feat: Update model profile and statistics tracking

- Updated model profile to support new cost parameters: costPerCachedInToken and costPerReasoningOutToken.
- Enhanced AgentRequestStatistic to track cached input tokens and reasoning output tokens.
- Modified CompletionStatistics to include cached input tokens and reasoning output tokens.
- Updated AzureAiClient, CachedClient, and other clients to handle new statistics fields.
- Introduced Timings class to capture detailed timing metrics for requests.
- Adjusted Usage class to include input and output token details.
- Updated properties files to include new cost parameters for cached and reasoning tokens.

src/main/java/de/extio/lmlib/client/CompletionStatistics.java

@@ -3,6 +3,6 @@ package de.extio.lmlib.client;
 import java.math.BigDecimal;
 import java.time.Duration;
 
-public record CompletionStatistics(int requests, Duration duration, long inTokens, long outTokens, BigDecimal cost, boolean cached) {
+public record CompletionStatistics(int requests, Duration duration, long inTokens, long cachedInTokens, long outTokens, long reasoningOutTokens, BigDecimal cost, boolean cached) {
 
 }

src/main/java/de/extio/lmlib/client/azureai/AzureAiClient.java

@@ -290,7 +290,9 @@ public final class AzureAiClient implements Client {
 				1,
 				Duration.between(start, LocalDateTime.now()),
 				promptTokens,
+				0,
 				completionTokens,
+				0,
 				new BigDecimal(promptTokens).multiply(modelProfile.costPerInToken()).add(new BigDecimal(completionTokens).multiply(modelProfile.costPerOutToken())),
 				false);
 		LOGGER.debug("{}", statistics);

src/main/java/de/extio/lmlib/client/cached/CachedClient.java

@@ -78,12 +78,12 @@ public class CachedClient implements Client {
 		
 		final var cachedCompletion = this.cacheRepository.get(key);
 		if (cachedCompletion != null) {
-			return new Completion(cachedCompletion.response(), cachedCompletion.reasoning(), CompletionFinishReason.DONE, new CompletionStatistics(0, Duration.ZERO, cachedCompletion.inTokens(), cachedCompletion.outTokens(), BigDecimal.ZERO, true));
+			return new Completion(cachedCompletion.response(), cachedCompletion.reasoning(), CompletionFinishReason.DONE, new CompletionStatistics(0, Duration.ZERO, cachedCompletion.inTokens(), cachedCompletion.cachedInTokens(), cachedCompletion.outTokens(), cachedCompletion.reasoningOutTokens(), BigDecimal.ZERO, true));
 		}
 		
 		final var completion = supplier.get();
 		if (completion != null) {
-			this.cacheRepository.put(key, new CachedCompletion(completion.response(), completion.reasoning(), completion.finishReason(), completion.statistics().inTokens(), completion.statistics().outTokens(), OffsetDateTime.now()));
+			this.cacheRepository.put(key, new CachedCompletion(completion.response(), completion.reasoning(), completion.finishReason(), completion.statistics().inTokens(), completion.statistics().cachedInTokens(), completion.statistics().outTokens(), completion.statistics().reasoningOutTokens(), OffsetDateTime.now()));
 		}
 		return completion;
 	}

src/main/java/de/extio/lmlib/client/cached/CachedCompletion.java

@@ -4,6 +4,6 @@ import java.time.OffsetDateTime;
 
 import de.extio.lmlib.client.CompletionFinishReason;
 
-public record CachedCompletion(String response, String reasoning, CompletionFinishReason finishReason, long inTokens, long outTokens, OffsetDateTime created) {
+public record CachedCompletion(String response, String reasoning, CompletionFinishReason finishReason, long inTokens, long cachedInTokens, long outTokens, long reasoningOutTokens, OffsetDateTime created) {
 	
 }

src/main/java/de/extio/lmlib/client/oai/completion/AbstractCompletionClient.java

@@ -140,32 +140,64 @@ public abstract class AbstractCompletionClient implements Client, DisposableBean
 		}
 	}
 	
-	protected de.extio.lmlib.client.CompletionStatistics createCompletionStatistics(final ModelProfile modelProfile, final LocalDateTime start, final Usage usage, final String prompt, final String response) {
+	protected de.extio.lmlib.client.CompletionStatistics createCompletionStatistics(final ModelProfile modelProfile, final LocalDateTime start, final Usage usage, final Timings timings, final String prompt, final String response, final String reasoning) {
 		int inTokens = 0;
+		int cachedInTokens = 0;
 		int outTokens = 0;
+		int reasoningOutTokens = 0;
+		
+		// Usage data has priority (official OAI spec)
 		if (usage != null && usage.getPromptTokens() > 0) {
-			inTokens = usage.getPromptTokens();
-			outTokens = usage.getCompletionTokens();
+			cachedInTokens = usage.getCachedTokens();
+			inTokens = usage.getPromptTokens() - cachedInTokens;
+			reasoningOutTokens = usage.getReasoningTokens();
+			outTokens = usage.getCompletionTokens() - reasoningOutTokens;
 		}
-		else {
-			if (prompt != null) {
-				inTokens = this.tokenizer.count(prompt, modelProfile);
+		// If no usage data, try to get token counts from timings (e.g. llama.cpp, prompt_n does not include cache_n)
+		if (timings != null) {
+			if (inTokens <= 0 && cachedInTokens <= 0 && (timings.getPromptN() > 0 || timings.getCacheN() > 0)) {
+				inTokens = Math.max(0, timings.getPromptN());
+				cachedInTokens = Math.max(0, timings.getCacheN());
 			}
-			if (response != null) {
-				outTokens = this.tokenizer.count(response, modelProfile);
+			else if (cachedInTokens <= 0 && timings.getCacheN() > 0) {
+				cachedInTokens = Math.max(0, timings.getCacheN());
+				inTokens -= cachedInTokens;
+			}
+			if (outTokens <= 0 && timings.getPredictedN() > 0) {
+				outTokens = timings.getPredictedN();
 			}
 		}
+		// Fallback to tokenizer estimation
+		if (inTokens <= 0 && prompt != null && !prompt.isBlank()) {
+			inTokens = this.tokenizer.count(prompt, modelProfile);
+		}
+		if (reasoningOutTokens <= 0 && reasoning != null && !reasoning.isBlank()) {
+			reasoningOutTokens = this.tokenizer.count(reasoning, modelProfile);
+			outTokens -= reasoningOutTokens;
+		}
+		if (outTokens <= 0 && response != null && !response.isBlank()) {
+			outTokens = this.tokenizer.count(response, modelProfile);
+		}
+		
+		// Calculate cost - inTokens and outTokens are already exclusive of cached/reasoning tokens
+		BigDecimal cost = BigDecimal.ZERO;
+		cost = cost.add(new BigDecimal(inTokens).multiply(modelProfile.costPerInToken()));
+		cost = cost.add(new BigDecimal(cachedInTokens).multiply(modelProfile.costPerCachedInToken()));
+		cost = cost.add(new BigDecimal(outTokens).multiply(modelProfile.costPerOutToken()));
+		cost = cost.add(new BigDecimal(reasoningOutTokens).multiply(modelProfile.costPerReasoningOutToken()));
 		
 		final var statistics = new de.extio.lmlib.client.CompletionStatistics(
 				1,
 				Duration.between(start, LocalDateTime.now()),
 				inTokens,
+				cachedInTokens,
 				outTokens,
-				new BigDecimal(inTokens).multiply(modelProfile.costPerInToken()).add(new BigDecimal(outTokens).multiply(modelProfile.costPerOutToken())),
+				reasoningOutTokens,
+				cost,
 				false);
 		LOGGER.debug("{}", statistics);
 		if (this.collectStatistics) {
-			this.totalStatistics.add(statistics.duration(), inTokens, outTokens, statistics.cost());
+			this.totalStatistics.add(statistics.duration(), inTokens, cachedInTokens, outTokens, reasoningOutTokens, statistics.cost());
 		}
 		return statistics;
 	}
@@ -187,15 +219,21 @@ public abstract class AbstractCompletionClient implements Client, DisposableBean
 		
 		private long inTokens;
 		
+		private long cachedInTokens;
+		
 		private long outTokens;
 		
+		private long reasoningOutTokens;
+		
 		private BigDecimal cost = BigDecimal.ZERO;
 		
-		synchronized void add(final Duration duration, final int inTokens, final int outTokens, final BigDecimal cost) {
+		synchronized void add(final Duration duration, final int inTokens, final int cachedInTokens, final int outTokens, final int reasoningOutTokens, final BigDecimal cost) {
 			this.requests++;
 			this.duration = this.duration.plus(duration);
 			this.inTokens += inTokens;
+			this.cachedInTokens += cachedInTokens;
 			this.outTokens += outTokens;
+			this.reasoningOutTokens += reasoningOutTokens;
 			this.cost = this.cost.add(cost);
 		}
 		
@@ -208,19 +246,24 @@ public abstract class AbstractCompletionClient implements Client, DisposableBean
 			builder.append(this.duration);
 			builder.append(", inTokens=");
 			builder.append(this.inTokens);
+			builder.append(", cachedInTokens=");
+			builder.append(this.cachedInTokens);
 			builder.append(", outTokens=");
 			builder.append(this.outTokens);
+			builder.append(", reasoningOutTokens=");
+			builder.append(this.reasoningOutTokens);
 			builder.append(", averageTps=");
-			builder.append(new DecimalFormat("0.##").format(((double) (this.outTokens + this.inTokens) / (double) this.requests) / ((double) this.duration.toMillis() / 1000.0 / (double) this.requests)));
+			final long totalOutTokens = this.outTokens + this.reasoningOutTokens;
+			builder.append(new DecimalFormat("0.##").format(((double) (totalOutTokens + inTokens) / (double) this.requests) / ((double) this.duration.toMillis() / 1000.0 / (double) this.requests)));
 			builder.append(", averageOutTps=");
-			builder.append(new DecimalFormat("0.##").format((double) this.outTokens / ((double) this.duration.toMillis() / 1000.0)));
+			builder.append(new DecimalFormat("0.##").format((double) totalOutTokens / ((double) this.duration.toMillis() / 1000.0)));
 			builder.append(", effectiveDuration=");
 			final var effectiveDuration = Duration.between(this.start, Instant.now());
 			builder.append(effectiveDuration);
 			builder.append(", effectiveTps=");
-			builder.append(new DecimalFormat("0.##").format(((double) (this.outTokens + this.inTokens) / (double) this.requests) / ((double) effectiveDuration.toMillis() / 1000.0 / (double) this.requests)));
+			builder.append(new DecimalFormat("0.##").format(((double) (totalOutTokens + inTokens) / (double) this.requests) / ((double) effectiveDuration.toMillis() / 1000.0 / (double) this.requests)));
 			builder.append(", effectiveOutTps=");
-			builder.append(new DecimalFormat("0.##").format((double) this.outTokens / ((double) effectiveDuration.toMillis() / 1000.0)));
+			builder.append(new DecimalFormat("0.##").format((double) totalOutTokens / ((double) effectiveDuration.toMillis() / 1000.0)));
 			builder.append(", cost=");
 			builder.append(new DecimalFormat("0.###").format(cost.doubleValue()));
 			builder.append("]");

src/main/java/de/extio/lmlib/client/oai/completion/AbstractCompletionRequest.java

@@ -16,6 +16,8 @@ public abstract class AbstractCompletionRequest {
 	
 	protected StreamOptions streamOptions;
 	
+	protected Boolean usage;
+	
 	@JsonProperty("model")
 	public String getModel() {
 		return this.model;
@@ -69,4 +71,13 @@ public abstract class AbstractCompletionRequest {
 	public void setStreamOptions(final StreamOptions streamOptions) {
 		this.streamOptions = streamOptions;
 	}
+	
+	@JsonProperty("usage")
+	public Boolean getUsage() {
+		return this.usage;
+	}
+	
+	public void setUsage(final Boolean usage) {
+		this.usage = usage;
+	}
 }

src/main/java/de/extio/lmlib/client/oai/completion/AbstractCompletionResponse.java

@@ -11,6 +11,8 @@ public abstract class AbstractCompletionResponse {
 	protected Integer created;
 	
 	protected Usage usage;
+	
+	protected Timings timings;
 
 	@JsonProperty("id")
 	public String getId() {
@@ -41,10 +43,19 @@ public abstract class AbstractCompletionResponse {
 	
 	@JsonProperty("usage")
 	public Usage getUsage() {
-		return usage;
+		return this.usage;
 	}
 
 	public void setUsage(final Usage usage) {
 		this.usage = usage;
 	}
+	
+	@JsonProperty("timings")
+	public Timings getTimings() {
+		return this.timings;
+	}
+	
+	public void setTimings(final Timings timings) {
+		this.timings = timings;
+	}
 }

src/main/java/de/extio/lmlib/client/oai/completion/Timings.java

@@ -0,0 +1,111 @@
+package de.extio.lmlib.client.oai.completion;
+
+import com.fasterxml.jackson.annotation.JsonProperty;
+
+public final class Timings {
+    
+    private int cacheN;
+    
+    private int predictedN;
+    
+    private double predictedMs;
+    
+    private double predictedPerTokenMs;
+    
+    private double predictedPerSecond;
+    
+    private int promptN;
+    
+    private double promptMs;
+    
+    private double promptPerTokenMs;
+    
+    private double promptPerSecond;
+    
+    @JsonProperty("cache_n")
+    public int getCacheN() {
+        return this.cacheN;
+    }
+    
+    public void setCacheN(final int cacheN) {
+        this.cacheN = cacheN;
+    }
+    
+    @JsonProperty("predicted_n")
+    public int getPredictedN() {
+        return this.predictedN;
+    }
+    
+    public void setPredictedN(final int predictedN) {
+        this.predictedN = predictedN;
+    }
+    
+    @JsonProperty("predicted_ms")
+    public double getPredictedMs() {
+        return this.predictedMs;
+    }
+    
+    public void setPredictedMs(final double predictedMs) {
+        this.predictedMs = predictedMs;
+    }
+    
+    @JsonProperty("predicted_per_token_ms")
+    public double getPredictedPerTokenMs() {
+        return this.predictedPerTokenMs;
+    }
+    
+    public void setPredictedPerTokenMs(final double predictedPerTokenMs) {
+        this.predictedPerTokenMs = predictedPerTokenMs;
+    }
+    
+    @JsonProperty("predicted_per_second")
+    public double getPredictedPerSecond() {
+        return this.predictedPerSecond;
+    }
+    
+    public void setPredictedPerSecond(final double predictedPerSecond) {
+        this.predictedPerSecond = predictedPerSecond;
+    }
+    
+    @JsonProperty("prompt_n")
+    public int getPromptN() {
+        return this.promptN;
+    }
+    
+    public void setPromptN(final int promptN) {
+        this.promptN = promptN;
+    }
+    
+    @JsonProperty("prompt_ms")
+    public double getPromptMs() {
+        return this.promptMs;
+    }
+    
+    public void setPromptMs(final double promptMs) {
+        this.promptMs = promptMs;
+    }
+    
+    @JsonProperty("prompt_per_token_ms")
+    public double getPromptPerTokenMs() {
+        return this.promptPerTokenMs;
+    }
+    
+    public void setPromptPerTokenMs(final double promptPerTokenMs) {
+        this.promptPerTokenMs = promptPerTokenMs;
+    }
+    
+    @JsonProperty("prompt_per_second")
+    public double getPromptPerSecond() {
+        return this.promptPerSecond;
+    }
+    
+    public void setPromptPerSecond(final double promptPerSecond) {
+        this.promptPerSecond = promptPerSecond;
+    }
+    
+    @Override
+    public String toString() {
+        return "Timings [cacheN=" + this.cacheN + ", predictedN=" + this.predictedN + ", predictedMs=" + this.predictedMs + ", predictedPerTokenMs=" + this.predictedPerTokenMs + ", predictedPerSecond=" + this.predictedPerSecond + ", promptN=" + this.promptN + ", promptMs=" + this.promptMs + ", promptPerTokenMs=" + this.promptPerTokenMs + ", promptPerSecond=" + this.promptPerSecond + "]";
+    }
+    
+}

src/main/java/de/extio/lmlib/client/oai/completion/Usage.java

@@ -10,36 +10,114 @@ public final class Usage {
     
     private int totalTokens;
     
+    private InputTokensDetails inputTokensDetails;
+    
+    private OutputTokensDetails outputTokensDetails;
+    
     @JsonProperty("completion_tokens")
     public int getCompletionTokens() {
-        return completionTokens;
+        return this.completionTokens;
     }
     
     public void setCompletionTokens(final int completionTokens) {
         this.completionTokens = completionTokens;
     }
     
+    @JsonProperty("output_tokens")
+    public void setOutputTokens(final int outputTokens) {
+        this.completionTokens = outputTokens;
+    }
+    
     @JsonProperty("prompt_tokens")
     public int getPromptTokens() {
-        return promptTokens;
+        return this.promptTokens;
     }
     
     public void setPromptTokens(final int promptTokens) {
         this.promptTokens = promptTokens;
     }
     
+    @JsonProperty("input_tokens")
+    public void setInputTokens(final int inputTokens) {
+        this.promptTokens = inputTokens;
+    }
+    
     @JsonProperty("total_tokens")
     public int getTotalTokens() {
-        return totalTokens;
+        return this.totalTokens;
     }
     
     public void setTotalTokens(final int totalTokens) {
         this.totalTokens = totalTokens;
     }
     
+    @JsonProperty("input_tokens_details")
+    public InputTokensDetails getInputTokensDetails() {
+        return this.inputTokensDetails;
+    }
+    
+    public void setInputTokensDetails(final InputTokensDetails inputTokensDetails) {
+        this.inputTokensDetails = inputTokensDetails;
+    }
+    
+    @JsonProperty("output_tokens_details")
+    public OutputTokensDetails getOutputTokensDetails() {
+        return this.outputTokensDetails;
+    }
+    
+    public void setOutputTokensDetails(final OutputTokensDetails outputTokensDetails) {
+        this.outputTokensDetails = outputTokensDetails;
+    }
+    
+    public int getCachedTokens() {
+        return this.inputTokensDetails != null ? this.inputTokensDetails.getCachedTokens() : 0;
+    }
+    
+    public int getReasoningTokens() {
+        return this.outputTokensDetails != null ? this.outputTokensDetails.getReasoningTokens() : 0;
+    }
+    
     @Override
     public String toString() {
-        return "Usage [completionTokens=" + completionTokens + ", promptTokens=" + promptTokens + ", totalTokens=" + totalTokens + "]";
+        return "Usage [completionTokens=" + this.completionTokens + ", promptTokens=" + this.promptTokens + ", totalTokens=" + this.totalTokens + ", inputTokensDetails=" + this.inputTokensDetails + ", outputTokensDetails=" + this.outputTokensDetails + "]";
+    }
+    
+    public static final class InputTokensDetails {
+        
+        private int cachedTokens;
+        
+        @JsonProperty("cached_tokens")
+        public int getCachedTokens() {
+            return this.cachedTokens;
+        }
+        
+        public void setCachedTokens(final int cachedTokens) {
+            this.cachedTokens = cachedTokens;
+        }
+        
+        @Override
+        public String toString() {
+            return "InputTokensDetails [cachedTokens=" + this.cachedTokens + "]";
+        }
+    }
+    
+    public static final class OutputTokensDetails {
+        
+        private int reasoningTokens;
+        
+        @JsonProperty("reasoning_tokens")
+        public int getReasoningTokens() {
+            return this.reasoningTokens;
+        }
+        
+        public void setReasoningTokens(final int reasoningTokens) {
+            this.reasoningTokens = reasoningTokens;
+        }
+        
+        @Override
+        public String toString() {
+            return "OutputTokensDetails [reasoningTokens=" + this.reasoningTokens + "]";
+        }
     }
     
 }

src/main/java/de/extio/lmlib/client/oai/completion/chat/ChatCompletionClient.java

@@ -52,6 +52,8 @@ public class ChatCompletionClient extends AbstractCompletionClient {
 		request.setStream(chunkConsumer != null);
 		if (chunkConsumer != null) {
 			this.configureStreamOptions(true, request::setStreamOptions);
+		} else {
+			request.setUsage(true);
 		}
 		if (modelProfile.reasoningEffort() != null && !modelProfile.reasoningEffort().isBlank()) {
 			request.setReasoning(new ChatCompletionRequest.ChatCompletionsRequestReasoning());
@@ -86,10 +88,14 @@ public class ChatCompletionClient extends AbstractCompletionClient {
 		final LocalDateTime start = LocalDateTime.now();
 		final var webClient = this.webClientBuilder.baseUrl(modelProfile.url()).build();
 		
-		final var response = webClient
+		var requestSpec = webClient
 				.method(HttpMethod.POST)
 				.uri(uriBuilder -> uriBuilder.path("/v1/chat/completions").build())
-				.header("Content-Type", "application/json")
+				.header("Content-Type", "application/json");
+		if (modelProfile.apiKey() != null && !modelProfile.apiKey().isBlank()) {
+			requestSpec = requestSpec.header("Authorization", "Bearer " + modelProfile.apiKey());
+		}
+		final var response = requestSpec
 				.bodyValue(requestBody)
 				.exchangeToMono(clientResponse -> {
 					if (clientResponse.statusCode().isError()) {
@@ -123,6 +129,9 @@ public class ChatCompletionClient extends AbstractCompletionClient {
 											if (streamResponse.getUsage() != null) {
 												chatCompletionResponse.setUsage(streamResponse.getUsage());
 											}
+											if (streamResponse.getTimings() != null) {
+												chatCompletionResponse.setTimings(streamResponse.getTimings());
+											}
 											if (streamResponse.getChoices() != null && !streamResponse.getChoices().isEmpty()) {
 												if (streamResponse.getChoices().getFirst().getFinishReason() != null) {
 													chatCompletionResponse.getChoices().getFirst().setFinishReason(streamResponse.getChoices().getFirst().getFinishReason());
@@ -165,7 +174,7 @@ public class ChatCompletionClient extends AbstractCompletionClient {
 		final var content = response.getChoices().getFirst().getMessage().getContent();
 		final var reasoning = response.getChoices().getFirst().getMessage().getReasoningContent();
 		final var finishReason = this.mapFinishReason(choice.getFinishReason());
-		final var statistics = createCompletionStatistics(modelProfile, start, response.getUsage(), null, content);
+		final var statistics = createCompletionStatistics(modelProfile, start, response.getUsage(), response.getTimings(), null, content, reasoning);
 		return new Completion(content, reasoning, finishReason, statistics);
 	}
 	

src/main/java/de/extio/lmlib/client/oai/completion/text/TextCompletionClient.java

@@ -57,6 +57,8 @@ public class TextCompletionClient extends AbstractCompletionClient {
 		request.setStream(chunkConsumer != null);
 		if (chunkConsumer != null) {
 			this.configureStreamOptions(true, request::setStreamOptions);
+		} else {
+			request.setUsage(true);
 		}
 		
 		String requestBody;
@@ -70,10 +72,14 @@ public class TextCompletionClient extends AbstractCompletionClient {
 		LOGGER.debug("Requesting completion at {}", modelProfile.url());
 		final LocalDateTime start = LocalDateTime.now();
 		final var webClient = this.webClientBuilder.baseUrl(modelProfile.url()).build();
-		final var response = webClient
+		var requestSpec = webClient
 				.method(HttpMethod.POST)
 				.uri(uriBuilder -> uriBuilder.path("/v1/completions").build())
-				.header("Content-Type", "application/json")
+				.header("Content-Type", "application/json");
+		if (modelProfile.apiKey() != null && !modelProfile.apiKey().isBlank()) {
+			requestSpec = requestSpec.header("Authorization", "Bearer " + modelProfile.apiKey());
+		}
+		final var response = requestSpec
 				.bodyValue(requestBody)
 				.exchangeToMono(clientResponse -> {
 					if (clientResponse.statusCode().isError()) {
@@ -112,6 +118,9 @@ public class TextCompletionClient extends AbstractCompletionClient {
 											if (streamResponse.getUsage() != null) {
 												CompletionResponse.setUsage(streamResponse.getUsage());
 											}
+											if (streamResponse.getTimings() != null) {
+												CompletionResponse.setTimings(streamResponse.getTimings());
+											}
 											if (streamResponse.getChoices() != null && !streamResponse.getChoices().isEmpty()) {
 												if (streamResponse.getChoices().getFirst().getFinishReason() != null) {
 													CompletionResponse.getChoices().getFirst().setFinishReason(streamResponse.getChoices().getFirst().getFinishReason());
@@ -172,7 +181,7 @@ public class TextCompletionClient extends AbstractCompletionClient {
 			}
 		}
 		
-		final var statistics = createCompletionStatistics(modelProfile, start, response.getUsage(), prompt, response.getContent());
+		final var statistics = createCompletionStatistics(modelProfile, start, response.getUsage(), response.getTimings(), prompt, content, reasoning);
 		return new Completion(content, reasoning, finishReason, statistics);
 	}