@empiricalrun/ai - npm Package Compare versions

Comparing version 0.2.0 to 0.3.0

CHANGELOG.md

 # @empiricalrun/ai
 
+## 0.3.0
+
+### Minor Changes
+
+- a94aa16: feat: add stop reason and token usage metrics to run output
+
 ## 0.2.0
@@ -4,0 +10,0 @@

dist/providers/anthropic/index.js

@@ -12,3 +12,3 @@ "use strict";
 const batchTaskManager = new utils_1.BatchTaskManager(5);
-const finishReaonReverseMap = new Map([
+const finishReasonReverseMap = new Map([
     ["end_turn", "stop"],
@@ -55,2 +55,3 @@ ["max_tokens", "length"],
     try {
+        const startedAt = Date.now();
         const response = await (0, promise_retry_1.default)((retry) => {
@@ -87,2 +88,5 @@ return anthropic.messages
         executionDone();
+        const latency = Date.now() - startedAt;
+        // renaming to terms used by openai, mistral
+        const { input_tokens: prompt_tokens, output_tokens: completion_tokens } = response.usage;
         return {
@@ -93,5 +97,10 @@ id: response.id,
             created: Date.now() / 1000,
+            usage: {
+                total_tokens: prompt_tokens + completion_tokens,
+                completion_tokens,
+                prompt_tokens,
+            },
             choices: [
                 {
-                    finish_reason: finishReaonReverseMap.get(response.stop_reason) || "stop",
+                    finish_reason: finishReasonReverseMap.get(response.stop_reason) || "stop",
                     index: 0,
@@ -105,2 +114,3 @@ message: {
             ],
+            latency,
         };
@@ -107,0 +117,0 @@ }

dist/providers/google/index.js

@@ -53,2 +53,3 @@ "use strict";
     try {
+        const startedAt = Date.now();
         const completion = await (0, promise_retry_1.default)((retry) => {
@@ -67,2 +68,18 @@ // TODO: move to model.startChat which support model config (e.g. temperature)
         executionDone();
+        const latency = Date.now() - startedAt;
+        const responseContent = completion.response.text();
+        let totalTokens = 0, promptTokens = 0, completionTokens = 0;
+        try {
+            [{ totalTokens: completionTokens }, { totalTokens: promptTokens }] =
+                await Promise.all([
+                    modelInstance.countTokens(responseContent),
+                    modelInstance.countTokens({
+                        contents,
+                    }),
+                ]);
+            totalTokens = completionTokens + promptTokens;
+        }
+        catch (e) {
+            console.warn(`Failed to fetch token usage for google:${model}`);
+        }
         const response = {
@@ -75,3 +92,3 @@ id: crypto_1.default.randomUUID(),
                     message: {
-                        content: completion.response.text(),
+                        content: responseContent,
                         role: "assistant",
@@ -84,3 +101,9 @@ },
             created: Date.now(),
+            usage: {
+                total_tokens: totalTokens,
+                prompt_tokens: promptTokens,
+                completion_tokens: completionTokens,
+            },
             model,
+            latency,
         };
@@ -87,0 +110,0 @@ return response;

dist/providers/mistral/index.js

@@ -22,2 +22,3 @@ "use strict";
         const mistralMessages = messages;
+        const startedAt = Date.now();
         // no retry needed as mistral internally handles it well
@@ -35,5 +36,6 @@ const completions = await mistralai.chat({
         executionDone();
+        const latency = Date.now() - startedAt;
         // typecasting as the only difference present in mistral interface is the it doesnt contain logprobs.
         // currently its not being used. hence typecasting it for now.
-        return completions;
+        return { ...completions, latency };
     }
@@ -40,0 +42,0 @@ catch (err) {

dist/providers/openai/index.js

@@ -19,2 +19,3 @@ "use strict";
     try {
+        const startedAt = Date.now();
         const completions = await (0, promise_retry_1.default)((retry) => {
@@ -39,3 +40,4 @@ return openai.chat.completions.create(body).catch((err) => {
         });
-        return completions;
+        const latency = Date.now() - startedAt;
+        return { ...completions, latency };
     }
@@ -42,0 +44,0 @@ catch (err) {

package.json

 {
   "name": "@empiricalrun/ai",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "publishConfig": {
@@ -9,3 +9,3 @@ "registry": "https://registry.npmjs.org/",
   "main": "dist/index.js",
-  "author": "Empirical.run Team <[e-mail]>",
+  "author": "Empirical Team <[e-mail]>",
   "license": "MIT",
@@ -18,3 +18,3 @@ "repository": {
     "@types/promise-retry": "^1.1.6",
-    "@empiricalrun/types": "0.2.0"
+    "@empiricalrun/types": "0.3.1"
   },
@@ -21,0 +21,0 @@ "dependencies": {

New alerts

License Policy Violation

License

This package is not allowed per your license policy. Review the package's license to ensure compliance.

Found 1 instance in 1 package

Fixed alerts

License Policy Violation

License

This package is not allowed per your license policy. Review the package's license to ensure compliance.

Found 1 instance in 1 package

Improved metrics

Total package byte prevSize

32967

increased by5.19%

Lines of code

722

increased by5.4%

No dependency changes