submission_id: meta-llama-llama-guard-3-8b_v1
developer_uid: chai_backend_admin
alignment_samples: 15473
alignment_score: -1.000422863111148
best_of: 1
celo_rating: 1114.37
display_name: meta-llama-llama-guard-3-8b_v1
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': 0.7601499499326027, 'latency_mean': 1.3154347467422485, 'latency_p50': 1.3176430463790894, 'latency_p90': 1.4248308420181275}, {'batch_size': 4, 'throughput': 1.5891962252285416, 'latency_mean': 2.5084204041957854, 'latency_p50': 2.5097827911376953, 'latency_p90': 2.768460512161255}, {'batch_size': 5, 'throughput': 1.728570284841039, 'latency_mean': 2.8757122457027435, 'latency_p50': 2.881162405014038, 'latency_p90': 3.25748131275177}, {'batch_size': 8, 'throughput': 1.9857104120064906, 'latency_mean': 4.009250607490539, 'latency_p50': 4.002607226371765, 'latency_p90': 4.499457597732544}, {'batch_size': 10, 'throughput': 2.0697143964632616, 'latency_mean': 4.796747711896896, 'latency_p50': 4.7729597091674805, 'latency_p90': 5.4019874095916744}, {'batch_size': 12, 'throughput': 2.0852526517106984, 'latency_mean': 5.702628554105758, 'latency_p50': 5.697651028633118, 'latency_p90': 6.298007178306579}, {'batch_size': 15, 'throughput': 2.081261351052274, 'latency_mean': 7.111022329330444, 'latency_p50': 7.154937744140625, 'latency_p90': 7.791962742805481}]
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_v1
model_num_parameters: 8030261248.0
model_repo: meta-llama/Llama-Guard-3-8B
model_size: 8B
num_battles: 15472
num_wins: 5134
propriety_score: 0.7420520231213873
propriety_total_count: 1384.0
ranking_group: single
status: inactive
submission_type: basic
throughput_3p7s: 1.94
timestamp: 2024-09-05T19:55:57+00:00
us_pacific_date: 2024-09-05
win_ratio: 0.33182523267838676
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-v1-mkmlizer
Waiting for job on meta-llama-llama-guard-3-8b-v1-mkmlizer to finish
meta-llama-llama-guard-3-8b-v1-mkmlizer: ╔═════════════════════════════════════════════════════════════════════╗
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║     _____            __           __                                ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║    / _/ /_ ___    __/ /  ___ ___ / /                                ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║   / _/ / // / |/|/ / _ \/ -_) -_) /                                 ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  /_//_/\_, /|__,__/_//_/\__/\__/_/                                  ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║       /___/                                                         ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║                                                                     ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  Version: 0.10.1                                                    ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  Copyright 2023 MK ONE TECHNOLOGIES Inc.                            ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  https://mk1.ai                                                     ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║                                                                     ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  The license key for the current software has been verified as      ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  belonging to:                                                      ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║                                                                     ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  Chai Research Corp.                                                ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  Account ID: 7997a29f-0ceb-4cc7-9adf-840c57b4ae6f                   ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║  Expiration: 2024-10-15 23:59:59                                    ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ║                                                                     ║
meta-llama-llama-guard-3-8b-v1-mkmlizer: ╚═════════════════════════════════════════════════════════════════════╝
meta-llama-llama-guard-3-8b-v1-mkmlizer: Downloaded to shared memory in 47.518s
meta-llama-llama-guard-3-8b-v1-mkmlizer: quantizing model to /dev/shm/model_cache, profile:s0, folder:/tmp/tmp6ekh0gia, device:0
meta-llama-llama-guard-3-8b-v1-mkmlizer: Saving flywheel model at /dev/shm/model_cache
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
meta-llama-llama-guard-3-8b-v1-mkmlizer: quantized model in 26.101s
meta-llama-llama-guard-3-8b-v1-mkmlizer: Processed model meta-llama/Llama-Guard-3-8B in 73.619s
meta-llama-llama-guard-3-8b-v1-mkmlizer: creating bucket guanaco-mkml-models
meta-llama-llama-guard-3-8b-v1-mkmlizer: Bucket 's3://guanaco-mkml-models/' created
meta-llama-llama-guard-3-8b-v1-mkmlizer: uploading /dev/shm/model_cache to s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v1
meta-llama-llama-guard-3-8b-v1-mkmlizer: cp /dev/shm/model_cache/config.json s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v1/config.json
meta-llama-llama-guard-3-8b-v1-mkmlizer: cp /dev/shm/model_cache/special_tokens_map.json s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v1/special_tokens_map.json
meta-llama-llama-guard-3-8b-v1-mkmlizer: cp /dev/shm/model_cache/tokenizer_config.json s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v1/tokenizer_config.json
meta-llama-llama-guard-3-8b-v1-mkmlizer: cp /dev/shm/model_cache/tokenizer.json s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v1/tokenizer.json
meta-llama-llama-guard-3-8b-v1-mkmlizer: cp /dev/shm/model_cache/flywheel_model.0.safetensors s3://guanaco-mkml-models/meta-llama-llama-guard-3-8b-v1/flywheel_model.0.safetensors
meta-llama-llama-guard-3-8b-v1-mkmlizer: 
Loading 0:   0%|          | 0/291 [00:00<?, ?it/s]
Loading 0:   2%|▏         | 5/291 [00:00<00:08, 33.99it/s]
Loading 0:   4%|▍         | 13/291 [00:00<00:05, 54.05it/s]
Loading 0:   7%|▋         | 19/291 [00:00<00:05, 49.86it/s]
Loading 0:   9%|▊         | 25/291 [00:00<00:05, 50.85it/s]
Loading 0:  11%|█         | 31/291 [00:00<00:04, 53.42it/s]
Loading 0:  13%|█▎        | 37/291 [00:00<00:05, 46.63it/s]
Loading 0:  14%|█▍        | 42/291 [00:00<00:05, 46.63it/s]
Loading 0:  17%|█▋        | 50/291 [00:01<00:04, 48.44it/s]
Loading 0:  20%|██        | 59/291 [00:01<00:04, 51.52it/s]
Loading 0:  23%|██▎       | 68/291 [00:01<00:04, 51.62it/s]
Loading 0:  26%|██▌       | 75/291 [00:01<00:03, 55.65it/s]
Loading 0:  28%|██▊       | 81/291 [00:01<00:03, 55.12it/s]
Loading 0:  30%|██▉       | 87/291 [00:01<00:05, 34.85it/s]
Loading 0:  33%|███▎      | 95/291 [00:02<00:05, 38.50it/s]
Loading 0:  35%|███▌      | 103/291 [00:02<00:04, 45.65it/s]
Loading 0:  37%|███▋      | 109/291 [00:02<00:04, 40.87it/s]
Loading 0:  39%|███▉      | 114/291 [00:02<00:04, 41.72it/s]
Loading 0:  42%|████▏     | 121/291 [00:02<00:03, 47.42it/s]
Loading 0:  44%|████▎     | 127/291 [00:02<00:03, 43.84it/s]
Loading 0:  45%|████▌     | 132/291 [00:02<00:03, 43.74it/s]
Loading 0:  48%|████▊     | 140/291 [00:03<00:03, 45.44it/s]
Loading 0:  51%|█████     | 149/291 [00:03<00:02, 48.31it/s]
Loading 0:  54%|█████▍    | 157/291 [00:03<00:02, 54.18it/s]
Loading 0:  56%|█████▌    | 163/291 [00:03<00:02, 50.49it/s]
Loading 0:  58%|█████▊    | 169/291 [00:03<00:02, 51.40it/s]
Loading 0:  60%|██████    | 175/291 [00:03<00:02, 52.67it/s]
Loading 0:  62%|██████▏   | 181/291 [00:03<00:02, 44.79it/s]
Loading 0:  64%|██████▍   | 187/291 [00:04<00:02, 36.73it/s]
Loading 0:  66%|██████▌   | 192/291 [00:04<00:02, 36.60it/s]
Loading 0:  67%|██████▋   | 196/291 [00:04<00:02, 37.15it/s]
Loading 0:  69%|██████▉   | 201/291 [00:04<00:02, 40.05it/s]
Loading 0:  71%|███████   | 206/291 [00:04<00:02, 41.82it/s]
Loading 0:  73%|███████▎  | 211/291 [00:04<00:01, 43.32it/s]
Loading 0:  74%|███████▍  | 216/291 [00:04<00:01, 44.77it/s]
Loading 0:  76%|███████▌  | 221/291 [00:04<00:01, 37.10it/s]
Loading 0:  79%|███████▊  | 229/291 [00:05<00:01, 46.30it/s]
Loading 0:  81%|████████  | 235/291 [00:05<00:01, 45.15it/s]
Loading 0:  82%|████████▏ | 240/291 [00:05<00:01, 45.85it/s]
Loading 0:  85%|████████▍ | 247/291 [00:05<00:00, 51.70it/s]
Loading 0:  87%|████████▋ | 253/291 [00:05<00:00, 49.12it/s]
Loading 0:  89%|████████▉ | 259/291 [00:05<00:00, 49.76it/s]
Loading 0:  91%|█████████▏| 266/291 [00:05<00:00, 45.06it/s]
Loading 0:  94%|█████████▍| 274/291 [00:05<00:00, 51.09it/s]
Loading 0:  96%|█████████▌| 280/291 [00:06<00:00, 49.20it/s]
Loading 0:  98%|█████████▊| 286/291 [00:06<00:00, 46.36it/s]
Loading 0: 100%|██████████| 291/291 [00:11<00:00,  3.53it/s]
                                                            

Job meta-llama-llama-guard-3-8b-v1-mkmlizer completed after 94.5s with status: succeeded
Stopping job with name meta-llama-llama-guard-3-8b-v1-mkmlizer
Pipeline stage MKMLizer completed in 96.28s
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-v1
Waiting for inference service meta-llama-llama-guard-3-8b-v1 to be ready
Failed to get response for submission chaiml-llama-8b-pairwis_8189_v19: ('http://zonemercy-lexical-nemo-1518-v18-predictor.tenant-chaiml-guanaco.k.chaiverse.com/v1/models/GPT-J-6B-lit-v2:predict', 'read tcp 127.0.0.1:46354->127.0.0.1:8080: read: connection reset by peer\n')
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
admin requested tearing down of nousresearch-meta-llama_4939_v19
Shutdown handler not registered because Python interpreter is not running in the main thread
Inference service meta-llama-llama-guard-3-8b-v1 ready after 150.84588098526s
run pipeline %s
Pipeline stage MKMLDeployer completed in 152.72s
run pipeline stage %s
run pipeline stage %s
Running pipeline stage MKMLDeleter
Running pipeline stage StressChecker
%s, retrying in %s seconds...
%s, retrying in %s seconds...
Received healthy response to inference request in 2.0696394443511963s
clean up pipeline due to error=%s
Shutdown handler de-registered
Received healthy response to inference request in 0.5578248500823975s
Received healthy response to inference request in 1.3681178092956543s
Received healthy response to inference request in 1.0186405181884766s
Received healthy response to inference request in 1.0533061027526855s
5 requests
0 failed requests
5th percentile: 0.6499879837036133
10th percentile: 0.7421511173248291
20th percentile: 0.9264773845672608
30th percentile: 1.0255736351013183
40th percentile: 1.039439868927002
50th percentile: 1.0533061027526855
60th percentile: 1.1792307853698731
70th percentile: 1.3051554679870605
80th percentile: 1.5084221363067627
90th percentile: 1.7890307903289795
95th percentile: 1.929335117340088
99th percentile: 2.0415785789489744
mean time: 1.213505744934082
Pipeline stage StressChecker completed in 10.81s
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 4.58s
Shutdown handler de-registered
meta-llama-llama-guard-3-8b_v1 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 0.12s
run pipeline stage %s
Running pipeline stage MKMLProfilerTemplater
Pipeline stage MKMLProfilerTemplater completed in 0.13s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeployer
Creating inference service meta-llama-llama-guard-3-8b-v1-profiler
Waiting for inference service meta-llama-llama-guard-3-8b-v1-profiler to be ready
Inference service meta-llama-llama-guard-3-8b-v1-profiler ready after 150.34625720977783s
Pipeline stage MKMLProfilerDeployer completed in 150.74s
run pipeline stage %s
Running pipeline stage MKMLProfilerRunner
kubectl cp /code/guanaco/guanaco_inference_services/src/inference_scripts tenant-chaiml-guanaco/meta-llama-llama-guaced2249432faa6ef9c8b5409f89345a9-deplohswwv:/code/chaiverse_profiler_1725566618 --namespace tenant-chaiml-guanaco
kubectl exec -it meta-llama-llama-guaced2249432faa6ef9c8b5409f89345a9-deplohswwv --namespace tenant-chaiml-guanaco -- sh -c 'cd /code/chaiverse_profiler_1725566618 && 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_1725566618/summary.json'
kubectl exec -it meta-llama-llama-guaced2249432faa6ef9c8b5409f89345a9-deplohswwv --namespace tenant-chaiml-guanaco -- bash -c 'cat /code/chaiverse_profiler_1725566618/summary.json'
Pipeline stage MKMLProfilerRunner completed in 899.43s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeleter
Checking if service meta-llama-llama-guard-3-8b-v1-profiler is running
Tearing down inference service meta-llama-llama-guard-3-8b-v1-profiler
Service meta-llama-llama-guard-3-8b-v1-profiler has been torndown
Pipeline stage MKMLProfilerDeleter completed in 1.79s
Shutdown handler de-registered
meta-llama-llama-guard-3-8b_v1 status is now inactive due to auto deactivation removed underperforming models

Usage Metrics

Latency Metrics