remove caching related code

Author: galkleinman

packages/ai-semantic-conventions/src/SemanticAttributes.ts

@@ -35,9 +35,6 @@ export const SpanAttributes = {
   LLM_USAGE_COMPLETION_TOKENS: "gen_ai.usage.completion_tokens",
   LLM_USAGE_INPUT_TOKENS: ATTR_GEN_AI_USAGE_INPUT_TOKENS,
   LLM_USAGE_OUTPUT_TOKENS: ATTR_GEN_AI_USAGE_OUTPUT_TOKENS,
-  LLM_USAGE_CACHE_CREATION_INPUT_TOKENS:
-    "gen_ai.usage.cache_creation_input_tokens",
-  LLM_USAGE_CACHE_READ_INPUT_TOKENS: "gen_ai.usage.cache_read_input_tokens",
 
   GEN_AI_AGENT_NAME: "gen_ai.agent.name",
 

packages/traceloop-sdk/recordings/Test-AI-SDK-Integration-with-Recording_156038438/should-capture-and-transform-cache-tokens-from-OpenAI-with-prompt-caching_4027203422/recording.har

@@ -32,9 +32,6 @@ const AI_PROMPT_MESSAGES = "ai.prompt.messages";
 const AI_PROMPT = "ai.prompt";
 const AI_USAGE_PROMPT_TOKENS = "ai.usage.promptTokens";
 const AI_USAGE_COMPLETION_TOKENS = "ai.usage.completionTokens";
-const AI_USAGE_CACHE_CREATION_INPUT_TOKENS =
-  "ai.usage.cacheCreationInputTokens";
-const AI_USAGE_CACHE_READ_INPUT_TOKENS = "ai.usage.cacheReadInputTokens";
 const AI_MODEL_PROVIDER = "ai.model.provider";
 const AI_PROMPT_TOOLS = "ai.prompt.tools";
 const AI_TELEMETRY_METADATA_PREFIX = "ai.telemetry.metadata.";
@@ -378,26 +375,6 @@ const calculateTotalTokens = (attributes: Record<string, any>): void => {
   }
 };
 
-const transformCacheCreationInputTokens = (
-  attributes: Record<string, any>,
-): void => {
-  if (AI_USAGE_CACHE_CREATION_INPUT_TOKENS in attributes) {
-    attributes[SpanAttributes.LLM_USAGE_CACHE_CREATION_INPUT_TOKENS] =
-      attributes[AI_USAGE_CACHE_CREATION_INPUT_TOKENS];
-    delete attributes[AI_USAGE_CACHE_CREATION_INPUT_TOKENS];
-  }
-};
-
-const transformCacheReadInputTokens = (
-  attributes: Record<string, any>,
-): void => {
-  if (AI_USAGE_CACHE_READ_INPUT_TOKENS in attributes) {
-    attributes[SpanAttributes.LLM_USAGE_CACHE_READ_INPUT_TOKENS] =
-      attributes[AI_USAGE_CACHE_READ_INPUT_TOKENS];
-    delete attributes[AI_USAGE_CACHE_READ_INPUT_TOKENS];
-  }
-};
-
 const transformVendor = (attributes: Record<string, any>): void => {
   if (AI_MODEL_PROVIDER in attributes) {
     const vendor = attributes[AI_MODEL_PROVIDER];
@@ -484,8 +461,6 @@ export const transformLLMSpans = (
   transformTools(attributes);
   transformPromptTokens(attributes);
   transformCompletionTokens(attributes);
-  transformCacheCreationInputTokens(attributes);
-  transformCacheReadInputTokens(attributes);
   calculateTotalTokens(attributes);
   transformVendor(attributes);
   transformTelemetryMetadata(attributes, spanName);

packages/traceloop-sdk/src/lib/tracing/ai-sdk-transformations.ts

@@ -269,168 +269,4 @@ describe("Test AI SDK Integration with Recording", function () {
     assert.ok(outputMessages[0].parts[0].content);
     assert.ok(typeof outputMessages[0].parts[0].content === "string");
   });
-
-  it("should capture and transform cache tokens from OpenAI with prompt caching", async function () {
-    this.timeout(30000);
-    const basePrompt =
-      "You are an expert AI assistant with comprehensive knowledge across numerous domains. " +
-      "Your responses should be accurate, detailed, and thoughtful. " +
-      "Always consider multiple perspectives and provide thorough explanations. " +
-      "\n\n" +
-      "## Guidelines for Different Domains:\n\n" +
-      "### Mathematics\n" +
-      "- Show all steps in your calculations\n" +
-      "- Explain the reasoning behind each step\n" +
-      "- Provide examples when helpful\n" +
-      "- Double-check your arithmetic\n" +
-      "\n" +
-      "### Science\n" +
-      "- Reference fundamental scientific principles\n" +
-      "- Cite recent discoveries when relevant\n" +
-      "- Explain complex concepts in accessible terms\n" +
-      "- Distinguish between established facts and theories\n" +
-      "\n" +
-      "### History\n" +
-      "- Consider broader historical context\n" +
-      "- Present multiple viewpoints\n" +
-      "- Acknowledge complexity and nuance\n" +
-      "- Connect historical events to modern implications\n" +
-      "\n" +
-      "### Literature\n" +
-      "- Analyze themes, symbolism, and motifs\n" +
-      "- Examine character development\n" +
-      "- Discuss literary devices and techniques\n" +
-      "- Place works in their cultural and historical context\n" +
-      "\n" +
-      "### Technology\n" +
-      "- Explain both practical applications and underlying concepts\n" +
-      "- Discuss benefits and potential drawbacks\n" +
-      "- Consider ethical implications\n" +
-      "- Keep up with current developments\n" +
-      "\n" +
-      "### Philosophy\n" +
-      "- Explore different schools of thought\n" +
-      "- Present arguments fairly and objectively\n" +
-      "- Examine ethical implications\n" +
-      "- Connect philosophical concepts to real-world scenarios\n" +
-      "\n";
-
-    const longSystemPrompt = basePrompt.repeat(30);
-
-    const result1 = await traceloop.withWorkflow(
-      { name: "test_cache_creation" },
-      async () => {
-        return await generateText({
-          messages: [
-            { role: "system", content: longSystemPrompt },
-            { role: "user", content: "What is 2+2?" },
-          ],
-          model: vercel_openai("gpt-4o-mini"),
-          experimental_telemetry: { isEnabled: true },
-        });
-      },
-    );
-
-    assert.ok(result1);
-    assert.ok(result1.text);
-
-    await traceloop.forceFlush();
-    const spans1 = memoryExporter.getFinishedSpans();
-    const firstCallSpan = spans1.find((span) => span.name === "text.generate");
-
-    assert.ok(firstCallSpan);
-    assert.ok(firstCallSpan.attributes[SpanAttributes.LLM_USAGE_INPUT_TOKENS]);
-    assert.ok(firstCallSpan.attributes[SpanAttributes.LLM_USAGE_OUTPUT_TOKENS]);
-
-    assert.strictEqual(
-      firstCallSpan.attributes["ai.usage.cachedInputTokens"],
-      undefined,
-    );
-    assert.strictEqual(
-      firstCallSpan.attributes["ai.usage.cacheCreationInputTokens"],
-      undefined,
-    );
-    assert.strictEqual(
-      firstCallSpan.attributes["ai.usage.cacheReadInputTokens"],
-      undefined,
-    );
-
-    if (
-      firstCallSpan.attributes[
-        SpanAttributes.LLM_USAGE_CACHE_CREATION_INPUT_TOKENS
-      ]
-    ) {
-      assert.ok(
-        Number(
-          firstCallSpan.attributes[
-            SpanAttributes.LLM_USAGE_CACHE_CREATION_INPUT_TOKENS
-          ],
-        ) > 0,
-        "Cache creation tokens should be > 0",
-      );
-    }
-
-    memoryExporter.reset();
-
-    const result2 = await traceloop.withWorkflow(
-      { name: "test_cache_read" },
-      async () => {
-        return await generateText({
-          messages: [
-            { role: "system", content: longSystemPrompt },
-            { role: "user", content: "What is 3+3?" },
-          ],
-          model: vercel_openai("gpt-4o-mini"),
-          experimental_telemetry: { isEnabled: true },
-        });
-      },
-    );
-
-    assert.ok(result2);
-    assert.ok(result2.text);
-
-    await traceloop.forceFlush();
-    const spans2 = memoryExporter.getFinishedSpans();
-    const secondCallSpan = spans2.find((span) => span.name === "text.generate");
-
-    assert.ok(secondCallSpan);
-    assert.ok(secondCallSpan.attributes[SpanAttributes.LLM_USAGE_INPUT_TOKENS]);
-    assert.ok(
-      secondCallSpan.attributes[SpanAttributes.LLM_USAGE_OUTPUT_TOKENS],
-    );
-
-    assert.strictEqual(
-      secondCallSpan.attributes["ai.usage.cachedInputTokens"],
-      undefined,
-    );
-    assert.strictEqual(
-      secondCallSpan.attributes["ai.usage.cacheCreationInputTokens"],
-      undefined,
-    );
-    assert.strictEqual(
-      secondCallSpan.attributes["ai.usage.cacheReadInputTokens"],
-      undefined,
-    );
-
-    if (
-      secondCallSpan.attributes[
-        SpanAttributes.LLM_USAGE_CACHE_READ_INPUT_TOKENS
-      ]
-    ) {
-      const cacheReadTokens = Number(
-        secondCallSpan.attributes[
-          SpanAttributes.LLM_USAGE_CACHE_READ_INPUT_TOKENS
-        ],
-      );
-      assert.ok(
-        cacheReadTokens > 0,
-        "Cache read tokens should be > 0 when cache is used",
-      );
-      assert.strictEqual(
-        cacheReadTokens,
-        6900,
-        "Expected 6900 cache read tokens from recording",
-      );
-    }
-  });
 });