Build: Release v2.80.28

src/lmstudio/streamer.ts

@@ -48,6 +48,19 @@ export class LMStudioStreamer implements IChatStreamer {
             signal: req.signal,
         });
 
+        // Bridge AbortSignal → prediction.cancel(): without this, an aborted
+        // request keeps generating on the LM Studio server. The orphaned
+        // prediction holds locks on the model handle, which is a known cause
+        // of "lock() request could not be registered" on the very next
+        // request — the reused handle is still bound to a dead prediction.
+        const onAbort = () => {
+            try { (prediction as any)?.cancel?.(); } catch { /* swallow — best effort */ }
+        };
+        if (req.signal) {
+            if (req.signal.aborted) onAbort();
+            else req.signal.addEventListener('abort', onAbort, { once: true });
+        }
+
         try {
             for await (const fragment of prediction as AsyncIterable<{ content: string }>) {
                 if (req.signal?.aborted) return;
@@ -59,6 +72,8 @@ export class LMStudioStreamer implements IChatStreamer {
             if (err?.name === 'AbortError') return;
             logError('LM Studio SDK chat stream failed.', { model: trimmedModel, error: err?.message ?? String(err) });
             throw err;
+        } finally {
+            req.signal?.removeEventListener?.('abort', onAbort);
         }
     }
 }