Merge branch 'feat/context-api-ga' into 'master'

feat(ark-runtime): add context chat api supports

See merge request iaasng/volcengine-java-sdk!296

Author: liuweiwei-xyz

volcengine-java-sdk-ark-runtime/src/main/java/com/volcengine/ark/runtime/Const.java

@@ -15,4 +15,9 @@ public class Const {
 
     public static final String RESOURCE_TYPE_BOT = "bot";
     public static final String RESOURCE_TYPE_ENDPOINT = "endpoint";
+
+    public static final String CONTEXT_MODE_SESSION = "session";
+    public static final String CONTEXT_MODE_COMMON_PREFIX = "common_prefix";
+    public static final String TRUNCATION_STRATEGY_TYPE_LAST_HISTORY_TOKENS = "last_history_tokens";
+    public static final String TRUNCATION_STRATEGY_TYPE_ROLLING_TOKENS = "rolling_tokens";
 }

volcengine-java-sdk-ark-runtime/src/main/java/com/volcengine/ark/runtime/exception/ArkHttpException.java

@@ -3,6 +3,7 @@
 import com.google.gson.Gson;
 
 public class ArkHttpException extends RuntimeException {
+    public static Integer INTERNAL_SERVICE_CODE = 500;
 
     public final int statusCode;
 
@@ -31,6 +32,7 @@ public String getMessage() {
     public String toString() {
         return "ArkHttpException{" +
                 "statusCode=" + statusCode +
+                ", message='" + super.getMessage() + '\'' +
                 ", code='" + code + '\'' +
                 ", param='" + param + '\'' +
                 ", type='" + type + '\'' +

volcengine-java-sdk-ark-runtime/src/main/java/com/volcengine/ark/runtime/interceptor/RequestIdInterceptor.java

@@ -1,6 +1,9 @@
 package com.volcengine.ark.runtime.interceptor;
 
 import com.volcengine.ark.runtime.Const;
+import com.volcengine.ark.runtime.exception.ArkAPIError;
+import com.volcengine.ark.runtime.exception.ArkException;
+import com.volcengine.ark.runtime.exception.ArkHttpException;
 import com.volcengine.version.Version;
 import okhttp3.Interceptor;
 import okhttp3.Request;
@@ -26,7 +29,14 @@ public Response intercept(Chain chain) throws IOException {
         requestBuilder.header("User-Agent", getUserAgent());
 
         Request request = requestBuilder.build();
-        return chain.proceed(request);
+
+        try {
+            return chain.proceed(request);
+        } catch (Exception e) {
+            String requestId = request.header(Const.CLIENT_REQUEST_HEADER);
+            ArkAPIError arkAPIError = new ArkAPIError(new ArkAPIError.ArkErrorDetails(e.getMessage(), "", "", ""));
+            throw new ArkHttpException(arkAPIError, e, ArkHttpException.INTERNAL_SERVICE_CODE, requestId);
+        }
     }
 
     private String genRequestId() {

volcengine-java-sdk-ark-runtime/src/main/java/com/volcengine/ark/runtime/interceptor/RetryInterceptor.java

@@ -21,18 +21,31 @@ public RetryInterceptor(int retryTimes) {
     }
 
     @Override
-    public Response intercept(Chain chain) throws IOException {
+    public Response intercept(Chain chain) throws RuntimeException, InterruptedIOException {
         Request request = chain.request();
 
-        // try the request
-        Response response = chain.proceed(request);
-
+        Response response = null;
         int tryCount = 0;
-        while ((response.code() >= 500 || response.code() == 429) && tryCount < retryTimes) {
-            tryCount++;
+        boolean shouldRetry;
+        Exception exception;
+        do {
+            if (response != null) {
+                response.close();
+            }
+            exception = null;
+
+            try {
+                response = chain.proceed(request);
+                shouldRetry = response.code() >= 500 || response.code() == 429;
+            } catch (Exception e) {
+                shouldRetry = true;
+                exception = e;
+            }
 
-            // retry the request
-            response.close();
+            tryCount++;
+            if (!(shouldRetry && tryCount <= retryTimes)) {
+                break;
+            }
 
             try {
                 double interval = retryInterval(retryTimes, retryTimes - tryCount) * 1000;
@@ -41,10 +54,12 @@ public Response intercept(Chain chain) throws IOException {
                 Thread.currentThread().interrupt();
                 throw new InterruptedIOException();
             }
-            response = chain.proceed(request);
-        }
+        } while (true);
 
-        return response;
+        if (response != null) {
+            return response;
+        }
+        throw new RuntimeException(exception);
     }
 
     public double retryInterval(int max, int remain) {

volcengine-java-sdk-ark-runtime/src/main/java/com/volcengine/ark/runtime/model/PromptTokensDetails.java

@@ -0,0 +1,26 @@
+package com.volcengine.ark.runtime.model;
+
+import com.fasterxml.jackson.annotation.JsonIgnoreProperties;
+import com.fasterxml.jackson.annotation.JsonProperty;
+
+@JsonIgnoreProperties(ignoreUnknown = true)
+public class PromptTokensDetails {
+
+    @JsonProperty("cached_tokens")
+    private Integer cachedTokens;
+
+    public Integer getCachedTokens() {
+        return cachedTokens;
+    }
+
+    public void setCachedTokens(Integer cachedTokens) {
+        this.cachedTokens = cachedTokens;
+    }
+
+    @Override
+    public String toString() {
+        return "PromptTokensDetails{" +
+                "cachedTokens=" + cachedTokens +
+                '}';
+    }
+}

volcengine-java-sdk-ark-runtime/src/main/java/com/volcengine/ark/runtime/model/Usage.java

@@ -23,12 +23,22 @@ public class Usage {
     @JsonProperty("total_tokens")
     long totalTokens;
 
+    @JsonProperty("prompt_tokens_details")
+    private PromptTokensDetails promptTokensDetails;
+
     public Usage(long promptTokens, long completionTokens, long totalTokens) {
         this.promptTokens = promptTokens;
         this.completionTokens = completionTokens;
         this.totalTokens = totalTokens;
     }
 
+    public Usage(long promptTokens, long completionTokens, long totalTokens, PromptTokensDetails promptTokensDetails) {
+        this.promptTokens = promptTokens;
+        this.completionTokens = completionTokens;
+        this.totalTokens = totalTokens;
+        this.promptTokensDetails = promptTokensDetails;
+    }
+
     public Usage() {}
 
     public long getPromptTokens() {
@@ -55,12 +65,21 @@ public void setTotalTokens(long totalTokens) {
         this.totalTokens = totalTokens;
     }
 
+    public PromptTokensDetails getPromptTokensDetails() {
+        return promptTokensDetails;
+    }
+
+    public void setPromptTokensDetails(PromptTokensDetails promptTokensDetails) {
+        this.promptTokensDetails = promptTokensDetails;
+    }
+
     @Override
     public String toString() {
         return "Usage{" +
                 "promptTokens=" + promptTokens +
                 ", completionTokens=" + completionTokens +
                 ", totalTokens=" + totalTokens +
+                ", promptTokensDetails=" + promptTokensDetails +
                 '}';
     }
 }

volcengine-java-sdk-ark-runtime/src/main/java/com/volcengine/ark/runtime/model/context/CreateContextRequest.java

@@ -0,0 +1,138 @@
+package com.volcengine.ark.runtime.model.context;
+
+import com.fasterxml.jackson.annotation.JsonIgnoreProperties;
+import com.fasterxml.jackson.annotation.JsonProperty;
+import com.volcengine.ark.runtime.model.completion.chat.ChatMessage;
+
+import java.util.List;
+
+@JsonIgnoreProperties(ignoreUnknown = true)
+public class CreateContextRequest {
+
+    @JsonProperty("model")
+    private String model;
+
+    @JsonProperty("mode")
+    private String mode;
+
+    @JsonProperty("messages")
+    private List<ChatMessage> messages;
+
+    @JsonProperty("ttl")
+    private Integer ttl;
+
+    @JsonProperty("truncation_strategy")
+    private TruncationStrategy truncationStrategy;
+
+    public CreateContextRequest() {
+    }
+
+    public CreateContextRequest(String model, String mode, List<ChatMessage> messages, Integer ttl, TruncationStrategy truncationStrategy) {
+        this.model = model;
+        this.mode = mode;
+        this.messages = messages;
+        this.ttl = ttl;
+        this.truncationStrategy = truncationStrategy;
+    }
+
+    public String getModel() {
+        return model;
+    }
+
+    public void setModel(String model) {
+        this.model = model;
+    }
+
+    public String getMode() {
+        return mode;
+    }
+
+    public void setMode(String mode) {
+        this.mode = mode;
+    }
+
+    public List<ChatMessage> getMessages() {
+        return messages;
+    }
+
+    public void setMessages(List<ChatMessage> messages) {
+        this.messages = messages;
+    }
+
+    public Integer getTtl() {
+        return ttl;
+    }
+
+    public void setTtl(Integer ttl) {
+        this.ttl = ttl;
+    }
+
+    public TruncationStrategy getTruncationStrategy() {
+        return truncationStrategy;
+    }
+
+    public void setTruncationStrategy(TruncationStrategy truncationStrategy) {
+        this.truncationStrategy = truncationStrategy;
+    }
+
+    @Override
+    public String toString() {
+        return "CreateContextRequest{" +
+                "model='" + model + '\'' +
+                ", mode='" + mode + '\'' +
+                ", messages=" + messages +
+                ", ttl=" + ttl +
+                ", truncationStrategy=" + truncationStrategy +
+                '}';
+    }
+
+    public static CreateContextRequest.Builder builder() {
+        return new Builder();
+    }
+
+    public static class Builder {
+        private String model;
+        private String mode;
+        private List<ChatMessage> messages;
+        private Integer ttl;
+        private TruncationStrategy truncationStrategy;
+
+        private Builder() {
+        }
+
+        public Builder model(String model) {
+            this.model = model;
+            return this;
+        }
+
+        public Builder mode(String mode) {
+            this.mode = mode;
+            return this;
+        }
+
+        public Builder messages(List<ChatMessage> messages) {
+            this.messages = messages;
+            return this;
+        }
+
+        public Builder ttl(Integer ttl) {
+            this.ttl = ttl;
+            return this;
+        }
+
+        public Builder truncationStrategy(TruncationStrategy truncationStrategy) {
+            this.truncationStrategy = truncationStrategy;
+            return this;
+        }
+
+        public CreateContextRequest build() {
+            CreateContextRequest createContextRequest = new CreateContextRequest();
+            createContextRequest.setModel(model);
+            createContextRequest.setMode(mode);
+            createContextRequest.setMessages(messages);
+            createContextRequest.setTtl(ttl);
+            createContextRequest.setTruncationStrategy(truncationStrategy);
+            return createContextRequest;
+        }
+    }
+}