Deprecate GLM and some Palmyra models (#2814)

setup.cfg

@@ -155,9 +155,6 @@ google =
 together =
     together~=1.1
 
-tsinghua =
-    icetk~=0.0.4
-
 yandex =
     sentencepiece~=0.1.97
 
@@ -174,7 +171,6 @@ models =
     crfm-helm[openai]
     crfm-helm[reka]
     crfm-helm[together]
-    crfm-helm[tsinghua]
     crfm-helm[yandex]
     crfm-helm[openvino]
 
@@ -232,6 +228,7 @@ heim =
 
     # HEIM models
     diffusers~=0.24.0
+    icetk~=0.0.4
     jax~=0.4.13
     jaxlib~=0.4.13
     crfm-helm[openai]

src/helm/config/model_deployments.yaml

@@ -2062,18 +2062,6 @@ model_deployments:
       args:
         together_model: togethercomputer/RedPajama-INCITE-7B-Instruct
 
-  ## Tsinghua
-  - name: together/glm
-    deprecated: true  # Removed from Together
-    model_name: tsinghua/glm
-    tokenizer_name: TsinghuaKEG/ice
-    max_sequence_length: 2048
-    max_request_length: 2049
-    client_spec:
-      class_name: "helm.clients.together_client.TogetherClient"
-    window_service_spec:
-      class_name: "helm.benchmark.window_services.ice_window_service.ICEWindowService"
-
   - name: thudm/cogview2
     model_name: thudm/cogview2
     tokenizer_name: openai/clip-vit-large-patch14
@@ -2112,22 +2100,6 @@ model_deployments:
     client_spec:
       class_name: "helm.clients.palmyra_client.PalmyraClient"
 
-  - name: writer/palmyra-instruct-30
-    model_name: writer/palmyra-instruct-30
-    tokenizer_name: writer/gpt2
-    max_sequence_length: 2048
-    max_sequence_and_generated_tokens_length: 2048
-    client_spec:
-      class_name: "helm.clients.palmyra_client.PalmyraClient"
-
-  - name: writer/palmyra-e
-    model_name: writer/palmyra-e
-    tokenizer_name: writer/gpt2
-    max_sequence_length: 2048
-    max_sequence_and_generated_tokens_length: 2048
-    client_spec:
-      class_name: "helm.clients.palmyra_client.PalmyraClient"
-
   - name: writer/silk-road
     model_name: writer/silk-road
     tokenizer_name: writer/gpt2

src/helm/config/model_metadata.yaml

@@ -2707,7 +2707,7 @@ models:
     release_date: 2022-08-04
     # Inference with echo=True is not feasible -- in the prompt encoding phase, they use
     # bidirectional attention and do not perform predictions on them.
-    tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, ABLATION_MODEL_TAG, NO_NEWLINES_TAG]
+    tags: [DEPRECATED_MODEL_TAG, TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, ABLATION_MODEL_TAG, NO_NEWLINES_TAG]
 
   - name: tsinghua/codegeex # NOT SUPPORTED
     display_name: CodeGeeX (13B)
@@ -2742,26 +2742,24 @@ models:
     tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
 
   - name: writer/palmyra-instruct-30
-    deprecated: true # Internal error
     display_name: InstructPalmyra (30B)
     description: InstructPalmyra (30B parameters) is trained using reinforcement learning techniques based on feedback from humans.
     creator_organization_name: Writer
     access: limited
     num_parameters: 30000000000
     release_date: 2023-02-16
     # Does not support echo
-    tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
+    tags: [DEPRECATED_MODEL_TAG, TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
 
   - name: writer/palmyra-e
-    deprecated: true # Internal error
     display_name: Palmyra E (30B)
     description: Palmyra E (30B)
     creator_organization_name: Writer
     access: limited
     num_parameters: 30000000000
     release_date: 2023-03-03
     # Does not support echo
-    tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
+    tags: [DEPRECATED_MODEL_TAG, TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
 
   - name: writer/silk-road
     display_name: Silk Road (35B)