diff --git a/llm-servers/sbert/gpu/Pipfile b/llm-servers/sbert/gpu/Pipfile index 1d809a2..a755386 100644 --- a/llm-servers/sbert/gpu/Pipfile +++ b/llm-servers/sbert/gpu/Pipfile @@ -13,7 +13,7 @@ einops = "~=0.8.0" fastapi = "~=0.115.7" pip = "~=25.0" sentence-transformers = "~=3.4.1" -torch = {version = "==2.5.1+cu124", index = "pytorch"} +torch = {version = "==2.6.0+cu124", index = "pytorch"} transformers = "~=4.48.1" uvicorn = "~=0.34.0" wheel = "~=0.45.1" diff --git a/llm-servers/sbert/gpu/Pipfile.lock b/llm-servers/sbert/gpu/Pipfile.lock index 956cc2a..0b5ddcd 100644 --- a/llm-servers/sbert/gpu/Pipfile.lock +++ b/llm-servers/sbert/gpu/Pipfile.lock @@ -1,7 +1,7 @@ { "_meta": { "hash": { - "sha256": "5b8f7b360234ab242ca44fab68c474183d97602992efd0a98887c4d07efa9fc4" + "sha256": "187f09257050774ae41445ea702ca278423cb559f4ef66c7c23704ee4567aee9" }, "pipfile-spec": 6, "requires": { @@ -171,19 +171,19 @@ }, "filelock": { "hashes": [ - "sha256:533dc2f7ba78dc2f0f531fc6c4940addf7b70a481e269a5a3b93be94ffbe8338", - "sha256:ee4e77401ef576ebb38cd7f13b9b28893194acc20a8e68e18730ba9c0e54660e" + "sha256:adbc88eabb99d2fec8c9c1b229b171f18afa655400173ddc653d5d01501fb9f2", + "sha256:c401f4f8377c4464e6db25fff06205fd89bdd83b65eb0488ed1b160f780e21de" ], "markers": "python_version >= '3.9'", - "version": "==3.17.0" + "version": "==3.18.0" }, "fsspec": { "hashes": [ - "sha256:670700c977ed2fb51e0d9f9253177ed20cbde4a3e5c0283cc5385b5870c8533f", - "sha256:b520aed47ad9804237ff878b504267a3b0b441e97508bd6d2d8774e3db85cee2" + "sha256:2daf8dc3d1dfa65b6aa37748d112773a7a08416f6c70d96b264c96476ecaf711", + "sha256:e52c77ef398680bbd6a98c0e628fbc469491282981209907bbc8aea76a04fdc6" ], - "markers": "python_version >= '3.8'", - "version": "==2024.12.0" + "markers": "python_version >= '3.9'", + "version": "==2025.3.2" }, "h11": { "hashes": [ @@ -211,11 +211,11 @@ }, "jinja2": { "hashes": [ - "sha256:8fefff8dc3034e27bb80d67c671eb8a9bc424c0ef4c0826edbff304cceff43bb", - "sha256:aba0f4dc9ed8013c424088f68a5c226f7d6097ed89b246d7749c2ec4175c6adb" + "sha256:0137fb05990d35f1275a587e9aee6d56da821fc83491a0fb838183be43f66d6d", + "sha256:85ece4451f492d0c13c5dd7c13a64681a86afae63a5f347908daf103ce6d2f67" ], "markers": "python_version >= '3.7'", - "version": "==3.1.5" + "version": "==3.1.6" }, "joblib": { "hashes": [ @@ -448,6 +448,15 @@ "markers": "python_version >= '3'", "version": "==12.3.1.170" }, + "nvidia-cusparselt-cu12": { + "hashes": [ + "sha256:0057c91d230703924c0422feabe4ce768841f9b4b44d28586b6f6d2eb86fbe70", + "sha256:067a7f6d03ea0d4841c85f0c6f1991c5dda98211f6302cb83a4ab234ee95bef8", + "sha256:df2c24502fd76ebafe7457dbc4716b2fec071aabaed4fb7691a201cde03704d9" + ], + "markers": "platform_system == 'Linux' and platform_machine == 'x86_64'", + "version": "==0.6.2" + }, "nvidia-nccl-cu12": { "hashes": [ "sha256:8579076d30a8c24988834445f8d633c697d42397e92ffc3f63fa26766d25e0a0" @@ -962,11 +971,11 @@ }, "setuptools": { "hashes": [ - "sha256:c5afc8f407c626b8313a86e10311dd3f661c6cd9c09d4bf8c15c0e11f9f2b0e6", - "sha256:e3982f444617239225d675215d51f6ba05f845d4eec313da4418fdbb56fb27e3" + "sha256:18fd474d4a82a5f83dac888df697af65afa82dec7323d09c3e37d1f14288da54", + "sha256:3e386e96793c8702ae83d17b853fb93d3e09ef82ec62722e61da5cd22376dcd8" ], "markers": "python_version >= '3.9'", - "version": "==75.8.0" + "version": "==78.1.0" }, "sniffio": { "hashes": [ @@ -1023,18 +1032,30 @@ }, "torch": { "hashes": [ - "sha256:3c3f705fb125edbd77f9579fa11a138c56af8968a10fc95834cdd9fdf4f1f1a6", - "sha256:6b2966ede9affe2fd69e0765691ca723ec870e0c34c7761f4d5b8e318383fdaf", - "sha256:6c8a7003ef1327479ede284b6e5ab3527d3900c2b2d401af15bcc50f2245a59f", - "sha256:6f99d8459369cfd6661c2aee14787592fe50156a33faf9ef643ba04e42d6543f", - "sha256:9036c4372dec409842a80965d94b7b0fb4298e0967ceb03336a42c83778faa6f", - "sha256:9dde30f399ca22137455cca4d47140dfb7f4176e2d16a9729fc044eebfadb13a", - "sha256:bf6484bfe5bc4f92a4a1a1bf553041505e19a911f717065330eb061afe0e14d7", - "sha256:d681b8be3fdc2cd41112310db3c3904f7c6a09a7ae28d042ae0af3af01c8fcda", - "sha256:e9bebf91ede89267577911da4b0709ac6113a0cff6a1c2202c046b1ec2a51601" + "sha256:09e06f9949e1a0518c5b09fe95295bc9661f219d9ecb6f9893e5123e10696628", + "sha256:265f70de5fd45b864d924b64be1797f86e76c8e48a02c2a3a6fc7ec247d2226c", + "sha256:2bb8987f3bb1ef2675897034402373ddfc8f5ef0e156e2d8cfc47cacafdda4a9", + "sha256:46763dcb051180ce1ed23d1891d9b1598e07d051ce4c9d14307029809c4d64f7", + "sha256:4874a73507a300a5d089ceaff616a569e7bb7c613c56f37f63ec3ffac65259cf", + "sha256:510c73251bee9ba02ae1cb6c9d4ee0907b3ce6020e62784e2d7598e0cfa4d6cc", + "sha256:56eeaf2ecac90da5d9e35f7f35eb286da82673ec3c582e310a8d1631a1c02341", + "sha256:683410f97984103148e31b38a8631acf31c3034c020c0f4d26171e7626d8317a", + "sha256:6860df13d9911ac158f4c44031609700e1eba07916fff62e21e6ffa0a9e01961", + "sha256:7979834102cd5b7a43cc64e87f2f3b14bd0e1458f06e9f88ffa386d07c7446e1", + "sha256:7e1448426d0ba3620408218b50aa6ada88aeae34f7a239ba5431f6c8774b1239", + "sha256:94fc63b3b4bedd327af588696559f68c264440e2503cc9e6954019473d74ae21", + "sha256:9a610afe216a85a8b9bc9f8365ed561535c93e804c2a317ef7fabcc5deda0989", + "sha256:9ea955317cfcd3852b1402b62af258ce735c2edeee42ca9419b6bc889e5ae053", + "sha256:a0d5e1b9874c1a6c25556840ab8920569a7a4137afa8a63a32cee0bc7d89bd4b", + "sha256:b789069020c5588c70d5c2158ac0aa23fd24a028f34a8b4fcb8fcb4d7efcf5fb", + "sha256:bb2c6c3e65049f081940f5ab15c9136c7de40d3f01192541c920a07c7c585b7e", + "sha256:c4f103a49830ce4c7561ef4434cc7926e5a5fe4e5eb100c19ab36ea1e2b634ab", + "sha256:ccbd0320411fe1a3b3fec7b4d3185aa7d0c52adac94480ab024b5c8f74a0bf1d", + "sha256:ff96f4038f8af9f7ec4231710ed4549da1bdebad95923953a25045dcf6fd87e2" ], - "index": "pytorch", - "version": "==2.5.1+cu124" + "index": "pypi", + "markers": "python_full_version >= '3.9.0'", + "version": "==2.6.0" }, "tqdm": { "hashes": [ @@ -1055,22 +1076,22 @@ }, "triton": { "hashes": [ - "sha256:0f34f6e7885d1bf0eaaf7ba875a5f0ce6f3c13ba98f9503651c1e6dc6757ed5c", - "sha256:6b0dd10a925263abbe9fa37dcde67a5e9b2383fc269fdf59f5657cac38c5d1d8", - "sha256:6dadaca7fc24de34e180271b5cf864c16755702e9f63a16f62df714a8099126a", - "sha256:aafa9a20cd0d9fee523cd4504aa7131807a864cd77dcf6efe7e981f18b8c6c11", - "sha256:c8182f42fd8080a7d39d666814fa36c5e30cc00ea7eeeb1a2983dbb4c99a0fdc" + "sha256:30ceed0eff2c4a73b14eb63e052992f44bbdf175f3fad21e1ac8097a772de7ee", + "sha256:8009a1fb093ee8546495e96731336a33fb8856a38e45bb4ab6affd6dbc3ba220", + "sha256:8d9b215efc1c26fa7eefb9a157915c92d52e000d2bf83e5f69704047e63f125c", + "sha256:b3e54983cd51875855da7c68ec05c05cf8bb08df361b1d5b69e05e40b0c9bd62", + "sha256:e5dfa23ba84541d7c0a531dfce76d8bcd19159d50a4a8b14ad01e91734a5c1b0" ], - "markers": "platform_system == 'Linux' and platform_machine == 'x86_64' and python_version < '3.13'", - "version": "==3.1.0" + "markers": "platform_system == 'Linux' and platform_machine == 'x86_64'", + "version": "==3.2.0" }, "typing-extensions": { "hashes": [ - "sha256:04e5ca0351e0f3f85c6853954072df659d0d13fac324d0072316b67d7794700d", - "sha256:1a7ead55c7e559dd4dee8856e3a88b41225abfe1ce8df57b7c13915fe121ffb8" + "sha256:a439e7c04b49fec3e5d3e2beaa21755cadbbdc391694e28ccdd36ca4a1408f8c", + "sha256:e6c81219bd689f51865d9e372991c540bda33a0379d5573cddb9a3a23f7caaef" ], "markers": "python_version >= '3.8'", - "version": "==4.12.2" + "version": "==4.13.2" }, "urllib3": { "hashes": [ diff --git a/llm-servers/sbert/gpu/requirements.txt b/llm-servers/sbert/gpu/requirements.txt index 822de16..4241d27 100644 --- a/llm-servers/sbert/gpu/requirements.txt +++ b/llm-servers/sbert/gpu/requirements.txt @@ -7,12 +7,12 @@ charset-normalizer==3.4.1; python_version >= '3.7' click==8.1.8; python_version >= '3.7' einops==0.8.0; python_version >= '3.8' fastapi==0.115.7; python_version >= '3.8' -filelock==3.17.0; python_version >= '3.9' -fsspec==2024.12.0; python_version >= '3.8' +filelock==3.18.0; python_version >= '3.9' +fsspec==2025.3.2; python_version >= '3.9' h11==0.14.0; python_version >= '3.7' huggingface-hub==0.28.0; python_full_version >= '3.8.0' idna==3.10; python_version >= '3.6' -jinja2==3.1.5; python_version >= '3.7' +jinja2==3.1.6; python_version >= '3.7' joblib==1.4.2; python_version >= '3.8' markupsafe==3.0.2; python_version >= '3.9' mpmath==1.3.0 @@ -27,6 +27,7 @@ nvidia-cufft-cu12==11.2.1.3; python_version >= '3' nvidia-curand-cu12==10.3.5.147; python_version >= '3' nvidia-cusolver-cu12==11.6.1.9; python_version >= '3' nvidia-cusparse-cu12==12.3.1.170; python_version >= '3' +nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' nvidia-nccl-cu12==2.21.5; python_version >= '3' nvidia-nvjitlink-cu12==12.4.127; python_version >= '3' nvidia-nvtx-cu12==12.4.127; python_version >= '3' @@ -42,17 +43,17 @@ safetensors==0.5.2; python_version >= '3.7' scikit-learn==1.6.1; python_version >= '3.9' scipy==1.15.1; python_version >= '3.10' sentence-transformers==3.4.1; python_version >= '3.9' -setuptools==75.8.0; python_version >= '3.9' +setuptools==78.1.0; python_version >= '3.9' sniffio==1.3.1; python_version >= '3.7' starlette==0.45.3; python_version >= '3.9' sympy==1.13.1; python_version >= '3.8' threadpoolctl==3.5.0; python_version >= '3.8' tokenizers==0.21.0; python_version >= '3.7' -torch==2.5.1+cu124 +torch==2.6.0; python_full_version >= '3.9.0' tqdm==4.67.1; python_version >= '3.7' transformers==4.48.1; python_full_version >= '3.9.0' -triton==3.1.0; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version < '3.13' -typing-extensions==4.12.2; python_version >= '3.8' +triton==3.2.0; platform_system == 'Linux' and platform_machine == 'x86_64' +typing-extensions==4.13.2; python_version >= '3.8' urllib3==2.3.0; python_version >= '3.9' uvicorn==0.34.0; python_version >= '3.9' wheel==0.45.1; python_version >= '3.8'