submission_id: meta-llama-llama-guard-3-8b_v3
developer_uid: chai_backend_admin
alignment_samples: 12754
alignment_score: -1.0859104477670847
best_of: 1
celo_rating: 1114.91
display_name: meta-llama-llama-guard-3-8b_v3
formatter: {'memory_template': "{bot_name}'s Persona: {memory}\n####\n", 'prompt_template': '{prompt}\n<START>\n', 'bot_template': '{bot_name}: {message}\n', 'user_template': '{user_name}: {message}\n', 'response_template': '{bot_name}:', 'truncate_by_message': False}
generation_params: {'temperature': 1.0, 'top_p': 1.0, 'min_p': 0.0, 'top_k': 40, 'presence_penalty': 0.0, 'frequency_penalty': 0.0, 'stopping_words': ['\n'], 'max_input_tokens': 2048, 'best_of': 1, 'max_output_tokens': 64}
gpu_counts: {'NVIDIA RTX A5000': 1}
is_internal_developer: True
language_model: meta-llama/Llama-Guard-3-8B
latencies: [{'batch_size': 1, 'throughput': 0.76401328921732, 'latency_mean': 1.3088135671615602, 'latency_p50': 1.3091638088226318, 'latency_p90': 1.4207462310791015}, {'batch_size': 4, 'throughput': 1.588916181637392, 'latency_mean': 2.505913132429123, 'latency_p50': 2.518923759460449, 'latency_p90': 2.902494144439697}, {'batch_size': 5, 'throughput': 1.734730542481028, 'latency_mean': 2.8670556104183196, 'latency_p50': 2.863649606704712, 'latency_p90': 3.298884558677673}, {'batch_size': 8, 'throughput': 1.98334873855821, 'latency_mean': 4.002783763408661, 'latency_p50': 4.0433491468429565, 'latency_p90': 4.528601503372192}, {'batch_size': 10, 'throughput': 2.0735282767910572, 'latency_mean': 4.784525997638703, 'latency_p50': 4.786468148231506, 'latency_p90': 5.360612678527832}, {'batch_size': 12, 'throughput': 2.0935162165556607, 'latency_mean': 5.68587018609047, 'latency_p50': 5.667144775390625, 'latency_p90': 6.318301057815551}, {'batch_size': 15, 'throughput': 2.095121900207965, 'latency_mean': 7.047889933586121, 'latency_p50': 7.063286066055298, 'latency_p90': 7.743583154678345}]
max_input_tokens: 2048
max_output_tokens: 64
model_architecture: LlamaForCausalLM
model_group: meta-llama/Llama-Guard-3
model_name: meta-llama-llama-guard-3-8b_v3
model_num_parameters: 8030261248.0
model_repo: meta-llama/Llama-Guard-3-8B
model_size: 8B
num_battles: 12753
num_wins: 4292
propriety_score: 0.7208646616541353
propriety_total_count: 1064.0
ranking_group: single
status: inactive
submission_type: basic
throughput_3p7s: 1.95
timestamp: 2024-09-06T01:10:36+00:00
us_pacific_date: 2024-09-05
win_ratio: 0.33654826315376773
Download Preference Data
Resubmit model
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 meta-llama-llama-guard-3-8b-v3-mkmlizer
Waiting for job on meta-llama-llama-guard-3-8b-v3-mkmlizer to finish
Connection pool is full, discarding connection: %s. Connection pool size: %s
meta-llama-llama-guard-3-8b-v3-mkmlizer: ╔═════════════════════════════════════════════════════════════════════╗
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║     _____            __           __                                ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║    / _/ /_ ___    __/ /  ___ ___ / /                                ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║   / _/ / // / |/|/ / _ \/ -_) -_) /                                 ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  /_//_/\_, /|__,__/_//_/\__/\__/_/                                  ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║       /___/                                                         ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║                                                                     ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  Version: 0.10.1                                                    ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  Copyright 2023 MK ONE TECHNOLOGIES Inc.                            ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  https://mk1.ai                                                     ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║                                                                     ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  The license key for the current software has been verified as      ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  belonging to:                                                      ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║                                                                     ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  Chai Research Corp.                                                ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  Account ID: 7997a29f-0ceb-4cc7-9adf-840c57b4ae6f                   ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║  Expiration: 2024-10-15 23:59:59                                    ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ║                                                                     ║
meta-llama-llama-guard-3-8b-v3-mkmlizer: ╚═════════════════════════════════════════════════════════════════════╝
meta-llama-llama-guard-3-8b-v3-mkmlizer: Downloaded to shared memory in 47.927s
meta-llama-llama-guard-3-8b-v3-mkmlizer: quantizing model to /dev/shm/model_cache, profile:s0, folder:/tmp/tmp9refnr2t, device:0
meta-llama-llama-guard-3-8b-v3-mkmlizer: Saving flywheel model at /dev/shm/model_cache
meta-llama-llama-guard-3-8b-v3-mkmlizer: quantized model in 25.873s
meta-llama-llama-guard-3-8b-v3-mkmlizer: Processed model meta-llama/Llama-Guard-3-8B in 73.800s
meta-llama-llama-guard-3-8b-v3-mkmlizer: creating bucket guanaco-mkml-models
meta-llama-llama-guard-3-8b-v3-mkmlizer: Bucket 's3://guanaco-mkml-models/' created
meta-llama-llama-guard-3-8b-v3-mkmlizer: uploading /dev/shm/model_cache to s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v3
meta-llama-llama-guard-3-8b-v3-mkmlizer: cp /dev/shm/model_cache/special_tokens_map.json s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v3/special_tokens_map.json
meta-llama-llama-guard-3-8b-v3-mkmlizer: cp /dev/shm/model_cache/config.json s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v3/config.json
meta-llama-llama-guard-3-8b-v3-mkmlizer: cp /dev/shm/model_cache/tokenizer_config.json s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v3/tokenizer_config.json
meta-llama-llama-guard-3-8b-v3-mkmlizer: cp /dev/shm/model_cache/tokenizer.json s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v3/tokenizer.json
meta-llama-llama-guard-3-8b-v3-mkmlizer: cp /dev/shm/model_cache/flywheel_model.0.safetensors s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v3/flywheel_model.0.safetensors
meta-llama-llama-guard-3-8b-v3-mkmlizer: 
Loading 0:   0%|          | 0/291 [00:00<?, ?it/s]
Loading 0:   2%|▏         | 5/291 [00:00<00:08, 33.85it/s]
Loading 0:   4%|▍         | 13/291 [00:00<00:05, 54.53it/s]
Loading 0:   7%|▋         | 19/291 [00:00<00:05, 47.46it/s]
Loading 0:   8%|▊         | 24/291 [00:00<00:05, 46.72it/s]
Loading 0:  11%|█         | 31/291 [00:00<00:04, 53.55it/s]
Loading 0:  13%|█▎        | 37/291 [00:00<00:05, 46.13it/s]
Loading 0:  14%|█▍        | 42/291 [00:00<00:05, 46.21it/s]
Loading 0:  17%|█▋        | 49/291 [00:00<00:04, 52.12it/s]
Loading 0:  19%|█▉        | 55/291 [00:01<00:04, 47.24it/s]
Loading 0:  21%|██        | 60/291 [00:01<00:04, 47.19it/s]
Loading 0:  23%|██▎       | 68/291 [00:01<00:04, 47.36it/s]
Loading 0:  26%|██▌       | 76/291 [00:01<00:04, 53.42it/s]
Loading 0:  28%|██▊       | 82/291 [00:01<00:04, 48.50it/s]
Loading 0:  30%|███       | 88/291 [00:01<00:05, 34.33it/s]
Loading 0:  32%|███▏      | 94/291 [00:02<00:05, 37.96it/s]
Loading 0:  34%|███▍      | 100/291 [00:02<00:04, 39.16it/s]
Loading 0:  36%|███▌      | 105/291 [00:02<00:04, 40.38it/s]
Loading 0:  38%|███▊      | 112/291 [00:02<00:03, 46.97it/s]
Loading 0:  41%|████      | 118/291 [00:02<00:03, 45.88it/s]
Loading 0:  42%|████▏     | 123/291 [00:02<00:03, 44.05it/s]
Loading 0:  45%|████▍     | 130/291 [00:02<00:03, 48.83it/s]
Loading 0:  47%|████▋     | 136/291 [00:03<00:03, 43.76it/s]
Loading 0:  48%|████▊     | 141/291 [00:03<00:03, 43.00it/s]
Loading 0:  51%|█████     | 148/291 [00:03<00:02, 48.73it/s]
Loading 0:  53%|█████▎    | 154/291 [00:03<00:03, 44.08it/s]
Loading 0:  55%|█████▍    | 159/291 [00:03<00:02, 44.48it/s]
Loading 0:  57%|█████▋    | 165/291 [00:03<00:02, 47.54it/s]
Loading 0:  58%|█████▊    | 170/291 [00:03<00:02, 47.11it/s]
Loading 0:  60%|██████    | 176/291 [00:03<00:02, 50.37it/s]
Loading 0:  63%|██████▎   | 182/291 [00:04<00:02, 44.59it/s]
Loading 0:  64%|██████▍   | 187/291 [00:04<00:02, 35.22it/s]
Loading 0:  66%|██████▌   | 191/291 [00:04<00:02, 35.99it/s]
Loading 0:  67%|██████▋   | 195/291 [00:04<00:02, 35.63it/s]
Loading 0:  69%|██████▉   | 202/291 [00:04<00:02, 43.75it/s]
Loading 0:  71%|███████▏  | 208/291 [00:04<00:01, 43.12it/s]
Loading 0:  73%|███████▎  | 213/291 [00:04<00:01, 44.04it/s]
Loading 0:  76%|███████▌  | 220/291 [00:04<00:01, 50.50it/s]
Loading 0:  78%|███████▊  | 226/291 [00:05<00:01, 49.44it/s]
Loading 0:  80%|███████▉  | 232/291 [00:05<00:01, 51.07it/s]
Loading 0:  82%|████████▏ | 239/291 [00:05<00:01, 48.50it/s]
Loading 0:  85%|████████▌ | 248/291 [00:05<00:00, 49.43it/s]
Loading 0:  88%|████████▊ | 256/291 [00:05<00:00, 55.47it/s]
Loading 0:  90%|█████████ | 262/291 [00:05<00:00, 51.47it/s]
Loading 0:  92%|█████████▏| 268/291 [00:05<00:00, 51.89it/s]
Loading 0:  94%|█████████▍| 274/291 [00:05<00:00, 53.06it/s]
Loading 0:  96%|█████████▌| 280/291 [00:06<00:00, 48.47it/s]
Loading 0:  98%|█████████▊| 285/291 [00:06<00:00, 47.45it/s]
Loading 0: 100%|█████████▉| 290/291 [00:11<00:00,  3.38it/s]
                                                            

Job meta-llama-llama-guard-3-8b-v3-mkmlizer completed after 95.0s with status: succeeded
Stopping job with name meta-llama-llama-guard-3-8b-v3-mkmlizer
Pipeline stage MKMLizer completed in 95.88s
run pipeline stage %s
Running pipeline stage MKMLTemplater
Pipeline stage MKMLTemplater completed in 0.09s
run pipeline stage %s
Running pipeline stage MKMLDeployer
Creating inference service meta-llama-llama-guard-3-8b-v3
Waiting for inference service meta-llama-llama-guard-3-8b-v3 to be ready
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Connection pool is full, discarding connection: %s. Connection pool size: %s
Inference service meta-llama-llama-guard-3-8b-v3 ready after 150.49059510231018s
Pipeline stage MKMLDeployer completed in 150.89s
run pipeline stage %s
Running pipeline stage StressChecker
Received healthy response to inference request in 1.670989990234375s
Received healthy response to inference request in 1.5140244960784912s
Received healthy response to inference request in 0.7133989334106445s
Received healthy response to inference request in 1.5993566513061523s
Received healthy response to inference request in 1.4056143760681152s
5 requests
0 failed requests
5th percentile: 0.8518420219421386
10th percentile: 0.9902851104736328
20th percentile: 1.267171287536621
30th percentile: 1.4272964000701904
40th percentile: 1.4706604480743408
50th percentile: 1.5140244960784912
60th percentile: 1.5481573581695556
70th percentile: 1.5822902202606202
80th percentile: 1.613683319091797
90th percentile: 1.642336654663086
95th percentile: 1.6566633224487304
99th percentile: 1.668124656677246
mean time: 1.3806768894195556
Pipeline stage StressChecker completed in 7.64s
run pipeline stage %s
Running pipeline stage TriggerMKMLProfilingPipeline
run_pipeline:run_in_cloud %s
starting trigger_guanaco_pipeline args=%s
Pipeline stage TriggerMKMLProfilingPipeline completed in 4.81s
Shutdown handler de-registered
meta-llama-llama-guard-3-8b_v3 status is now deployed due to DeploymentManager action
Shutdown handler registered
run pipeline %s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeleter
Skipping teardown as no inference service was successfully deployed
Pipeline stage MKMLProfilerDeleter completed in 0.14s
run pipeline stage %s
Running pipeline stage MKMLProfilerTemplater
Pipeline stage MKMLProfilerTemplater completed in 0.12s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeployer
Creating inference service meta-llama-llama-guard-3-8b-v3-profiler
Waiting for inference service meta-llama-llama-guard-3-8b-v3-profiler to be ready
Inference service meta-llama-llama-guard-3-8b-v3-profiler ready after 150.34655570983887s
Pipeline stage MKMLProfilerDeployer completed in 150.72s
run pipeline stage %s
Running pipeline stage MKMLProfilerRunner
kubectl cp /code/guanaco/guanaco_inference_services/src/inference_scripts tenant-chaiml-guanaco/meta-llama-llama-gua86216e2e383020240037e62ee9690e82-deplor9zf5:/code/chaiverse_profiler_1725585493 --namespace tenant-chaiml-guanaco
kubectl exec -it meta-llama-llama-gua86216e2e383020240037e62ee9690e82-deplor9zf5 --namespace tenant-chaiml-guanaco -- sh -c 'cd /code/chaiverse_profiler_1725585493 && python profiles.py profile --best_of_n 1 --auto_batch 5 --batches 1,5,10,15,20,25,30,35,40,45,50,55,60,65,70,75,80,85,90,95,100,105,110,115,120,125,130,135,140,145,150,155,160,165,170,175,180,185,190,195 --samples 200 --input_tokens 2048 --output_tokens 64 --summary /code/chaiverse_profiler_1725585493/summary.json'
kubectl exec -it meta-llama-llama-gua86216e2e383020240037e62ee9690e82-deplor9zf5 --namespace tenant-chaiml-guanaco -- bash -c 'cat /code/chaiverse_profiler_1725585493/summary.json'
Pipeline stage MKMLProfilerRunner completed in 895.46s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeleter
Checking if service meta-llama-llama-guard-3-8b-v3-profiler is running
Tearing down inference service meta-llama-llama-guard-3-8b-v3-profiler
Service meta-llama-llama-guard-3-8b-v3-profiler has been torndown
Pipeline stage MKMLProfilerDeleter completed in 1.67s
Shutdown handler de-registered
meta-llama-llama-guard-3-8b_v3 status is now inactive due to auto deactivation removed underperforming models

Usage Metrics

Latency Metrics