sentencetransformers

index.yaml

@@ -48,80 +48,80 @@
       uri: https://huggingface.co/TheBloke/leo-hessianai-13B-chat-bilingual-GGUF/resolve/main/leo-hessianai-13b-chat-bilingual.Q4_K_M.gguf
   
 
-- url: https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/base.yaml
-  name: text-embedding-gte-german
-  license: mit
-  urls:
-    - https://huggingface.co/bert-base-german-cased
-  tags:
-    - license:mit
-    - region:us
-  overrides:
-    parameters:
-        model: text-embeddings/gte/gte-base-german.bin
-    embeddings: true
-    gpu_layers: 12      
-  files:
-    - filename: text-embeddings/gte/gte-base-german.bin
-      sha256: "56a21938415b06a68b870e4b1b3413cdd532ae6456fefef1ee5a852faf52f806"
-      uri: https://huggingface.co/bert-base-german-cased/resolve/main/pytorch_model.bin
+# - url: https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/base.yaml
+#   name: text-embedding-gte-german
+#   license: mit
+#   urls:
+#     - https://huggingface.co/bert-base-german-cased
+#   tags:
+#     - license:mit
+#     - region:us
+#   overrides:
+#     parameters:
+#         model: text-embeddings/gte/gte-base-german.bin
+#     embeddings: true
+#     gpu_layers: 12      
+#   files:
+#     - filename: text-embeddings/gte/gte-base-german.bin
+#       sha256: "56a21938415b06a68b870e4b1b3413cdd532ae6456fefef1ee5a852faf52f806"
+#       uri: https://huggingface.co/bert-base-german-cased/resolve/main/pytorch_model.bin
 
-- url: https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/base.yaml
-  name: text-embedding-gte-base
-  license: mit
-  urls:
-    - https://huggingface.co/maikaarda/gte-base-ggml
-  tags:
-    - license:mit
-    - region:us
-  overrides:
-    parameters:
-        model: text-embeddings/gte/gte-base-ggml-f16.bin
-    embeddings: true
-    gpu_layers: 12      
-  files:
-    - filename: text-embeddings/gte/gte-base-ggml-f16.bin
-      sha256: ""
-      uri: https://huggingface.co/maikaarda/gte-base-ggml/resolve/main/ggml-model-f16.bin
+# - url: https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/base.yaml
+#   name: text-embedding-gte-base
+#   license: mit
+#   urls:
+#     - https://huggingface.co/maikaarda/gte-base-ggml
+#   tags:
+#     - license:mit
+#     - region:us
+#   overrides:
+#     parameters:
+#         model: text-embeddings/gte/gte-base-ggml-f16.bin
+#     embeddings: true
+#     gpu_layers: 12      
+#   files:
+#     - filename: text-embeddings/gte/gte-base-ggml-f16.bin
+#       sha256: ""
+#       uri: https://huggingface.co/maikaarda/gte-base-ggml/resolve/main/ggml-model-f16.bin
 
-- url: https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/base.yaml
-  name: text-embedding-gte-large
-  license: mit
-  urls:
-    - https://huggingface.co/maikaarda/gte-large-ggml
-  tags:
-    - license:mit
-    - region:us
-  overrides:
-    parameters:
-        model: text-embeddings/gte/gte-large-ggml-model-q4_0.bin
-    embeddings: true
-    gpu_layers: 12      
-  files:
-    - filename: text-embeddings/gte/gte-large-ggml-model-q4_0.bin
-      sha256: ""
-      uri: https://huggingface.co/maikaarda/gte-large-ggml/resolve/main/gte-large-ggml-model-q4_0.bin
+# - url: https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/base.yaml
+#   name: text-embedding-gte-large
+#   license: mit
+#   urls:
+#     - https://huggingface.co/maikaarda/gte-large-ggml
+#   tags:
+#     - license:mit
+#     - region:us
+#   overrides:
+#     parameters:
+#         model: text-embeddings/gte/gte-large-ggml-model-q4_0.bin
+#     embeddings: true
+#     gpu_layers: 12      
+#   files:
+#     - filename: text-embeddings/gte/gte-large-ggml-model-q4_0.bin
+#       sha256: ""
+#       uri: https://huggingface.co/maikaarda/gte-large-ggml/resolve/main/gte-large-ggml-model-q4_0.bin
 
-- url: https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/base.yaml
-  name: text-embedding-all-miniLM-L6
-  license: mit
-  urls:
-    - https://huggingface.co/skeskinen/ggml
-  tags:
-    - license:mit
-    - region:us
-  overrides:
-    parameters:
-        model: all-miniLM-L6.bin
-    embeddings: true
-    backend: sentencetransformers
-    # gpu_layers: 2      
-  files:
-    - filename: all-miniLM-L6.bin
-      sha256: ""
-      uri: https://huggingface.co/skeskinen/ggml/resolve/main/all-MiniLM-L6-v2/ggml-model-q4_0.bin
+# - url: https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/base.yaml
+#   name: text-embedding-all-miniLM-L6
+#   license: mit
+#   urls:
+#     - https://huggingface.co/skeskinen/ggml
+#   tags:
+#     - license:mit
+#     - region:us
+#   overrides:
+#     parameters:
+#         model: all-miniLM-L6.bin
+#     embeddings: true
+#     backend: sentencetransformers
+#     # gpu_layers: 2      
+#   files:
+#     - filename: all-miniLM-L6.bin
+#       sha256: ""
+#       uri: https://huggingface.co/skeskinen/ggml/resolve/main/all-MiniLM-L6-v2/ggml-model-q4_0.bin
 
-- url: "https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/bert-embeddings.yaml"
+- url: "https://gitlab.kit.edu/kit/aifb/BIS/infrastruktur/localai/localai-model-gallery/-/raw/main/text-embeddings.yaml"
   name: "text-embedding-ada-002"
   license: other
 

bert-embeddings.yaml → text-embeddings.yaml

@@ -3,13 +3,13 @@ license: "Apache 2.0"
 urls:
 - https://huggingface.co/skeskinen/ggml
 description: |
-    Bert model that can be used for embeddings
+    GTE model that can be used for embeddings
 config_file: |
     parameters:
-      model: text-embeddings/bert/bert-MiniLM-L6-v2q4_0.bin
-    backend: bert-embeddings
+      model: ""thenlper/gte-large""
+    backend: sentencetransformers
     embeddings: true
-files:
-- filename: "text-embeddings/bert/bert-MiniLM-L6-v2q4_0.bin"
-  sha256: "a5a174d8772c8a569faf9f3136c441f2c3855b5bf35ed32274294219533feaad"
-  uri: "https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin"
+# files:
+# - filename: "text-embeddings/bert/bert-MiniLM-L6-v2q4_0.bin"
+#   sha256: "a5a174d8772c8a569faf9f3136c441f2c3855b5bf35ed32274294219533feaad"
+#   uri: "https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin"
\ No newline at end of file