Fixing and improving...

eclipse/predictive-powers/pom.xml

@@ -18,7 +18,7 @@
 		<!-- Library versions -->
 		<jackson.version>2.15.3</jackson.version>
 		<jackson.jsonschema.version>1.0.39</jackson.jsonschema.version>
-		<lombok.version>1.18.26</lombok.version>
+		<lombok.version>1.18.32</lombok.version>
 		<retrofit.version>2.9.0</retrofit.version>
 		<tika.version>2.9.0</tika.version>
 		<djl.tokenizers.version>0.26.0</djl.tokenizers.version>

eclipse/predictive-powers/src/main/java/io/github/mzattera/predictivepowers/ApiClient.java

@@ -74,7 +74,7 @@ public static OkHttpClient getDefaultHttpClient(int readTimeoutMillis, int maxRe
 		if (readTimeoutMillis < 0)
 			throw new IllegalArgumentException("Read time out must be >= 0");
 		if (keepAliveDurationMillis < 0)
-			throw new IllegalArgumentException("Keep alive druaation out be >= 0");
+			throw new IllegalArgumentException("Keep alive duration must be >= 0");
 		if (maxIdleConnectionsMillis < 0)
 			throw new IllegalArgumentException("Max idel connections must must be >= 0");
 
@@ -99,9 +99,11 @@ public Response intercept(Chain chain) throws IOException {
 //					System.out.println("SERVER BUSY check: " + (retries < maxRetries) + " timeOut:" + readTimeoutMillis
 //							+ " elapsed: " + elapsed + " " + (readTimeoutMillis <= elapsed));
 
+				// TODO URGENT check list of OpenAI, Anthropic and Hugging face errors and let client customize accordingly.
+
 				while ((retries < maxRetries) && ((readTimeoutMillis == 0) || (elapsed < readTimeoutMillis))
 						&& ((response.code() == 429) || (response.code() == 500) || (response.code() == 503)
-								|| (response.code() == 504))) {
+								|| (response.code() == 504)|| (response.code() == 529))) {
 
 					// Waits and retries as server is temporarily unavailable
 

eclipse/predictive-powers/src/main/java/io/github/mzattera/predictivepowers/anthropic/services/AnthropicChatService.java

@@ -207,11 +207,7 @@ public void setMaxNewTokens(Integer maxNewTokens) {
 		defaultReq.setMaxTokens(maxNewTokens);
 	}
 
-	/**
-	 * This method counts number of tokens that are consumed at each request to
-	 * provide bot instructions (personality). This is the minimum size each request
-	 * to the API will take.
-	 */
+	@Override
 	public int getBaseTokens() {
 		// Instructions tokens
 		String cmd = getPersonality();

eclipse/predictive-powers/src/main/java/io/github/mzattera/predictivepowers/anthropic/services/AnthropicModelService.java

@@ -208,6 +208,8 @@ public ToolCallResult invoke(@NonNull ToolCall call) throws Exception {
 		}
 	}
 
+	// TODO URGENT Below code is to test the tokenizer; remove it.
+
 	// Capability providing the functions to the agent
 	private final static Capability DEFAULT_CAPABILITY = new Toolset();
 	static {

eclipse/predictive-powers/src/main/java/io/github/mzattera/predictivepowers/huggingface/services/HuggingFaceChatService.java

@@ -138,6 +138,11 @@ public void setMaxNewTokens(Integer maxNewTokens) {
 		defaultReq.getParameters().setMaxLength(maxNewTokens);
 	}
 
+	@Override
+	public int getBaseTokens() {
+		return 0;
+	}
+
 	@Override
 	public ChatCompletion chat(String msg) {
 		return chat(msg, defaultReq);

eclipse/predictive-powers/src/main/java/io/github/mzattera/predictivepowers/openai/services/OpenAiChatService.java

@@ -217,16 +217,7 @@ public void setMaxNewTokens(Integer maxNewTokens) {
 		defaultReq.setMaxTokens(maxNewTokens);
 	}
 
-	/**
-	 * This method counts number of tokens that are consumed at each request to
-	 * provide bot instructions (personality) and list tools it can use. This is the
-	 * minimum size each request to the API will take. In addition, each request
-	 * will consume the tokens needed to encode messages, which include tool calls
-	 * and their corresponding replies.
-	 * 
-	 * @return Number of tokens in the request including bot personality and tools
-	 *         (functions) descriptions, but excluding any other message.
-	 */
+	@Override
 	public int getBaseTokens() {
 		List<OpenAiChatMessage> old = defaultReq.getMessages();
 

eclipse/predictive-powers/src/main/java/io/github/mzattera/predictivepowers/services/ChatService.java

@@ -148,6 +148,14 @@ public interface ChatService extends AiService {
 	 */
 	void setMaxNewTokens(Integer maxNewTokens);
 
+	/**
+	 * This method returns number of tokens that are consumed at each request to
+	 * provide directions (e.g. personality) to this service. This is the minimum
+	 * token size of each request to the API, in addition to token consumed by
+	 * messages.
+	 */
+	public int getBaseTokens();
+
 	/**
 	 * Starts a new chat, clearing current conversation.
 	 */

eclipse/predictive-powers/src/main/java/io/github/mzattera/util/IdUtil.java

@@ -0,0 +1,26 @@
+/**
+ * 
+ */
+package io.github.mzattera.util;
+
+import java.util.concurrent.atomic.AtomicLong;
+
+/**
+ * Generates unique long IDs.
+ * IDs are guaranteed to be unique inside the JVM.ight not be unique over subsequent runs.
+ */
+public final class IdUtil {
+
+	private static final AtomicLong counter = new AtomicLong(System.nanoTime());
+
+	private IdUtil() {
+	}
+
+	public static long getLongId() {
+		return counter.getAndIncrement();
+	}
+
+	public static String getStringId() {
+		return Long.toString(getLongId());
+	}
+}