ABAP® client library to use the watsonx.ai Runtime (formerly Watson Machine Learning) and the watsonx.ai APIs.
This is a community SDK written by ABAPers for the ABAPers in the Open Source community, to provide easy usage of IBM® watsonx™ services in innovation initiatives combined with any SAP® application which is run on SAP NetWeaver® 7.50 and above, such as SAP Business Suite or SAP S/4HANA® on-premises edition. It is the choice and responsibility of application developers how this community SDK is used.
Additionally, as the ABAP SDK is a community release it is not updated with the same schedule as IBM-supported SDKs. Please see more information in Support and Feedback.
| abap-sdk-nwas-x | abap-sdk-btp-x |
|---|---|
| for SAP NetWeaver® AS ABAP 7.50+ | for SAP BTP ABAP Environment |
| tested on SAP Business Suite and SAP S/4HANA on-premises edition | tested with SAP BTP ABAP Environment 2405 |
this repository |
Table of Contents
Version v1.5.0 of the SDK has been released - see what's changed in the migration guide.
- You need an IBM Cloud account.
The client library is provided as abapGit repository. Proceed as follows to clone the ABAP SDK code to your SAP system.
- Install abapGit, using the abapGit Docs
- Use abapGit to clone the ABAP SDK Git repository into your SAP system.
- Assign the ABAP SDK to the package
ZIBMXwhen performing setup of the abapGit repository.
The following SAP profile parameter settings are recommended.
icm/HTTPS/client_sni_enabled = TRUE
ssl/ciphersuites = 135:PFS:HIGH::EC_P256:EC_HIGH
ssl/client_ciphersuites = 150:PFS:HIGH::EC_P256:EC_HIGH
The client library respects the ABAP proxy settings. If you are using a central proxy server that is not already configured in your SAP system, proceed as follows.
- Logon to the SAP system and call transaction SICF.
- Click Execute (or press F8).
- Select menu item Client → Proxy Settings.
- On tab Global Settings: Select Proxy Setting is Active and No Proxy Setting for Local Server.
- On tabs HTTP Protocol and HTTPS Protocol specify the proxy Host Name and Port. If the proxy server requires logon credentials, also specify User Name and Password.
- Click Execute (F8) and OK.
Communication between SAP and the IBM Cloud is secured by the Transport Layer Security (TLS) protocol, also known as Secure Sockets Layer (SSL). SSL requires certificates that must be stored on the SAP application server in the Personal Security Environment (PSE). Transaction STRUST is used to maintain the PSE.
If a call to an IBM Cloud service failed due to a missing SSL certificate, check the ICM trace file to find the required certificate. To do so, call transaction SMICM and select menu item Goto → Trace File → Display End. Alternatively, call transaction AL11 and navigate to file DIR_HOME/dev_icm. Find last entry in trace file that indicates error SSSLERR_PEER_CERT_UNTRUSTED, as shown in the example below.
[Thr nn] Peer not trusted
[Thr nn] Certificate:
[Thr nn] Certificate:
[Thr nn] Subject: CN=*.watsonplatform.net, O=INTERNATIONAL BUSINESS MACHINES CORPORATION, L=Armonk, SP=New York, C=US
[Thr nn] Issuer: CN=GeoTrust RSA CA 2018, OU=www.digicert.com, O=DigiCert Inc, C=US
[Thr nn] Serial Number: 05:1C:46:1B:6E:C2:29:48:BD:F0:03:B3:A6:87:1A:5B
[Thr nn] Verification result:
[...]
[Thr nn] SignerCert:
[Thr nn] Certificate:
[Thr nn] Subject: CN=GeoTrust RSA CA 2018, OU=www.digicert.com, O=DigiCert Inc, C=US
[Thr nn] Issuer: CN=DigiCert Global Root CA, OU=www.digicert.com, O=DigiCert Inc, C=US
[Thr nn] Serial Number: 05:46:FE:18:23:F7:E1:94:1D:A3:9F:CE:14:C4:61:73
[Thr nn] Verification result:
[...]
[Thr nn] SignerCert:
[Thr nn] Certificate:
[Thr nn] Subject: CN=DigiCert Global Root CA, OU=www.digicert.com, O=DigiCert Inc, C=US
[...]
[Thr nn] << ---------- End of Secu-SSL Errorstack ----------
[Thr nn] Target Hostname="stream.watsonplatform.net"
[Thr nn] SSL NI-hdl 463: local=9.152.20.145:56542 peer=169.48.115.62:443
[Thr nn] <<- ERROR: SapSSLSessionStartNB(sssl_hdl=3fef0000c70)==SSSLERR_PEER_CERT_UNTRUSTED
[Thr nn] *** ERROR => SSL handshake with stream.watsonplatform.net:443 failed: SSSLERR_PEER_CERT_UNTRUSTED (-102)
[Thr nn] The peer's X.509 Certificate (chain) is untrusted
The SSL error stack shows the SSL certificate chain, which in the example above looks like follows:
| Certificate Authority Type | Certificate Authority (CA) |
|---|---|
| end-user | *.watsonplatform.net |
| intermediate | GeoTrust RSA CA 2018 |
| root | DigiCert Global Root CA |
You can install any of these SSL certificates. However, it is
recommended to install the root CA certificate.
Download the
appropriate SSL certificate from the provider's website, for example
https://www.digicert.com/digicert-root-certificates.htm. Store the
certificate file on your computer.
Proceed as follows to add the downloaded SSL certificate to your SAP system's Standard SSL Client PSE.
- Logon to the SAP system and call transaction STRUST.
- Switch to edit mode (press according tool bar icon).
- If a local PSE file does not exist already, create it by
right-clicking on
SSL client SSL Client (Standard)and selecting Create from context menu. Keep all default settings in next popup dialog. - In Certificate section, click Import (alternatively select menu item Certificate → Import). Choose the certificate file that you have downloaded and import the certificate.
- Click button Add to Certificate List.
- Click Save (F3).
It is recommended to restart the ICM after a new SSL certificate has been applied to the PSE. To do so, proceed as follows.
- Logon to the SAP system and start transaction SMICM.
- If applicable, ensure that the ICM is restarted automatically when ended: Administration → ICM → Restart → Yes.
- Select menu item Administration → ICM → Exit Soft → Global.
Before you can access a specific service from your SAP system, you must create a service instance in the IBM Cloud and obtain an API key. The API key can be shared among multiple end users of the SAP system, but you must ensure that the selected plan is sufficient to support the expected number of calls or the expected data volume.
All currently supported services support IAM authentication (see
below). Service credentials consist of an API key and a URL. Both the
API key and the URL are character values that need to be provided as parameters
i_apikey and i_url to method zcl_ibmx_service_ext=>get_instance.
You can store the values with your application, but it is suggested to do that in an encrypted format. Using cloud services usually creates costs based on usage for the owner of the service instance, and anyone with the credentials can use the service instance at the owner's expenses. If you want to distribute the costs over multiple cost centers, you need to create a service instance and provide service credentials for each cost center separately.
Service credentials and other parameters that must be specified at
watsonx service wrapper ABAP class instantiation can also be provided
in table ZIBMX_CONFIG. This table has three keys:
| Table Key | Description |
|---|---|
| SERVICE | The ABAP class name without prefix ZCL_IBMX_ |
| INSTANCE_UID | ID chosen by application developer that must be provided by application as parameter to method zcl_ibmx_service_ext=>get_instance() |
| PARAM | The parameter name |
List of configuration parameters in table ZIBMX_CONFIG
| Parameter Name | Default Value | Description |
|---|---|---|
| URL | service-dependent | watsonx service url |
| APIKEY | watsonx service API keys | |
| PROXY_HOST | Proxy server | |
| PROXY_PORT | Proxy server port | |
| AUTH_NAME | service-dependent | Authorization, IAM or basicAuth |
| SSL_ID | CLIENT | SSL identity, defines PSE for SSL certificates: CLIENT or ANONYM |
Identity and Access Management (IAM) is a bearer-token based
authentication method. Token management is either performed by the
ABAP SDK or must be implemented by the SDK user.
If a value for
i_apikey is provided by the caller in method
zcl_ibmx_service_ext=>get_instance(), the ABAP SDK generates a
bearer-token under the cover when needed and refreshes it when it is
about to expire.
If i_apikey is not provided for method
zcl_ibmx_service_ext=>get_instance(), the ABAP SDK user must
implement an individual token management. Before a service method is
called the first time, a valid bearer-token must be provided to the
watsonx service wrapper ABAP class instance as follows:
lo_service_class->set_bearer_token( i_bearer_token = '...' ).Afterwards, service methods can be called as long as the provided token is valid. When the token has expired, the method above must be called again with a new (refreshed) bearer-token as parameter.
The client library is delivered as package ZIBMX. Once the Git
repository has been cloned to the SAP system, a watsonx service
instance is wrapped by an ABAP class instance.
The following
watsonx services are currently supported:
| Service | ABAP Class Name |
|---|---|
| watsonx.ai Runtime | ZCL_IBMX_WML_V4 |
| watsonx.ai | ZCL_IBMX_WATSONX_AI_ML_V1 |
Using the client library requires two steps:
-
Create an instance of the watsonx service wrapper ABAP class by calling method
zcl_ibmx_service_ext=>get_instance.data: lo_service_class type <ABAP Class Name>. zcl_ibmx_service_ext=>get_instance( exporting i_url = <url> i_apikey = <api key> ... importing eo_instance = lo_service_class ).
Note
Depending on the API operation, the<url>parameter may vary as described in Endpoint URLs. If you use API operations with different endpoint URLs in the same program, you need to create an instance of the watsonx service wrapper ABAP class for each endpoint URL and use the instances accordingly in your program. -
Call the watsonx service API endpoint by invoking the corresponding class method.
try. lo_service_class->method( exporting is_input = ... importing es_output = ... ). catch zcx_ibmx_service_exception into data(lo_service_exception). ... endtry.
watsonx.ai Example (AI Sevices)
* Z_WATSONX_AI_SERVICES_DEMO
* This sample code deploys arbitrary Python code as Python AI Service on a watsonx deployment space
* and calls the deployed function.
* It also returns a CURL command that can be called to invoke the function endpoint url.
*
* Copy this code into
* - an ABAP report and remove all comment prefixes "[PRG]
* - a Console Application (RAP) and remove all comment prefixes "[RAP]
* Adjust credentials and run code.
constants:
" Prerequisite: watsonx deployment space must be created and id specified here:
c_url type string value `https://eu-de.ml.cloud.ibm.com`, " <- ADJUST
c_apikey type string value `KHds7uuH...A7892Gt6`, " <- ADJUST
c_space_id type string value `c3a99662...21349082`, " <- ADJUST
c_function_name type string value `demo-ai-service-function`,
c_serving_name type string value `abapsdk_test`,
c_version type string value `2024-03-14`.
" convert Python function code to GZIP file (ZIP with file header 0x04034b50 is not supported)
try.
" some Python code that bounces the received data (http method, path parameter, payload (= json request body)) as response body
data(lx_code) = zcl_ibmx_service=>convert_string_to_utf8(
i_string =
`def ai_service_function(context):` && cl_abap_char_utilities=>newline &&
` def generate(context) -> dict:` && cl_abap_char_utilities=>newline &&
` payload = context.get_json()` && cl_abap_char_utilities=>newline &&
` path_suffix = context.get_path_suffix()` && cl_abap_char_utilities=>newline &&
` method = context.get_method()` && cl_abap_char_utilities=>newline &&
` return {'body': {'method': method, 'path': path_suffix, 'payload': payload}}` && cl_abap_char_utilities=>newline &&
` return generate`
).
cl_abap_gzip=>compress_binary_with_header(
exporting
raw_in = lx_code
importing
gzip_out = data(lx_gzipfile)
).
catch cx_root into data(lo_zip_exception).
"[RAP] out->write( lo_zip_exception->get_longtext( ) ). exit.
"[PRG] message lo_zip_exception type 'E'.
endtry.
" instantiate watsonx.ai wrapper class (explicit type declaration is required)
data: lo_watsonx_ai type ref to zcl_ibmx_watsonx_ai_ml_v1.
zcl_ibmx_service_ext=>get_instance(
exporting
i_url = c_url
i_apikey = c_apikey
i_version = c_version
importing
eo_instance = lo_watsonx_ai ).
" create Python function as asset in watsonx deployment space
try.
lo_watsonx_ai->ai_services_create(
exporting
i_aiservicerequest = value zcl_ibmx_watsonx_ai_ml_v1=>t_aiservice_request(
space_id = c_space_id
name = c_function_name
description = 'demo'
software_spec = value zcl_ibmx_watsonx_ai_ml_v1=>t_software_spec_rel(
name = `runtime-24.1-py3.11`
)
)
importing
e_response = data(ls_result_create)
).
catch zcx_ibmx_service_exception into data(lo_create_exception).
"[RAP] out->write( lo_create_exception->get_longtext( ) ). exit.
"[PRG] message lo_create_exception type 'E'.
endtry.
data(lv_aiservice_id) = ls_result_create-metadata-id.
" upload AI Service Python function code
try.
lo_watsonx_ai->ai_services_upload_code(
exporting
i_id = lv_aiservice_id
i_space_id = c_space_id
i_upload_code = lx_gzipfile
i_contenttype = 'application/gzip'
importing
e_response = data(ls_result_upload)
).
catch zcx_ibmx_service_exception into data(lo_upload_exception).
"[RAP] out->write( lo_upload_exception->get_longtext( ) ). exit.
"[PRG] message lo_upload_exception type 'E'.
endtry.
" deploy AI Service Python function
try.
lo_watsonx_ai->create_deployment(
exporting
i_deploymentresourceprototype = value zcl_ibmx_watsonx_ai_ml_v1=>t_deployment_resrc_prototype(
space_id = c_space_id
name = `Deployment for ` && c_function_name
asset = value zcl_ibmx_watsonx_ai_ml_v1=>t_rel(
id = lv_aiservice_id
)
hardware_spec = value zcl_ibmx_watsonx_ai_ml_v1=>t_hardware_spec(
name = 'S'
)
online = value zcl_ibmx_watsonx_ai_ml_v1=>t_online_deployment(
parameters = value zcl_ibmx_watsonx_ai_ml_v1=>t_online_deployment_parameters(
serving_name = c_serving_name
)
)
)
importing
e_response = data(ls_deployment_result)
).
catch zcx_ibmx_service_exception into data(lo_deployment_exception).
" http status 400 (bad request) -> delete existing deployment, if exists, and retry
"[RAP] out->write( lo_deployment_exception->get_longtext( ) ). exit.
"[PRG] message lo_deployment_exception type 'E'.
endtry.
data(lv_deployment_id) = ls_deployment_result-metadata-id.
" wait until deployment is complete
do 20 times.
try.
lo_watsonx_ai->deployments_get(
exporting
i_space_id = c_space_id
i_deployment_id = lv_deployment_id
importing
e_response = data(ls_deployment)
).
catch zcx_ibmx_service_exception into data(lo_deployment_get_exception).
"[RAP] out->write( lo_deployment_get_exception->get_longtext( ) ). exit.
"[PRG] message lo_deployment_get_exception type 'E'.
endtry.
if not ls_deployment-entity-status-state eq 'initializing'.
exit.
endif.
wait up to 5 seconds.
enddo.
" INFERENCE
" declare structure of request response
data:
begin of lv_response,
begin of payload,
message type string,
end of payload,
path type string,
method type string,
end of lv_response.
" call AI Service function endpoint and parse response
data(lv_url) = `/ml/v4/deployments/` && c_serving_name && `/ai_service/mypath`.
data(ls_request_prop) = value zcl_ibmx_service=>ts_request_prop(
url = value zcl_ibmx_service=>ts_url(
path = lv_url
)
body = `{ "message": "This message was sent via ABAP" }`
header_content_type = 'application/json'
).
try.
data(lo_response) = lo_watsonx_ai->http_post( i_request_prop = ls_request_prop ).
data(lv_json) = zcl_ibmx_service=>get_response_string( lo_response ).
zcl_ibmx_service=>parse_json(
exporting
i_json = lv_json
changing
c_abap = lv_response ).
catch zcx_ibmx_service_exception into data(lo_inference_exception).
"[RAP] out->write( lo_inference_exception->get_longtext( ) ). exit.
"[PRG] message lo_inference_exception type 'E'.
endtry.
" compile CURL command to test deployment endpoint
data(lv_curl_command) =
`curl -X POST -H "Authorization: Bearer ${IAM_ACCESS_TOKEN}" `
&& `-d '{ "message": "This message was sent via CURL" }' `
&& `"` && ls_deployment-entity-status-inference[ 1 ]-url && `/mypath"`.
" write success message
data(lv_msg) = `Received message "` && lv_response-payload-message && `" from deployed AI Service function. Try CURL command below.`.
"[RAP] out->write( lv_msg ).
"[RAP] out->write( lv_curl_command ).
"[PRG] write: lv_msg.
"[PRG] write: / lv_curl_command.watsonx.ai Example (Text Generation)
* Z_WATSONX_AI_DEMO
* This sample code runs text generation with a watsonx.ai foundation model.
*
* Copy this code into
* - an ABAP report and remove all comment prefixes "[PRG]
* - a Console Application (RAP) and remove all comment prefixes "[RAP]
* Adjust credentials and run code.
" instantiate watsonx.ai wrapper class (explicit type declaration is required)
data: lo_watsonx_ai type ref to zcl_ibmx_watsonx_ai_ml_v1.
zcl_ibmx_service_ext=>get_instance(
exporting
i_url = 'https://eu-de.ml.cloud.ibm.com' " <- ADJUST
i_apikey = 'X7Ziqsan...FfOAtAno' " <- ADJUST
i_version = '2023-05-29'
importing
eo_instance = lo_watsonx_ai ).
" run text generation
try.
data(lv_prompt) = `The highest mountain in Europe is`.
lo_watsonx_ai->text_generation(
exporting
i_textgenrequest = value zcl_ibmx_watsonx_ai_ml_v1=>t_text_gen_request(
" prompt
input = lv_prompt
" model parameters
model_id = 'ibm/granite-13b-chat-v2'
parameters = value zcl_ibmx_watsonx_ai_ml_v1=>t_text_gen_parameters(
decoding_method = 'greedy'
max_new_tokens = 20
repetition_penalty = '1.05'
stop_sequences = value #( ( `. ` ) ) " stop after first sentence
)
project_id = '3e606d0c...459b7c38' " <- ADJUST
)
importing
e_response = data(ls_generated_document) ).
catch zcx_ibmx_service_exception into data(lo_service_exception).
"[RAP] out->write( lo_service_exception->get_longtext( ) ). exit.
"[PRG] message lo_service_exception type 'E'.
endtry.
" display generated text
"[RAP] out->write( lv_prompt ).
"[PRG] write lv_prompt.
loop at ls_generated_document-results into data(ls_result).
"[RAP] out->write( ls_result-generated_text ).
"[PRG] write / ls_result-generated_text.
endloop.GitHub Pages contain the ABAP Client Library for watsonx API Reference.
The ABAP SDK is a community SDK for IBM watsonx, created by the IBM watsonx development community and SAP's ABAP development community – written by ABAPers from IBM Systems. Therefore, as a community release it is not updated with the same schedule as IBM-supported SDKs and does not include support by IBM.
The code in this repository is generated from the OpenAPI specs of the supported services. Hence, pull requests must not be integrated directly into the code. If you would like to request a new feature or report a bug, please raise an issue instead of sending a pull request.
This library is licensed under the Apache 2.0 license.