deps: update llama.cpp (#99)

- this fixes timeouts when running llama 3.1
- bypass storing failed jobs in redis if they were ran with external processor

skynet/modules/ttt/summaries/jobs.py

@@ -137,6 +137,9 @@ async def _run_job(job: Job) -> None:
 
     await update_job(job_id=job.id, start=start, status=JobStatus.RUNNING, worker_id=worker_id)
 
+    customer_id = job.metadata.customer_id
+    processor = get_job_processor(customer_id)  # may have changed since job was created
+
     # add to running jobs list if not already there (which may occur on multiple worker disconnects while running the same job)
     if job.id not in await db.lrange(RUNNING_JOBS_KEY, 0, -1):
         await db.rpush(RUNNING_JOBS_KEY, job.id)
@@ -147,10 +150,8 @@ async def _run_job(job: Job) -> None:
         result = job.payload.text
     else:
         try:
-            customer_id = job.metadata.customer_id
             options = get_credentials(customer_id)
             secret = options.get('secret')
-            processor = get_job_processor(customer_id)  # may have changed since job was created
 
             if processor == Processors.OPENAI:
                 log.info(f"Forwarding inference to OpenAI for customer {customer_id}")
@@ -176,15 +177,17 @@ async def _run_job(job: Job) -> None:
             has_failed = True
             result = str(e)
 
+    should_expire = not has_failed or processor != Processors.LOCAL
+
     updated_job = await update_job(
-        expires=redis_exp_seconds if not has_failed else None,
+        expires=redis_exp_seconds if should_expire else None,
         job_id=job.id,
         end=time.time(),
         status=JobStatus.ERROR if has_failed else JobStatus.SUCCESS,
         result=result,
     )
 
-    if has_failed:
+    if not should_expire:
         await db.rpush(ERROR_JOBS_KEY, job.id)
         SUMMARY_ERROR_COUNTER.inc()