Add new Palmyra models

src/helm/benchmark/window_services/palmyra_window_service.py

@@ -43,3 +43,9 @@ class LongerPalmyraWindowService(PalmyraWindowService):
     @property
     def max_sequence_length(self) -> int:
         return 8192
+
+
+class Palmyra32KWindowService(PalmyraWindowService):
+    @property
+    def max_sequence_length(self) -> int:
+        return 32000

src/helm/config/model_deployments.yaml

@@ -1641,4 +1641,40 @@ model_deployments:
       args: {}
     window_service_spec:
       class_name: "helm.benchmark.window_services.palmyra_window_service.LongerPalmyraWindowService"
+      args: {}
+
+  - name: writer/palmyra-x-v2
+    model_name: writer/palmyra-x-v2
+    tokenizer_name: writer/gpt2
+    max_sequence_length: 6000
+    max_sequence_and_generated_tokens_length: 7024
+    client_spec:
+      class_name: "helm.proxy.clients.palmyra_client.PalmyraClient"
+      args: {}
+    window_service_spec:
+      class_name: "helm.benchmark.window_services.palmyra_window_service.LongerPalmyraWindowService"
+      args: {}
+
+  - name: writer/palmyra-x-v3
+    model_name: writer/palmyra-x-v3
+    tokenizer_name: writer/gpt2
+    max_sequence_length: 6000
+    max_sequence_and_generated_tokens_length: 7024
+    client_spec:
+      class_name: "helm.proxy.clients.palmyra_client.PalmyraClient"
+      args: {}
+    window_service_spec:
+      class_name: "helm.benchmark.window_services.palmyra_window_service.LongerPalmyraWindowService"
+      args: {}
+
+  - name: writer/palmyra-x-32k
+    model_name: writer/palmyra-x-32k
+    tokenizer_name: writer/gpt2
+    max_sequence_length: 28000
+    max_sequence_and_generated_tokens_length: 30048
+    client_spec:
+      class_name: "helm.proxy.clients.palmyra_client.PalmyraClient"
+      args: {}
+    window_service_spec:
+      class_name: "helm.benchmark.window_services.palmyra_window_service.Palmyra32KWindowService"
       args: {}

src/helm/config/model_metadata.yaml

@@ -1386,6 +1386,36 @@ models:
     # Does not support echo
     tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
 
+  - name: writer/palmyra-x-v2
+    display_name: Palmyra X V2 (33B)
+    description: Palmyra-X V2 (33B parameters) is a Transformer-based model, which is trained on extremely large-scale pre-training data. The pre-training data more than 2 trillion tokens types are diverse and cover a wide range of areas, used FlashAttention-2.
+    creator_organization_name: Writer
+    access: limited
+    num_parameters: 33000000000
+    release_date: 2023-12-01
+    # Does not support echo
+    tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
+
+  - name: writer/palmyra-x-v3
+    display_name: Palmyra X V3 (72B)
+    description: Palmyra-X V3 (72B parameters) is a Transformer-based model, which is trained on extremely large-scale pre-training data. It is trained via unsupervised learning and DPO and use multiquery attention.
+    creator_organization_name: Writer
+    access: limited
+    num_parameters: 72000000000
+    release_date: 2023-12-01
+    # Does not support echo
+    tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
+
+  - name: writer/palmyra-x-32k
+    display_name: Palmyra X-32K (33B)
+    description: Palmyra-X-32K (33B parameters) is a Transformer-based model, which is trained on large-scale pre-training data. The pre-training data types are diverse and cover a wide range of areas. These data types are used in conjunction and the alignment mechanism to extend context window.
+    creator_organization_name: Writer
+    access: limited
+    num_parameters: 33000000000
+    release_date: 2023-12-01
+    # Does not support echo
+    tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG]
+
 
 
   # Yandex