SageMaker connection

The credentials of that connection will be used to connect to SageMaker

Friendly name

Model purpose

Handling mode

Some models such as some Llama2 versions require header X-Amzn-SageMaker-Custom-Attributes with value accept_eula=true to indicate EULA acceptation

Custom HTTP headers

Custom Query Template*

<span class="help-inline">Template for query, in JSON format. You may use the following placeholders: __PROMPT__, __TOPP__, __TOPK__, __TEMPERATURE__, __MAX_TOKENS__, __FREQUENCY_PENALTY__, __PRESENCE_PENALTY__, __LOGIT_BIAS__</span>
                                    </div>
                                </div>
                                <div class="control-group" ng-if="useFullyCustomHandling()">
                                    <label for="responseJsonPath" class="control-label">Response*</label>
                                    <div class="controls">
                                        <input id="responseJsonPath" type="text" ng-model="connection.params.responseJsonPath" required />
                                        <span class="help-inline">JSON Path of response. If the response is a JSON object and the response is in field reply, it will be `$.reply' (without the quotes).</span>
                                    </div>
                                </div>
                                <div class="control-group" ng-if="useFullyCustomHandling()">
                                    <label for="finishReasonJsonPath" class="control-label">Finish reason</label>
                                    <div class="controls">
                                        <input id="finishReasonJsonPath" type="text" ng-model="connection.params.finishReasonJsonPath" />
                                        <span class="help-inline">Optional JSON Path of the finish reason. If the response is a JSON object and the finish reason is in field finishReason, it will be `$.finishReason' (without the quotes).</span>
                                    </div>
                                </div>
                                <div class="control-group" ng-if="useFullyCustomHandling()">
                                    <label for="promptTokensJsonPath" class="control-label">Prompt Tokens</label>
                                    <div class="controls">
                                        <input id="promptTokensJsonPath" type="text" ng-model="connection.params.promptTokensJsonPath" />
                                        <span class="help-inline">Optional JSON Path of prompt tokens. If the response is a JSON object and the prompt tokens are in field promptTokens, it will be `$.promptTokens' (without the quotes).</span>
                                    </div>
                                </div>
                                <div class="control-group" ng-if="useFullyCustomHandling()">
                                    <label for="completionTokensJsonPath" class="control-label">Completion Tokens</label>
                                    <div class="controls">
                                        <input id="completionTokensJsonPath" type="text" ng-model="connection.params.completionTokensJsonPath" />
                                        <span class="help-inline">Optional JSON Path of completion tokens. If the response is a JSON object and the completion tokens are in field completionTokens, it will be `$.completionTokens' (without the quotes).</span>
                                    </div>
                                </div>
                                <h2 class="settings-section-title">LLM Common Parameters</h2>
                                <div class="control-group">
                                    <label for="max-parallelism" class="control-label">Max parallelism</label>
                                    <div class="controls">
                                        <input id="max-parallelism" type="number" force-integer step="1" min="1" max="128" ng-model="connection.params.maxParallelism" />
                                    </div>
                                </div>
                                <div class="control-group">
                                    <label for="maxTokensLimit" class="control-label">Max tokens limit</label>
                                    <div class="controls">
                                        <input id="maxTokensLimit" type="number" force-integer step="1" min="1" max="2048" ng-model="connection.params.sageMakerModel.maxTokensLimit" />
                                    </div>
                                </div>
                                <h2 class="settings-section-title">Network & Tuning</h2>

<div class="control-group">
                                    <label for="networkTimeout" class="control-label">Network timeout</label>
                                    <div class="controls">
                                        <input id="networkTimeout" type="number" force-integer min="1000" ng-model="connection.params.networkSettings.queryTimeoutMS" />
                                        <span class="help-inline">In milliseconds</span>
                                    </div>
                                </div>
                                <div class="control-group">
                                    <label for="maxRetries" class="control-label">Max retries</label>
                                    <div class="controls">
                                        <input id="maxRetries" type="number" force-integer min="0" ng-model="connection.params.networkSettings.maxRetries" />
                                    </div>
                                </div>
                                <div include-no-scope="/templates/admin/fragments/connection-global-proxy.html" />
                            </div>
                            <div class="span6">
                                <div include-no-scope="/templates/admin/fragments/connection-flags-security-only.html" />
                                <div include-no-scope="/templates/admin/fragments/llm-connection-usage-control.html" />
                            </div>
                        </div>
                    </form>
                    <div include-no-scope="/templates/admin/cloud-source-ips.html" />
                </div>
                <div class="generic-white-box mtop20" ng-if="appConfig.customFieldsMap['CONNECTION'].length > 0">
                    <div include-no-scope="/templates/admin/fragments/connection-custom-fields.html" />
                </div>
            </div>
        </div>
    </div>
    <div include-no-scope=/templates/admin/fragments/llm-connections-error-handler.html />
</div>

Connection

Endpoint

Fully Custom Handling Parameters