosted.org/packages/9c/78/9994ab4cc9b18c48e089f08c85028ea8c60ffa0d5868d7e42c842b9ca80e/multidict-6.3.1-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:ebd121433f5d8707379f4fc0e6b4bf67b0b7cd1a7132e097ead2713c8d661a41", size = 260515 },
|
1092 |
+
{ url = "https://files.pythonhosted.org/packages/86/aa/d40c0bce043fa2903e7d3f9e5a2402fd55850933bc81f86a08efe78e303d/multidict-6.3.1-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:31f94d64672487570c7c2bbcff74311055066e013545714b938786843eb54ef8", size = 255476 },
|
1093 |
+
{ url = "https://files.pythonhosted.org/packages/86/1b/20580f901b260c2d6733e5ec3e1e227e04330de966b567b3f3e102567bd0/multidict-6.3.1-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:739fe3fde8b8aca7219048f8bda17901afb8710c93307dc0d740014d3481b36b", size = 251311 },
|
1094 |
+
{ url = "https://files.pythonhosted.org/packages/9c/44/db7cb84b604ee1912d5ba908def729adc60413448e789247db98992c149f/multidict-6.3.1-cp312-cp312-win32.whl", hash = "sha256:891a94a056de2d904cc30f40ec1d111aebb09abd33089a34631ff5a19e0167b2", size = 35031 },
|
1095 |
+
{ url = "https://files.pythonhosted.org/packages/32/bb/2931b3d6a2b57b5a7dbb819c2b5c55d3170c54098009224872d4b6ae40d3/multidict-6.3.1-cp312-cp312-win_amd64.whl", hash = "sha256:d9844e0f93405a9c5bc2106d48cf82e022e18685baebea74cc5057ca2009799e", size = 38453 },
|
1096 |
+
{ url = "https://files.pythonhosted.org/packages/57/99/624da94f4deb41a75b5e08752270ecdb1ff871c1b1539705c0aef02aa7a2/multidict-6.3.1-py3-none-any.whl", hash = "sha256:2d45b070b33fa1d0a9a7650469997713e3a4f5cd9eb564332d5d0206cf61efc5", size = 10348 },
|
1097 |
]
|
1098 |
|
1099 |
[[package]]
|
|
|
2348 |
[[package]]
|
2349 |
name = "yourbench"
|
2350 |
version = "0.2.0"
|
2351 |
+
source = { git = "https://github.com/huggingface/yourbench.git?branch=main#334778a3bcff0ddcec7fb15dd145a9b23e772ed6" }
|
2352 |
dependencies = [
|
2353 |
{ name = "asyncio" },
|
2354 |
{ name = "bert-score" },
|
|
|
2373 |
]
|
2374 |
|
2375 |
[[package]]
|
2376 |
+
name = "yourbench-advanced"
|
2377 |
version = "0.1.0"
|
2378 |
source = { editable = "." }
|
2379 |
dependencies = [
|
yourbench_space/app.py
CHANGED
@@ -28,8 +28,8 @@ project_description = """
|
|
28 |
# 🚀 YourBench
|
29 |
### Dynamic Benchmark Generation from Your Documents
|
30 |
|
31 |
-
- Create zero-shot benchmarks from your documents — no manual labeling
|
32 |
-
- Evaluate top open models and publish a leaderboard in one click
|
33 |
- Run locally or explore the [source on GitHub](https://github.com/huggingface/yourbench)
|
34 |
|
35 |
⚠️ **Important:** This app uses your Hugging Face token for inference and uploads — you are responsible for any usage costs
|
|
|
28 |
# 🚀 YourBench
|
29 |
### Dynamic Benchmark Generation from Your Documents
|
30 |
|
31 |
+
- Create zero-shot benchmarks from your documents — no manual labeling
|
32 |
+
- Evaluate top open models and publish a leaderboard in one click
|
33 |
- Run locally or explore the [source on GitHub](https://github.com/huggingface/yourbench)
|
34 |
|
35 |
⚠️ **Important:** This app uses your Hugging Face token for inference and uploads — you are responsible for any usage costs
|
yourbench_space/evaluation.py
CHANGED
@@ -18,7 +18,7 @@ def create_eval_file(eval_ds_name: str):
|
|
18 |
subprocess.run(["lighteval", "tasks", "create", str(template_path), task_name, eval_ds_name])
|
19 |
|
20 |
|
21 |
-
async def run_process(args: list, custom_env
|
22 |
process = await asyncio.create_subprocess_exec(
|
23 |
*args, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE, env=custom_env
|
24 |
)
|
@@ -32,7 +32,7 @@ async def run_process(args: list, custom_env = None) -> dict:
|
|
32 |
return {"pid": process.pid, "stdout": stdout.decode(), "stderr": stderr.decode()}
|
33 |
|
34 |
|
35 |
-
async def run_evaluations(eval_ds_name: str, org: str, custom_env
|
36 |
task_name = eval_ds_name.replace("/", "_")
|
37 |
tasks = []
|
38 |
for model_name, provider in INIT_MODELS:
|
|
|
18 |
subprocess.run(["lighteval", "tasks", "create", str(template_path), task_name, eval_ds_name])
|
19 |
|
20 |
|
21 |
+
async def run_process(args: list, custom_env=None) -> dict:
|
22 |
process = await asyncio.create_subprocess_exec(
|
23 |
*args, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE, env=custom_env
|
24 |
)
|
|
|
32 |
return {"pid": process.pid, "stdout": stdout.decode(), "stderr": stderr.decode()}
|
33 |
|
34 |
|
35 |
+
async def run_evaluations(eval_ds_name: str, org: str, custom_env=None) -> list:
|
36 |
task_name = eval_ds_name.replace("/", "_")
|
37 |
tasks = []
|
38 |
for model_name, provider in INIT_MODELS:
|
yourbench_space/leaderboard_space/utils.py
CHANGED
@@ -6,11 +6,13 @@ from env import TASK, MODELS, ORG_NAME
|
|
6 |
import gradio as gr
|
7 |
from datasets import Dataset, load_dataset
|
8 |
|
|
|
9 |
KNOWN_METRIC_LABELS = {
|
10 |
"accuracy": "Accuracy",
|
11 |
"accuracy_stderr": "Accuracy (stderr)",
|
12 |
}
|
13 |
|
|
|
14 |
def aggregate_results() -> list:
|
15 |
"""Extract scores for each model and return list of result dictionaries."""
|
16 |
all_results = []
|
@@ -134,7 +136,7 @@ def samples_to_box_display(samples: list[dict[str, Any]], example_index: int = 0
|
|
134 |
--text-color: #333333;
|
135 |
--border-color: #ddd;
|
136 |
}
|
137 |
-
|
138 |
@media (prefers-color-scheme: dark) {
|
139 |
:root {
|
140 |
--primary-bg: #2a2a2a;
|
@@ -144,43 +146,43 @@ def samples_to_box_display(samples: list[dict[str, Any]], example_index: int = 0
|
|
144 |
--border-color: #555;
|
145 |
}
|
146 |
}
|
147 |
-
|
148 |
.box-container {
|
149 |
max-width: 800px;
|
150 |
margin: 0 auto;
|
151 |
color: var(--text-color);
|
152 |
}
|
153 |
-
|
154 |
.gold-box {
|
155 |
background: var(--gold-bg);
|
156 |
padding: 20px;
|
157 |
border-radius: 10px;
|
158 |
margin-bottom: 20px;
|
159 |
}
|
160 |
-
|
161 |
.model-box {
|
162 |
background: var(--primary-bg);
|
163 |
padding: 20px;
|
164 |
margin-bottom: 20px;
|
165 |
border-radius: 10px;
|
166 |
}
|
167 |
-
|
168 |
.content-section {
|
169 |
background: var(--secondary-bg);
|
170 |
padding: 15px;
|
171 |
border-radius: 5px;
|
172 |
margin-top: 10px;
|
173 |
}
|
174 |
-
|
175 |
.metric-row {
|
176 |
padding: 5px;
|
177 |
border-bottom: 1px solid var(--border-color);
|
178 |
}
|
179 |
-
|
180 |
h2, h3 {
|
181 |
color: var(--text-color);
|
182 |
}
|
183 |
-
|
184 |
pre, code {
|
185 |
white-space: pre-wrap;
|
186 |
word-wrap: break-word;
|
@@ -307,4 +309,4 @@ def update_examples(samples_ix: int = 0) -> tuple[str, str, str]:
|
|
307 |
samples_to_box_display(easy_samples, samples_ix),
|
308 |
samples_to_box_display(hard_samples, samples_ix),
|
309 |
samples_to_box_display(all_samples, samples_ix),
|
310 |
-
)
|
|
|
6 |
import gradio as gr
|
7 |
from datasets import Dataset, load_dataset
|
8 |
|
9 |
+
|
10 |
KNOWN_METRIC_LABELS = {
|
11 |
"accuracy": "Accuracy",
|
12 |
"accuracy_stderr": "Accuracy (stderr)",
|
13 |
}
|
14 |
|
15 |
+
|
16 |
def aggregate_results() -> list:
|
17 |
"""Extract scores for each model and return list of result dictionaries."""
|
18 |
all_results = []
|
|
|
136 |
--text-color: #333333;
|
137 |
--border-color: #ddd;
|
138 |
}
|
139 |
+
|
140 |
@media (prefers-color-scheme: dark) {
|
141 |
:root {
|
142 |
--primary-bg: #2a2a2a;
|
|
|
146 |
--border-color: #555;
|
147 |
}
|
148 |
}
|
149 |
+
|
150 |
.box-container {
|
151 |
max-width: 800px;
|
152 |
margin: 0 auto;
|
153 |
color: var(--text-color);
|
154 |
}
|
155 |
+
|
156 |
.gold-box {
|
157 |
background: var(--gold-bg);
|
158 |
padding: 20px;
|
159 |
border-radius: 10px;
|
160 |
margin-bottom: 20px;
|
161 |
}
|
162 |
+
|
163 |
.model-box {
|
164 |
background: var(--primary-bg);
|
165 |
padding: 20px;
|
166 |
margin-bottom: 20px;
|
167 |
border-radius: 10px;
|
168 |
}
|
169 |
+
|
170 |
.content-section {
|
171 |
background: var(--secondary-bg);
|
172 |
padding: 15px;
|
173 |
border-radius: 5px;
|
174 |
margin-top: 10px;
|
175 |
}
|
176 |
+
|
177 |
.metric-row {
|
178 |
padding: 5px;
|
179 |
border-bottom: 1px solid var(--border-color);
|
180 |
}
|
181 |
+
|
182 |
h2, h3 {
|
183 |
color: var(--text-color);
|
184 |
}
|
185 |
+
|
186 |
pre, code {
|
187 |
white-space: pre-wrap;
|
188 |
word-wrap: break-word;
|
|
|
309 |
samples_to_box_display(easy_samples, samples_ix),
|
310 |
samples_to_box_display(hard_samples, samples_ix),
|
311 |
samples_to_box_display(all_samples, samples_ix),
|
312 |
+
)
|