Skip to content

Commit e39187c

Browse files
ylwu-amznylwu-amzn
authored
add blueprint for Cluade 3.7 on Bedrock (#3584)
Signed-off-by: Yaliang Wu <ylwu@amazon.com>
1 parent dbfe0c2 commit e39187c

File tree

1 file changed

+349
-0
lines changed

1 file changed

+349
-0
lines changed

‎docs/remote_inference_blueprints/bedrock_connector_anthropic_claude3.7_blueprint.md

Lines changed: 349 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,349 @@
1+
# Bedrock connector blueprint example for Claude 3.7 model
2+
3+
Anthropic's Claude 3.7 Sonnet model is now available on Amazon Bedrock. For more details, check out this [blog](https://aws.amazon.com/blogs/aws/anthropics-claude-3-7-sonnet-the-first-hybrid-reasoning-model-is-now-available-in-amazon-bedrock/).
4+
5+
Claude 3.7 is Anthropic's first hybrid reasoning model, supporting two modes: standard and extended thinking. This doc covers both modes.
6+
7+
## 1. Add connector endpoint to trusted URLs:
8+
9+
Note: This step is only necessary for OpenSearch versions prior to 2.11.0.
10+
11+
```json
12+
PUT /_cluster/settings
13+
{
14+
"persistent": {
15+
"plugins.ml_commons.trusted_connector_endpoints_regex": [
16+
"^https://bedrock-runtime\\..*[a-z0-9-]\\.amazonaws\\.com/.*$"
17+
]
18+
}
19+
}
20+
```
21+
22+
## 2. Standard mode
23+
### 2.1 Create connector
24+
25+
If you are using self-managed Opensearch, you should supply AWS credentials:
26+
27+
Note:
28+
1. User needs to use [inference profile](https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-support.html) for invocation of this model. We can see the profile ID for Claude 3.7 is `us.anthropic.claude-3-7-sonnet-20250219-v1:0` for three available US regions `us-east-1`, `us-east-2`, `us-west-2`.
29+
30+
```json
31+
POST /_plugins/_ml/connectors/_create
32+
{
33+
"name": "Amazon Bedrock claude v3.7",
34+
"description": "Test connector for Amazon Bedrock claude v3.7",
35+
"version": 1,
36+
"protocol": "aws_sigv4",
37+
"credential": {
38+
"access_key": "<PLEASE ADD YOUR AWS ACCESS KEY HERE>",
39+
"secret_key": "<PLEASE ADD YOUR AWS SECRET KEY HERE>",
40+
"session_token": "<PLEASE ADD YOUR AWS SECURITY TOKEN HERE>"
41+
},
42+
"parameters": {
43+
"region": "<PLEASE ADD YOUR AWS REGION HERE>",
44+
"service_name": "bedrock",
45+
"max_tokens": 8000,
46+
"temperature": 1,
47+
"anthropic_version": "bedrock-2023-05-31",
48+
"model": "us.anthropic.claude-3-7-sonnet-20250219-v1:0"
49+
},
50+
"actions": [
51+
{
52+
"action_type": "predict",
53+
"method": "POST",
54+
"headers": {
55+
"content-type": "application/json"
56+
},
57+
"url": "https://bedrock-runtime.${parameters.region}.amazonaws.com/model/${parameters.model}/invoke",
58+
"request_body": "{ \"anthropic_version\": \"${parameters.anthropic_version}\", \"max_tokens\": ${parameters.max_tokens}, \"temperature\": ${parameters.temperature}, \"messages\": ${parameters.messages} }"
59+
}
60+
]
61+
}
62+
```
63+
64+
If using the AWS Opensearch Service, you can provide an IAM role arn that allows access to the bedrock service.
65+
Refer to this [AWS doc](https://docs.aws.amazon.com/opensearch-service/latest/developerguide/ml-amazon-connector.html)
66+
67+
```json
68+
POST /_plugins/_ml/connectors/_create
69+
{
70+
"name": "Amazon Bedrock claude v3.7",
71+
"description": "Test connector for Amazon Bedrock claude v3.7",
72+
"version": 1,
73+
"protocol": "aws_sigv4",
74+
"credential": {
75+
"roleArn": "<PLEASE ADD YOUR AWS ROLE ARN HERE>"
76+
},
77+
"parameters": {
78+
"region": "<PLEASE ADD YOUR AWS REGION HERE>",
79+
"service_name": "bedrock",
80+
"max_tokens": 8000,
81+
"temperature": 1,
82+
"anthropic_version": "bedrock-2023-05-31",
83+
"model": "us.anthropic.claude-3-7-sonnet-20250219-v1:0"
84+
},
85+
"actions": [
86+
{
87+
"action_type": "predict",
88+
"method": "POST",
89+
"headers": {
90+
"content-type": "application/json"
91+
},
92+
"url": "https://bedrock-runtime.${parameters.region}.amazonaws.com/model/${parameters.model}/invoke",
93+
"request_body": "{ \"anthropic_version\": \"${parameters.anthropic_version}\", \"max_tokens\": ${parameters.max_tokens}, \"temperature\": ${parameters.temperature}, \"messages\": ${parameters.messages} }"
94+
}
95+
]
96+
}
97+
```
98+
99+
Sample response:
100+
```json
101+
{
102+
"connector_id": "fa5tP5UBX2k07okSp89B"
103+
}
104+
```
105+
106+
### 2.2 Register model
107+
108+
```json
109+
POST /_plugins/_ml/models/_register?deploy=true
110+
{
111+
"name": "anthropic.claude-v3.7",
112+
"function_name": "remote",
113+
"description": "claude v3.7 model",
114+
"connector_id": "fa5tP5UBX2k07okSp89B"
115+
}
116+
```
117+
118+
Sample response:
119+
```json
120+
{
121+
"task_id": "fq5uP5UBX2k07okSFM__",
122+
"status": "CREATED",
123+
"model_id": "f65uP5UBX2k07okSFc8P"
124+
}
125+
```
126+
127+
### 2.3 Test model inference
128+
129+
```json
130+
POST /_plugins/_ml/models/f65uP5UBX2k07okSFc8P/_predict
131+
{
132+
"parameters": {
133+
"messages": [
134+
{
135+
"role": "user",
136+
"content": [
137+
{
138+
"type": "text",
139+
"text": "hello world"
140+
}
141+
]
142+
}
143+
]
144+
}
145+
}
146+
```
147+
148+
Sample response:
149+
```json
150+
{
151+
"inference_results": [
152+
{
153+
"output": [
154+
{
155+
"name": "response",
156+
"dataAsMap": {
157+
"id": "msg_bdrk_012spGFGr4CcD1PWb2TSfYut",
158+
"type": "message",
159+
"role": "assistant",
160+
"model": "claude-3-7-sonnet-20250219",
161+
"content": [
162+
{
163+
"type": "text",
164+
"text": "Hello! It's nice to meet you. How can I help you today?"
165+
}
166+
],
167+
"stop_reason": "end_turn",
168+
"stop_sequence": null,
169+
"usage": {
170+
"input_tokens": 9.0,
171+
"cache_creation_input_tokens": 0.0,
172+
"cache_read_input_tokens": 0.0,
173+
"output_tokens": 19.0
174+
}
175+
}
176+
}
177+
],
178+
"status_code": 200
179+
}
180+
]
181+
}
182+
```
183+
184+
## 3. Extended thinking mode
185+
186+
Extended thinking mode allows Claude 3.7 to perform more in-depth reasoning before providing a response.
187+
188+
### 3.1 Create connector
189+
190+
```json
191+
POST /_plugins/_ml/connectors/_create
192+
{
193+
"name": "Amazon Bedrock claude v3.7",
194+
"description": "Test connector for Amazon Bedrock claude v3.7",
195+
"version": 1,
196+
"protocol": "aws_sigv4",
197+
"credential": {
198+
"access_key": "<PLEASE ADD YOUR AWS ACCESS KEY HERE>",
199+
"secret_key": "<PLEASE ADD YOUR AWS SECRET KEY HERE>",
200+
"session_token": "<PLEASE ADD YOUR AWS SECURITY TOKEN HERE>"
201+
},
202+
"parameters": {
203+
"region": "<PLEASE ADD YOUR AWS REGION HERE>",
204+
"service_name": "bedrock",
205+
"max_tokens": 8000,
206+
"temperature": 1,
207+
"anthropic_version": "bedrock-2023-05-31",
208+
"model": "us.anthropic.claude-3-7-sonnet-20250219-v1:0",
209+
"budget_tokens": 1024
210+
},
211+
"actions": [
212+
{
213+
"action_type": "predict",
214+
"method": "POST",
215+
"headers": {
216+
"content-type": "application/json"
217+
},
218+
"url": "https://bedrock-runtime.${parameters.region}.amazonaws.com/model/${parameters.model}/invoke",
219+
"request_body": "{ \"anthropic_version\": \"${parameters.anthropic_version}\", \"max_tokens\": ${parameters.max_tokens}, \"temperature\": ${parameters.temperature}, \"messages\": ${parameters.messages}, \"thinking\": {\"type\": \"enabled\", \"budget_tokens\": ${parameters.budget_tokens} } }"
220+
}
221+
]
222+
}
223+
```
224+
225+
If using the AWS Opensearch Service, you can provide an IAM role arn that allows access to the bedrock service.
226+
Refer to this [AWS doc](https://docs.aws.amazon.com/opensearch-service/latest/developerguide/ml-amazon-connector.html)
227+
228+
```json
229+
POST /_plugins/_ml/connectors/_create
230+
{
231+
"name": "Amazon Bedrock claude v3.7",
232+
"description": "Test connector for Amazon Bedrock claude v3.7",
233+
"version": 1,
234+
"protocol": "aws_sigv4",
235+
"credential": {
236+
"roleArn": "<PLEASE ADD YOUR AWS ROLE ARN HERE>"
237+
},
238+
"parameters": {
239+
"region": "<PLEASE ADD YOUR AWS REGION HERE>",
240+
"service_name": "bedrock",
241+
"max_tokens": 8000,
242+
"temperature": 1,
243+
"anthropic_version": "bedrock-2023-05-31",
244+
"model": "us.anthropic.claude-3-7-sonnet-20250219-v1:0",
245+
"budget_tokens": 1024
246+
},
247+
"actions": [
248+
{
249+
"action_type": "predict",
250+
"method": "POST",
251+
"headers": {
252+
"content-type": "application/json"
253+
},
254+
"url": "https://bedrock-runtime.${parameters.region}.amazonaws.com/model/${parameters.model}/invoke",
255+
"request_body": "{ \"anthropic_version\": \"${parameters.anthropic_version}\", \"max_tokens\": ${parameters.max_tokens}, \"temperature\": ${parameters.temperature}, \"messages\": ${parameters.messages}, \"thinking\": {\"type\": \"enabled\", \"budget_tokens\": ${parameters.budget_tokens} } }"
256+
}
257+
]
258+
}
259+
```
260+
261+
Sample response:
262+
```json
263+
{
264+
"connector_id": "1652P5UBX2k07okSys_J"
265+
}
266+
```
267+
268+
### 3.2 Register model:
269+
270+
```json
271+
POST /_plugins/_ml/models/_register?deploy=true
272+
{
273+
"name": "anthropic.claude-v3.7",
274+
"function_name": "remote",
275+
"description": "claude v3.7 model",
276+
"connector_id": "1652P5UBX2k07okSys_J"
277+
}
278+
```
279+
280+
Sample response:
281+
```json
282+
{
283+
"task_id": "5K53P5UBX2k07okSXc-7",
284+
"status": "CREATED",
285+
"model_id": "5a53P5UBX2k07okSXc_M"
286+
}
287+
```
288+
289+
### 3.3 Test model inference
290+
291+
```json
292+
POST /_plugins/_ml/models/5a53P5UBX2k07okSXc_M/_predict
293+
{
294+
"parameters": {
295+
"messages": [
296+
{
297+
"role": "user",
298+
"content": [
299+
{
300+
"type": "text",
301+
"text": "hello world"
302+
}
303+
]
304+
}
305+
]
306+
}
307+
}
308+
```
309+
310+
Sample response:
311+
```json
312+
{
313+
"inference_results": [
314+
{
315+
"output": [
316+
{
317+
"name": "response",
318+
"dataAsMap": {
319+
"id": "msg_bdrk_01TqgZsyqsxhNGAGVjRjCP6N",
320+
"type": "message",
321+
"role": "assistant",
322+
"model": "claude-3-7-sonnet-20250219",
323+
"content": [
324+
{
325+
"type": "thinking",
326+
"thinking": "This is a simple greeting phrase \"hello world\" which is often the first program someone writes when learning a new programming language. The person could be:\n1. Simply greeting me casually\n2. Making a reference to programming\n3. Testing if I'm working\n\nI'll respond with a friendly greeting that acknowledges the \"hello world\" phrase and its connection to programming culture, while being conversational.",
327+
"signature": "EqgBCkYQARgCIkBAw8igmzDbjCXk1nLyWHFoS46Pi0VzkRj0IOySgDccftUjmMHmfM7nEWxPnjsgT3SEpFDDcS4Sj5gAiWjbejGTEgyCZKvfxX/J+ss/cM8aDH7t4SO42Cmpg+heNyIw2jyV6CD0lsx+mJAh6CuJtxXrfrp/VAu8FERnH2JTovSm0U3KYf+vRwfJNMdb3AsnKhAFuA24st/+8w6g5Xwxdqce"
328+
},
329+
{
330+
"type": "text",
331+
"text": "Hello! It's nice to meet you. \"Hello world\" is such a classic phrase - it's often the first program many people write when learning to code! How are you doing today? Is there something I can help you with?"
332+
}
333+
],
334+
"stop_reason": "end_turn",
335+
"stop_sequence": null,
336+
"usage": {
337+
"input_tokens": 37.0,
338+
"cache_creation_input_tokens": 0.0,
339+
"cache_read_input_tokens": 0.0,
340+
"output_tokens": 143.0
341+
}
342+
}
343+
}
344+
],
345+
"status_code": 200
346+
}
347+
]
348+
}
349+
```

0 commit comments

Comments
 (0)