|
| 1 | +/* |
| 2 | + * Copyright 2024 the original author or authors. |
| 3 | + * |
| 4 | + * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | + * you may not use this file except in compliance with the License. |
| 6 | + * You may obtain a copy of the License at |
| 7 | + * |
| 8 | + * https://www.apache.org/licenses/LICENSE-2.0 |
| 9 | + * |
| 10 | + * Unless required by applicable law or agreed to in writing, software |
| 11 | + * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | + * See the License for the specific language governing permissions and |
| 14 | + * limitations under the License. |
| 15 | + */ |
| 16 | +package org.springframework.ai.oci; |
| 17 | + |
| 18 | +import java.util.ArrayList; |
| 19 | +import java.util.List; |
| 20 | +import java.util.Objects; |
| 21 | +import java.util.concurrent.atomic.AtomicInteger; |
| 22 | + |
| 23 | +import com.oracle.bmc.generativeaiinference.GenerativeAiInference; |
| 24 | +import com.oracle.bmc.generativeaiinference.model.DedicatedServingMode; |
| 25 | +import com.oracle.bmc.generativeaiinference.model.EmbedTextDetails; |
| 26 | +import com.oracle.bmc.generativeaiinference.model.EmbedTextResult; |
| 27 | +import com.oracle.bmc.generativeaiinference.model.OnDemandServingMode; |
| 28 | +import com.oracle.bmc.generativeaiinference.model.ServingMode; |
| 29 | +import com.oracle.bmc.generativeaiinference.requests.EmbedTextRequest; |
| 30 | +import io.micrometer.observation.ObservationRegistry; |
| 31 | +import org.springframework.ai.chat.metadata.EmptyUsage; |
| 32 | +import org.springframework.ai.document.Document; |
| 33 | +import org.springframework.ai.embedding.AbstractEmbeddingModel; |
| 34 | +import org.springframework.ai.embedding.Embedding; |
| 35 | +import org.springframework.ai.embedding.EmbeddingOptions; |
| 36 | +import org.springframework.ai.embedding.EmbeddingRequest; |
| 37 | +import org.springframework.ai.embedding.EmbeddingResponse; |
| 38 | +import org.springframework.ai.embedding.EmbeddingResponseMetadata; |
| 39 | +import org.springframework.ai.embedding.observation.DefaultEmbeddingModelObservationConvention; |
| 40 | +import org.springframework.ai.embedding.observation.EmbeddingModelObservationContext; |
| 41 | +import org.springframework.ai.embedding.observation.EmbeddingModelObservationConvention; |
| 42 | +import org.springframework.ai.embedding.observation.EmbeddingModelObservationDocumentation; |
| 43 | +import org.springframework.ai.model.ModelOptionsUtils; |
| 44 | +import org.springframework.ai.observation.conventions.AiProvider; |
| 45 | +import org.springframework.util.Assert; |
| 46 | + |
| 47 | +/** |
| 48 | + * {@link org.springframework.ai.embedding.EmbeddingModel} implementation that uses the |
| 49 | + * OCI GenAI Embedding API. |
| 50 | + * |
| 51 | + * @author Anders Swanson |
| 52 | + * @since 1.0.0 |
| 53 | + */ |
| 54 | +public class OCIEmbeddingModel extends AbstractEmbeddingModel { |
| 55 | + |
| 56 | + // The OCI GenAI API has a batch size of 96 for embed text requests. |
| 57 | + private static final int EMBEDTEXT_BATCH_SIZE = 96; |
| 58 | + |
| 59 | + private static final EmbeddingModelObservationConvention DEFAULT_OBSERVATION_CONVENTION = new DefaultEmbeddingModelObservationConvention(); |
| 60 | + |
| 61 | + private final GenerativeAiInference genAi; |
| 62 | + |
| 63 | + private final OCIEmbeddingOptions options; |
| 64 | + |
| 65 | + private final ObservationRegistry observationRegistry; |
| 66 | + |
| 67 | + private final EmbeddingModelObservationConvention observationConvention = DEFAULT_OBSERVATION_CONVENTION; |
| 68 | + |
| 69 | + public OCIEmbeddingModel(GenerativeAiInference genAi, OCIEmbeddingOptions options) { |
| 70 | + this(genAi, options, ObservationRegistry.NOOP); |
| 71 | + } |
| 72 | + |
| 73 | + public OCIEmbeddingModel(GenerativeAiInference genAi, OCIEmbeddingOptions options, |
| 74 | + ObservationRegistry observationRegistry) { |
| 75 | + Assert.notNull(genAi, "com.oracle.bmc.generativeaiinference.GenerativeAiInferenceClient must not be null"); |
| 76 | + Assert.notNull(options, "options must not be null"); |
| 77 | + Assert.notNull(observationRegistry, "observationRegistry must not be null"); |
| 78 | + this.genAi = genAi; |
| 79 | + this.options = options; |
| 80 | + this.observationRegistry = observationRegistry; |
| 81 | + } |
| 82 | + |
| 83 | + @Override |
| 84 | + public EmbeddingResponse call(EmbeddingRequest request) { |
| 85 | + Assert.notEmpty(request.getInstructions(), "At least one text is required!"); |
| 86 | + OCIEmbeddingOptions runtimeOptions = mergeOptions(request.getOptions(), options); |
| 87 | + List<EmbedTextRequest> embedTextRequests = createRequests(request.getInstructions(), runtimeOptions); |
| 88 | + |
| 89 | + EmbeddingModelObservationContext context = EmbeddingModelObservationContext.builder() |
| 90 | + .embeddingRequest(request) |
| 91 | + .provider(AiProvider.OCI_GENAI.value()) |
| 92 | + .requestOptions(runtimeOptions) |
| 93 | + .build(); |
| 94 | + |
| 95 | + return EmbeddingModelObservationDocumentation.EMBEDDING_MODEL_OPERATION |
| 96 | + .observation(this.observationConvention, DEFAULT_OBSERVATION_CONVENTION, () -> context, |
| 97 | + this.observationRegistry) |
| 98 | + .observe(() -> embedAllWithContext(embedTextRequests, context)); |
| 99 | + } |
| 100 | + |
| 101 | + @Override |
| 102 | + public float[] embed(Document document) { |
| 103 | + return embed(document.getContent()); |
| 104 | + } |
| 105 | + |
| 106 | + private EmbeddingResponse embedAllWithContext(List<EmbedTextRequest> embedTextRequests, |
| 107 | + EmbeddingModelObservationContext context) { |
| 108 | + String modelId = null; |
| 109 | + AtomicInteger index = new AtomicInteger(0); |
| 110 | + List<Embedding> embeddings = new ArrayList<>(); |
| 111 | + for (EmbedTextRequest embedTextRequest : embedTextRequests) { |
| 112 | + EmbedTextResult embedTextResult = genAi.embedText(embedTextRequest).getEmbedTextResult(); |
| 113 | + if (modelId == null) { |
| 114 | + modelId = embedTextResult.getModelId(); |
| 115 | + } |
| 116 | + for (List<Float> e : embedTextResult.getEmbeddings()) { |
| 117 | + float[] data = toFloats(e); |
| 118 | + embeddings.add(new Embedding(data, index.getAndIncrement())); |
| 119 | + } |
| 120 | + } |
| 121 | + EmbeddingResponseMetadata metadata = new EmbeddingResponseMetadata(); |
| 122 | + metadata.setModel(modelId); |
| 123 | + metadata.setUsage(new EmptyUsage()); |
| 124 | + EmbeddingResponse embeddingResponse = new EmbeddingResponse(embeddings, metadata); |
| 125 | + context.setResponse(embeddingResponse); |
| 126 | + return embeddingResponse; |
| 127 | + } |
| 128 | + |
| 129 | + private ServingMode servingMode(OCIEmbeddingOptions embeddingOptions) { |
| 130 | + return switch (embeddingOptions.getServingMode()) { |
| 131 | + case "dedicated" -> DedicatedServingMode.builder().endpointId(embeddingOptions.getModel()).build(); |
| 132 | + case "on-demand" -> OnDemandServingMode.builder().modelId(embeddingOptions.getModel()).build(); |
| 133 | + default -> throw new IllegalArgumentException( |
| 134 | + "unknown serving mode for OCI embedding model: " + embeddingOptions.getServingMode()); |
| 135 | + }; |
| 136 | + } |
| 137 | + |
| 138 | + private List<EmbedTextRequest> createRequests(List<String> inputs, OCIEmbeddingOptions embeddingOptions) { |
| 139 | + int size = inputs.size(); |
| 140 | + List<EmbedTextRequest> requests = new ArrayList<>(); |
| 141 | + for (int i = 0; i < inputs.size(); i += EMBEDTEXT_BATCH_SIZE) { |
| 142 | + List<String> batch = inputs.subList(i, Math.min(i + EMBEDTEXT_BATCH_SIZE, size)); |
| 143 | + requests.add(createRequest(batch, embeddingOptions)); |
| 144 | + } |
| 145 | + return requests; |
| 146 | + } |
| 147 | + |
| 148 | + private EmbedTextRequest createRequest(List<String> inputs, OCIEmbeddingOptions embeddingOptions) { |
| 149 | + EmbedTextDetails embedTextDetails = EmbedTextDetails.builder() |
| 150 | + .servingMode(servingMode(embeddingOptions)) |
| 151 | + .compartmentId(embeddingOptions.getCompartment()) |
| 152 | + .inputs(inputs) |
| 153 | + .truncate(Objects.requireNonNullElse(embeddingOptions.getTruncate(), EmbedTextDetails.Truncate.End)) |
| 154 | + .build(); |
| 155 | + return EmbedTextRequest.builder().embedTextDetails(embedTextDetails).build(); |
| 156 | + } |
| 157 | + |
| 158 | + private OCIEmbeddingOptions mergeOptions(EmbeddingOptions embeddingOptions, OCIEmbeddingOptions defaultOptions) { |
| 159 | + if (embeddingOptions instanceof OCIEmbeddingOptions) { |
| 160 | + OCIEmbeddingOptions dynamicOptions = ModelOptionsUtils.merge(embeddingOptions, defaultOptions, |
| 161 | + OCIEmbeddingOptions.class); |
| 162 | + if (dynamicOptions != null) { |
| 163 | + return dynamicOptions; |
| 164 | + } |
| 165 | + } |
| 166 | + return defaultOptions; |
| 167 | + } |
| 168 | + |
| 169 | + private float[] toFloats(List<Float> embedding) { |
| 170 | + float[] floats = new float[embedding.size()]; |
| 171 | + for (int i = 0; i < embedding.size(); i++) { |
| 172 | + floats[i] = embedding.get(i); |
| 173 | + } |
| 174 | + return floats; |
| 175 | + } |
| 176 | + |
| 177 | +} |
0 commit comments