Experimental TypeScript/JavaScript runtime

cog-fixture/cog.yaml

@@ -0,0 +1,4 @@
+# Example cog.yaml for JavaScript/TypeScript predictor
+predict: "./example_predictor.ts:simplePredictFunction"
+concurrency:
+  max: 2

cog-fixture/example_predictor.ts

@@ -0,0 +1,10 @@
+// Example function-based predictor
+export function simplePredictFunction(input) {
+  const multiplier = input.multiplier || 1;
+  return {
+    original: input.text,
+    uppercase: input.text.toUpperCase(),
+    length: input.text.length * multiplier,
+    timestamp: new Date().toISOString(),
+  };
+}

deno-coglet/README.md

@@ -0,0 +1,86 @@
+# Deno Coglet
+
+A Deno/TypeScript implementation of the Cog runtime interface that allows JavaScript/TypeScript predictors to work with the Cog Go runner.
+
+## Overview
+
+This implementation provides the same file-based IPC interface as the Python coglet, enabling the Go runner to manage JavaScript/TypeScript prediction functions and classes.
+
+## Architecture
+
+The communication protocol uses:
+- **File-based IPC** for exchanging requests and responses
+- **HTTP status updates** to notify the Go runner of state changes
+- **Formatted logging** for proper log routing
+
+## Usage
+
+### Running the Coglet
+
+```bash
+deno run --allow-read --allow-write --allow-net --allow-env \
+  coglet.ts \
+  --ipc-url http://localhost:8080 \
+  --working-dir /tmp/work
+```
+
+### Creating a Predictor
+
+You can create predictors as functions or classes:
+
+```typescript
+// Function predictor
+export function myPredictor(input: { text: string }) {
+  return { result: input.text.toUpperCase() };
+}
+
+// Class predictor with setup
+export class MyPredictor {
+  private model: any;
+
+  async setup() {
+    // Initialize your model here
+    this.model = await loadModel();
+  }
+
+  async predict(input: { prompt: string }) {
+    return await this.model.generate(input.prompt);
+  }
+}
+```
+
+### Integration with Go Runner
+
+To use this with the Go runner, you would need to modify the runner to use Deno instead of Python:
+
+```go
+// In runner.go
+args := []string{
+    "run",
+    "--allow-read", "--allow-write", "--allow-net", "--allow-env",
+    "/path/to/coglet.ts",
+    "--ipc-url", ipcUrl,
+    "--working-dir", workingDir,
+}
+cmd := exec.Command("deno", args...)
+```
+
+## File Interface
+
+- `config.json` - Initial configuration from Go
+- `request-{id}.json` - Prediction requests
+- `response-{id}-{epoch}.json` - Prediction responses
+- `cancel-{id}` - Cancel specific prediction
+- `stop` - Shutdown signal
+- `setup_result.json` - Setup status
+- `openapi.json` - API schema
+
+## Testing
+
+Run the test script to see it in action:
+
+```bash
+./test_runner.sh
+```
+
+This will start a mock IPC server, run the coglet with an example predictor, and send a test prediction request.