@auxot/worker-cli 0.3.4 → 0.3.5

package/dist/index.js

@@ -458,6 +458,10 @@ var WebSocketConnection = class {
               this.isReconnecting = false;
               this.retryDelay = INITIAL_RETRY_DELAY;
               this.startHeartbeat();
+              if (this.capabilities && this.capabilities.ctx_size > 0) {
+                console.log("  Re-sending config after reconnection...");
+                this.sendConfig(this.capabilities);
+              }
               resolve();
             } else {
               const errorMessage = message.error || "Authentication failed";
@@ -807,7 +811,7 @@ async function processJob(job, llamaUrl, capabilities, abortSignal, onToken) {
   try {
     while (true) {
       if (abortSignal.aborted) {
-        break;
+        throw new Error("Job cancelled by user");
       }
       const { done, value } = await reader.read();
       if (done)
@@ -817,8 +821,9 @@ async function processJob(job, llamaUrl, capabilities, abortSignal, onToken) {
       parser.feed(text);
     }
   } catch (error) {
-    if (error instanceof Error && error.name === "AbortError") {
-      console.log(`[Job ${job.job_id}] Cancelled - returning partial response`);
+    if (error instanceof Error && (error.name === "AbortError" || error.message === "Job cancelled by user")) {
+      console.log(`[Job ${job.job_id}] Cancelled`);
+      throw error;
     } else {
       throw error;
     }
@@ -1193,11 +1198,12 @@ async function spawnLlamaCpp(options) {
     host,
     gpuLayers
   };
+  const totalContextSize = contextSize * parallelism;
   const args2 = [
     "--model",
     modelPath,
     "--ctx-size",
-    contextSize.toString(),
+    totalContextSize.toString(),
     "--parallel",
     parallelism.toString(),
     "--port",
@@ -1214,7 +1220,8 @@ async function spawnLlamaCpp(options) {
   if (gpuLayers !== void 0) {
     args2.push("--n-gpu-layers", gpuLayers.toString());
   }
-  console.log(`[llama.cpp] Spawning process: ${binaryPath} ${args2.join(" ")}`);
+  console.log(`[llama.cpp] Spawning with ${parallelism} parallel slots, ${contextSize.toLocaleString()} context per slot (${totalContextSize.toLocaleString()} total)`);
+  console.log(`[llama.cpp] Command: ${binaryPath} ${args2.join(" ")}`);
   const childProcess = spawn(binaryPath, args2, {
     stdio: ["ignore", "pipe", "pipe"],
     // stdin: ignore, stdout/stderr: pipe
@@ -5885,7 +5892,6 @@ async function main() {
             wsConnection.sendToken(job.job_id, token);
           }
         );
-        const wasCancelled = abortController.signal.aborted;
         wsConnection.sendComplete(
           job.job_id,
           result.fullResponse,
@@ -5894,14 +5900,15 @@ async function main() {
           result.outputTokens,
           result.tool_calls
         );
-        if (!wasCancelled) {
-        }
       } catch (error) {
-        console.error(`\u2717 Job ${job.job_id} failed:`, error);
-        wsConnection.sendError(
-          job.job_id,
-          error instanceof Error ? error.message : "Unknown error"
-        );
+        const errorMessage = error instanceof Error ? error.message : "Unknown error";
+        if (errorMessage === "Job cancelled by user" || error instanceof Error && error.name === "AbortError") {
+          console.log(`[Job ${job.job_id}] Cancelled by user`);
+          wsConnection.sendError(job.job_id, "Job cancelled by user");
+        } else {
+          console.error(`\u2717 Job ${job.job_id} failed:`, error);
+          wsConnection.sendError(job.job_id, errorMessage);
+        }
       } finally {
         activeJobs.delete(job.job_id);
       }