[chore] support two types of cross-region inference (global, geographic) (#2146)

Co-authored-by: Yuanbo Li <ybalbert@amazon.com>

Author: ybalbert001

models/bedrock/manifest.yaml

@@ -1,4 +1,4 @@
-version: 0.0.53
+version: 0.0.54
 type: plugin
 author: langgenius
 name: bedrock

models/bedrock/models/llm/amazon-nova.yaml

@@ -29,14 +29,18 @@ parameter_rules:
       - Nova Micro
   - name: cross-region
     label:
-      zh_Hans: 使用跨区域推理
-      en_US: Use Cross-Region Inference
-    type: boolean
+      zh_Hans: 跨区域推理
+      en_US: Cross-Region Inference
+    type: string
     required: true
-    default: true
+    default: geographic
+    options:
+      - disabled
+      - geographic
+      - global
     help:
-      zh_Hans: 跨区域推理会自动选择您所在地理区域 AWS 区域 内的最佳位置来处理您的推理请求。
-      en_US: Cross-Region inference automatically selects the optimal AWS Region within your geography to process your inference request.
+      zh_Hans: 跨区域推理会自动选择 AWS 区域内的最佳位置来处理您的推理请求。地理跨区域限于您所在地理区域，全球跨区域可跨所有区域。
+      en_US: Cross-Region inference automatically selects the optimal AWS Region to process your inference request. Geographic limits to your geography, Global spans all regions.
   - name: system_cache_checkpoint
     label:
       zh_Hans: 缓存系统提示词

models/bedrock/models/llm/anthropic-claude.yaml

@@ -23,8 +23,9 @@ parameter_rules:
       zh_Hans: 指定模型名称
       en_US: specify model name
     required: true
-    default: Claude 3.7 Sonnet
+    default: Claude 4.5 Sonnet
     options:
+      - Claude 4.5 Opus
       - Claude 4.5 Haiku
       - Claude 4.5 Sonnet
       - Claude 4.0 Sonnet
@@ -39,14 +40,18 @@ parameter_rules:
       - Claude 3 Opus
   - name: cross-region
     label:
-      zh_Hans: 使用跨区域推理
-      en_US: Use Cross-Region Inference
-    type: boolean
+      zh_Hans: 跨区域推理
+      en_US: Cross-Region Inference
+    type: string
     required: true
-    default: true
+    default: global
+    options:
+      - disabled
+      - geographic
+      - global
     help:
-      zh_Hans: 跨区域推理会自动选择您所在地理区域 AWS 区域 内的最佳位置来处理您的推理请求。
-      en_US: Cross-Region inference automatically selects the optimal AWS Region within your geography to process your inference request.
+      zh_Hans: 跨区域推理会自动选择 AWS 区域内的最佳位置来处理您的推理请求。地理跨区域限于您所在地理区域，全球跨区域可跨所有区域。
+      en_US: Cross-Region inference automatically selects the optimal AWS Region to process your inference request. Geographic limits to your geography, Global spans all regions.
   - name: system_cache_checkpoint
     label:
       zh_Hans: 缓存系统提示词

models/bedrock/models/llm/cohere.yaml

@@ -28,14 +28,18 @@ parameter_rules:
       - Command Light
   - name: cross-region
     label:
-      zh_Hans: 使用跨区域推理
-      en_US: Use Cross-Region Inference
-    type: boolean
+      zh_Hans: 跨区域推理
+      en_US: Cross-Region Inference
+    type: string
     required: true
-    default: true
+    default: geographic
+    options:
+      - disabled
+      - geographic
+      - global
     help:
-      zh_Hans: 跨区域推理会自动选择您所在地理区域 AWS 区域 内的最佳位置来处理您的推理请求。
-      en_US: Cross-Region inference automatically selects the optimal AWS Region within your geography to process your inference request.
+      zh_Hans: 跨区域推理会自动选择 AWS 区域内的最佳位置来处理您的推理请求。地理跨区域限于您所在地理区域，全球跨区域可跨所有区域。
+      en_US: Cross-Region inference automatically selects the optimal AWS Region to process your inference request. Geographic limits to your geography, Global spans all regions.
   - name: max_new_tokens
     use_template: max_tokens
     required: true

models/bedrock/models/llm/deepseek.yaml

@@ -27,14 +27,18 @@ parameter_rules:
       - DeepSeek V3.1
   - name: cross-region
     label:
-      zh_Hans: 使用跨区域推理
-      en_US: Use Cross-Region Inference
-    type: boolean
+      zh_Hans: 跨区域推理
+      en_US: Cross-Region Inference
+    type: string
     required: true
-    default: false
+    default: disabled
+    options:
+      - disabled
+      - geographic
+      - global
     help:
-      zh_Hans: 跨区域推理会自动选择您所在地理区域 AWS 区域 内的最佳位置来处理您的推理请求。
-      en_US: Cross-Region inference automatically selects the optimal AWS Region within your geography to process your inference request.
+      zh_Hans: 跨区域推理会自动选择 AWS 区域内的最佳位置来处理您的推理请求。地理跨区域限于您所在地理区域，全球跨区域可跨所有区域。
+      en_US: Cross-Region inference automatically selects the optimal AWS Region to process your inference request. Geographic limits to your geography, Global spans all regions.
   - name: max_tokens
     use_template: max_tokens
     required: true

models/bedrock/models/llm/llm.py

@@ -291,20 +291,12 @@ def _get_model_info(self, model: str, credentials: dict, model_parameters: dict)
             # Get region prefix for model ID construction
             region_name = credentials['aws_region']
             region_prefix = None
+            cross_region = model_parameters.pop('cross-region', 'disabled')
 
-            if model_parameters.pop('cross-region', False):
+            if cross_region in ('geographic', 'global'):
                 # Cross-region inference enabled
-                # Check if the model supports global prefix (currently mainly Claude 4 series)
-                supports_global = any(model_id.startswith(prefix) for prefix in [
-                    'anthropic.claude-sonnet-4', 'anthropic.claude-sonnet-4-5'
-                ])
-                
-                if supports_global:
-                    # Prefer using global prefix
-                    region_prefix = model_ids.get_region_area(region_name, prefer_global=True)
-                else:
-                    # Use traditional regional prefix
-                    region_prefix = model_ids.get_region_area(region_name, prefer_global=False)
+                prefer_global = (cross_region == 'global')
+                region_prefix = model_ids.get_region_area(region_name, prefer_global=prefer_global)
 
                 if not region_prefix:
                     raise InvokeError(f'Failed to get cross-region inference prefix for region {region_name}')
@@ -313,14 +305,6 @@ def _get_model_info(self, model: str, credentials: dict, model_parameters: dict)
                     raise InvokeError(f"Model {model_id} doesn't support cross-region inference")
 
                 model_id = "{}.{}".format(region_prefix, model_id)
-            elif model_ids.is_support_cross_region(model_id):
-                # Cross-region inference not enabled, but still add region prefix for all models
-                region_prefix = model_ids.get_region_area(region_name, prefer_global=False)
-                
-                if not region_prefix:
-                    raise InvokeError(f'Failed to get region prefix for region {region_name}')
-
-                model_id = "{}.{}".format(region_prefix, model_id)
 
 
             model_info = BedrockLargeLanguageModel._find_model_info(model_id)

models/bedrock/models/llm/meta-llama.yaml

@@ -30,14 +30,18 @@ parameter_rules:
       - Llama 3.2 90B Instruct
   - name: cross-region
     label:
-      zh_Hans: 使用跨区域推理
-      en_US: Use Cross-Region Inference
-    type: boolean
+      zh_Hans: 跨区域推理
+      en_US: Cross-Region Inference
+    type: string
     required: true
-    default: true
+    default: geographic
+    options:
+      - disabled
+      - geographic
+      - global
     help:
-      zh_Hans: 跨区域推理会自动选择您所在地理区域 AWS 区域 内的最佳位置来处理您的推理请求。
-      en_US: Cross-Region inference automatically selects the optimal AWS Region within your geography to process your inference request.
+      zh_Hans: 跨区域推理会自动选择 AWS 区域内的最佳位置来处理您的推理请求。地理跨区域限于您所在地理区域，全球跨区域可跨所有区域。
+      en_US: Cross-Region inference automatically selects the optimal AWS Region to process your inference request. Geographic limits to your geography, Global spans all regions.
   - name: temperature
     use_template: temperature
 

models/bedrock/models/llm/model_ids.py

@@ -8,6 +8,7 @@
 
 BEDROCK_MODEL_IDS = {
     'anthropic claude': {
+        'Claude 4.5 Opus': 'anthropic.claude-opus-4-5-20251101-v1:0',
         'Claude 4.5 Haiku': 'anthropic.claude-haiku-4-5-20251001-v1:0',
         'Claude 4.5 Sonnet': 'anthropic.claude-sonnet-4-5-20250929-v1:0',
         'Claude 4.0 Sonnet': 'anthropic.claude-sonnet-4-20250514-v1:0',

models/bedrock/models/llm/openai.yaml

@@ -26,14 +26,18 @@ parameter_rules:
       - GPT OSS 20B
   - name: cross-region
     label:
-      zh_Hans: 使用跨区域推理
-      en_US: Use Cross-Region Inference
-    type: boolean
+      zh_Hans: 跨区域推理
+      en_US: Cross-Region Inference
+    type: string
     required: true
-    default: false
+    default: disabled
+    options:
+      - disabled
+      - geographic
+      - global
     help:
-      zh_Hans: 跨区域推理会自动选择您所在地理区域 AWS 区域 内的最佳位置来处理您的推理请求。
-      en_US: Cross-Region inference automatically selects the optimal AWS Region within your geography to process your inference request.
+      zh_Hans: 跨区域推理会自动选择 AWS 区域内的最佳位置来处理您的推理请求。地理跨区域限于您所在地理区域，全球跨区域可跨所有区域。
+      en_US: Cross-Region inference automatically selects the optimal AWS Region to process your inference request. Geographic limits to your geography, Global spans all regions.
   - name: max_tokens
     use_template: max_tokens
     required: true

models/bedrock/models/llm/qwen.yaml

@@ -28,14 +28,18 @@ parameter_rules:
       - Qwen3 Coder 30B
   - name: cross-region
     label:
-      zh_Hans: 使用跨区域推理
-      en_US: Use Cross-Region Inference
-    type: boolean
+      zh_Hans: 跨区域推理
+      en_US: Cross-Region Inference
+    type: string
     required: true
-    default: false
+    default: disabled
+    options:
+      - disabled
+      - geographic
+      - global
     help:
-      zh_Hans: 跨区域推理会自动选择您所在地理区域 AWS 区域 内的最佳位置来处理您的推理请求。
-      en_US: Cross-Region inference automatically selects the optimal AWS Region within your geography to process your inference request.
+      zh_Hans: 跨区域推理会自动选择 AWS 区域内的最佳位置来处理您的推理请求。地理跨区域限于您所在地理区域，全球跨区域可跨所有区域。
+      en_US: Cross-Region inference automatically selects the optimal AWS Region to process your inference request. Geographic limits to your geography, Global spans all regions.
   - name: max_tokens
     use_template: max_tokens
     required: true