Shutdown handler not registered because Python interpreter is not running in the main thread
run pipeline %s
run pipeline stage %s
Running pipeline stage MKMLizer
Starting job with name rinen0721-llama1009-v1-mkmlizer
Waiting for job on rinen0721-llama1009-v1-mkmlizer to finish
rinen0721-llama1009-v1-mkmlizer: ╔═════════════════════════════════════════════════════════════════════╗
rinen0721-llama1009-v1-mkmlizer: ║ _____ __ __ ║
rinen0721-llama1009-v1-mkmlizer: ║ / _/ /_ ___ __/ / ___ ___ / / ║
rinen0721-llama1009-v1-mkmlizer: ║ / _/ / // / |/|/ / _ \/ -_) -_) / ║
rinen0721-llama1009-v1-mkmlizer: ║ /_//_/\_, /|__,__/_//_/\__/\__/_/ ║
rinen0721-llama1009-v1-mkmlizer: ║ /___/ ║
rinen0721-llama1009-v1-mkmlizer: ║ ║
rinen0721-llama1009-v1-mkmlizer: ║ Version: 0.11.12 ║
rinen0721-llama1009-v1-mkmlizer: ║ Copyright 2023 MK ONE TECHNOLOGIES Inc. ║
rinen0721-llama1009-v1-mkmlizer: ║ https://mk1.ai ║
rinen0721-llama1009-v1-mkmlizer: ║ ║
rinen0721-llama1009-v1-mkmlizer: ║ The license key for the current software has been verified as ║
rinen0721-llama1009-v1-mkmlizer: ║ belonging to: ║
rinen0721-llama1009-v1-mkmlizer: ║ ║
rinen0721-llama1009-v1-mkmlizer: ║ Chai Research Corp. ║
rinen0721-llama1009-v1-mkmlizer: ║ Account ID: 7997a29f-0ceb-4cc7-9adf-840c57b4ae6f ║
rinen0721-llama1009-v1-mkmlizer: ║ Expiration: 2024-10-15 23:59:59 ║
rinen0721-llama1009-v1-mkmlizer: ║ ║
rinen0721-llama1009-v1-mkmlizer: ╚═════════════════════════════════════════════════════════════════════╝
rinen0721-llama1009-v1-mkmlizer: Downloaded to shared memory in 40.127s
rinen0721-llama1009-v1-mkmlizer: quantizing model to /dev/shm/model_cache, profile:s0, folder:/tmp/tmpsdxztodj, device:0
rinen0721-llama1009-v1-mkmlizer: Saving flywheel model at /dev/shm/model_cache
rinen0721-llama1009-v1-mkmlizer: /opt/conda/lib/python3.10/site-packages/mk1/flywheel/functional/loader.py:55: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
rinen0721-llama1009-v1-mkmlizer: tensors = torch.load(model_shard_filename, map_location=torch.device(self.device), mmap=True)
rinen0721-llama1009-v1-mkmlizer: quantized model in 28.439s
rinen0721-llama1009-v1-mkmlizer: Processed model rinen0721/llama1009 in 68.567s
rinen0721-llama1009-v1-mkmlizer: creating bucket guanaco-mkml-models
rinen0721-llama1009-v1-mkmlizer: Bucket 's3://guanaco-mkml-models/' created
rinen0721-llama1009-v1-mkmlizer: uploading /dev/shm/model_cache to s3://guanaco-mkml-models/rinen0721-llama1009-v1
rinen0721-llama1009-v1-mkmlizer: cp /dev/shm/model_cache/special_tokens_map.json s3://guanaco-mkml-models/rinen0721-llama1009-v1/special_tokens_map.json
rinen0721-llama1009-v1-mkmlizer: cp /dev/shm/model_cache/config.json s3://guanaco-mkml-models/rinen0721-llama1009-v1/config.json
rinen0721-llama1009-v1-mkmlizer: cp /dev/shm/model_cache/tokenizer_config.json s3://guanaco-mkml-models/rinen0721-llama1009-v1/tokenizer_config.json
rinen0721-llama1009-v1-mkmlizer: cp /dev/shm/model_cache/tokenizer.json s3://guanaco-mkml-models/rinen0721-llama1009-v1/tokenizer.json
Job rinen0721-llama1009-v1-mkmlizer completed after 110.4s with status: succeeded
Stopping job with name rinen0721-llama1009-v1-mkmlizer
Pipeline stage MKMLizer completed in 111.14s
run pipeline stage %s
Running pipeline stage MKMLTemplater
Pipeline stage MKMLTemplater completed in 0.18s
run pipeline stage %s
Running pipeline stage MKMLDeployer
Creating inference service rinen0721-llama1009-v1
Waiting for inference service rinen0721-llama1009-v1 to be ready
Inference service rinen0721-llama1009-v1 ready after 130.65059208869934s
Pipeline stage MKMLDeployer completed in 131.24s
run pipeline stage %s
Running pipeline stage StressChecker
Received healthy response to inference request in 1.7643921375274658s
Received healthy response to inference request in 1.599656105041504s
Received healthy response to inference request in 1.2803153991699219s
Received healthy response to inference request in 1.2243387699127197s
Received healthy response to inference request in 1.2433147430419922s
5 requests
0 failed requests
5th percentile: 1.2281339645385743
10th percentile: 1.2319291591644288
20th percentile: 1.2395195484161377
30th percentile: 1.2507148742675782
40th percentile: 1.26551513671875
50th percentile: 1.2803153991699219
60th percentile: 1.4080516815185546
70th percentile: 1.5357879638671874
80th percentile: 1.6326033115386964
90th percentile: 1.698497724533081
95th percentile: 1.7314449310302733
99th percentile: 1.7578026962280273
mean time: 1.4224034309387208
Pipeline stage StressChecker completed in 8.88s
Shutdown handler de-registered
rinen0721-llama1009_v1 status is now deployed due to DeploymentManager action
rinen0721-llama1009_v1 status is now inactive due to auto deactivation removed underperforming models
rinen0721-llama1009_v1 status is now torndown due to DeploymentManager action