@elizaos/training 2.0.0-alpha.41 → 2.0.0-alpha.44

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@elizaos/training",
-  "version": "2.0.0-alpha.41",
+  "version": "2.0.0-alpha.44",
   "description": "ElizaOS RL training pipeline with benchmarking and model publishing support",
   "main": "./src/index.ts",
   "types": "./src/index.ts",
@@ -53,5 +53,5 @@
     "bun-types": "^1.3.2",
     "typescript": "^5.9.3"
   },
-  "gitHead": "b3e37e421bcd49b6bc7a34373edc7b3b3a282b8b"
+  "gitHead": "2b27a4e70ebdf054b117b87ed9e8f9f709fe006b"
 }

package/research-output/training-runs/training-run-1773726941205.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T05:55:41.205Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 139
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 2130
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 0
+    }
+  ],
+  "totalDuration": 2270,
+  "success": true
+}

package/research-output/training-runs/training-run-1773742857616.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T10:20:57.616Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 132
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 1528
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 0
+    }
+  ],
+  "totalDuration": 1660,
+  "success": true
+}

package/research-output/training-runs/training-run-1773742946977.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T10:22:26.977Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 55
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 1215
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 0
+    }
+  ],
+  "totalDuration": 1270,
+  "success": true
+}

package/research-output/training-runs/training-run-1773743278891.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T10:27:58.891Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 315
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 3870
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 0
+    }
+  ],
+  "totalDuration": 4185,
+  "success": true
+}

package/research-output/training-runs/training-run-1773743409754.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T10:30:09.754Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 145
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 2265
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 1
+    }
+  ],
+  "totalDuration": 2412,
+  "success": true
+}

package/research-output/training-runs/training-run-1773743651086.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T10:34:11.086Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 138
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 1809
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 1
+    }
+  ],
+  "totalDuration": 1949,
+  "success": true
+}

package/research-output/training-runs/training-run-1773743782883.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T10:36:22.883Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 40
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 1138
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 0
+    }
+  ],
+  "totalDuration": 1178,
+  "success": true
+}

package/research-output/training-runs/training-run-1773755075895.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T13:44:35.895Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 138
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 2068
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 0
+    }
+  ],
+  "totalDuration": 2208,
+  "success": true
+}

package/research-output/training-runs/training-run-1773755142682.json

@@ -0,0 +1,38 @@
+{
+  "timestamp": "2026-03-17T13:45:42.682Z",
+  "config": {
+    "skipTraining": true,
+    "skipBenchmark": true,
+    "ticks": 100,
+    "archetype": "trader",
+    "verbose": false
+  },
+  "results": [
+    {
+      "name": "Check Prerequisites",
+      "success": true,
+      "message": "Prerequisites satisfied (Ollama: yes)",
+      "details": {
+        "python": true,
+        "trainingDir": true,
+        "mlx": false,
+        "ollama": true
+      },
+      "duration": 132
+    },
+    {
+      "name": "Install Dependencies",
+      "success": true,
+      "message": "Python dependencies installed",
+      "duration": 2688
+    },
+    {
+      "name": "Train Model",
+      "success": true,
+      "message": "Training skipped (--skip-training)",
+      "duration": 0
+    }
+  ],
+  "totalDuration": 2820,
+  "success": true
+}

package/scripts/rank_trajectories.ts

@@ -67,7 +67,6 @@ async function main() {
     // Initialize Judge Runtime
     const character = {
         name: 'JudgeAgent',
-        modelProvider: "openai" as any,
         bio: ['I am an impartial AI judge.'],
         settings: {
             secrets: {
@@ -138,7 +137,18 @@ async function main() {
 
     console.log(`Found ${lines.length} trajectories to rank.`);
 
-    const scoredTrajectories = [];
+    interface ScoredTrajectory {
+        trajectoryId?: string;
+        steps?: Array<{
+            action?: { parameters?: { text?: string } };
+        }>;
+        metadata?: { task?: string };
+        score?: number;
+        reasoning?: string;
+        isScored?: boolean;
+    }
+
+    const scoredTrajectories: ScoredTrajectory[] = [];
 
     // Clear output file first if overwriting
     if (fs.existsSync(outputFile)) {
@@ -147,12 +157,12 @@ async function main() {
 
     for (const line of lines) {
         try {
-            const trajectory = JSON.parse(line);
+            const trajectory = JSON.parse(line) as ScoredTrajectory;
             const { steps, metadata } = trajectory;
             const task = metadata?.task || 'Unknown Task';
 
             // Extract the last step's action/response
-            const lastStep = steps[steps.length - 1];
+            const lastStep = steps && steps.length > 0 ? steps[steps.length - 1] : undefined;
             const response = lastStep?.action?.parameters?.text || "No response found";
 
             console.log(`Ranking trajectory ${trajectory.trajectoryId}...`);
@@ -177,11 +187,15 @@ Return ONLY valid JSON.
             const resultText = typeof result === 'string' ? result : result.text;
 
             // Parse JSON
-            let scoreData;
+            interface ScoreData {
+                score: number;
+                reasoning: string;
+            }
+            let scoreData: ScoreData;
             try {
                 // simple cleanup for markdown code blocks
                 const jsonStr = resultText.replace(/```json/g, '').replace(/```/g, '').trim();
-                scoreData = JSON.parse(jsonStr);
+                scoreData = JSON.parse(jsonStr) as ScoreData;
             } catch (e) {
                 console.warn(`Failed to parse judge output for ${trajectory.trajectoryId}: ${resultText}`);
                 scoreData = { score: 0, reasoning: "Parse Error" };

package/scripts/run_task_benchmark.ts

@@ -50,8 +50,7 @@ class BenchmarkRuntimeManager implements IAgentRuntimeManager {
         // Create a new runtime
         const character = {
             name: 'BenchmarkAgent',
-            modelProvider: "openai" as any,
-            bio: 'A helpful assistant for benchmarking.',
+            bio: ['A helpful assistant for benchmarking.'],
             settings: {
                 secrets: {
                     OPENAI_API_KEY: process.env.OPENAI_API_KEY || ''
@@ -60,13 +59,7 @@ class BenchmarkRuntimeManager implements IAgentRuntimeManager {
         };
 
         const runtime = new AgentRuntime({
-            token: process.env.OPENAI_API_KEY || '',
-            modelProvider: "openai" as any,
             character,
-            plugins: [],
-            providers: [],
-            actions: [],
-            evaluators: [],
         });
 
         // We must initialize with allowNoDatabase to avoid DB error
@@ -116,7 +109,7 @@ class BenchmarkTaskInteractor implements ITaskInteractor {
 
             const userMemory: Memory = {
                 id: messageId as `${string}-${string}-${string}-${string}-${string}`,
-                userId: userId as `${string}-${string}-${string}-${string}-${string}`,
+                entityId: userId as `${string}-${string}-${string}-${string}-${string}`,
                 agentId: runtime.agentId,
                 roomId: roomId as `${string}-${string}-${string}-${string}-${string}`,
                 content: {
@@ -143,6 +136,7 @@ Assistant:`;
             // Signature: generateText(input: string, options?: GenerateTextOptions)
             const result = await runtime.generateText(context, {
                 modelType: ModelType.TEXT_SMALL,
+                stopSequences: [],
             });
             // Handle both string and object return types for safety
             const response = typeof result === 'string' ? result : result.text;
@@ -227,11 +221,11 @@ async function main() {
         agentService: new BenchmarkAgentService(),
         agentRuntimeManager: new BenchmarkRuntimeManager(),
         autonomousCoordinator: {
-            executeAutonomousTick: async () => ({ success: true })
-        } as any,
+            executeAutonomousTick: async () => ({ success: true }),
+        },
         llmCaller: {
-            callGroqDirect: async () => "mock response"
-        } as any,
+            callGroqDirect: async () => "mock response",
+        },
     });
 
     // Import task interactor config