submission_id: jic062-dpo-v1-4-nemo-c1000_v2
developer_uid: chace9580
best_of: 8
celo_rating: 1256.77
display_name: jic062-dpo-v1-4-nemo-c1000_v2
family_friendly_score: 0.0
formatter: {'memory_template': '[INST]system\n{memory}[/INST]\n', 'prompt_template': '[INST]user\n{prompt}[/INST]\n', 'bot_template': '[INST]assistant\n{bot_name}: {message}[/INST]\n', 'user_template': '[INST]user\n{user_name}: {message}[/INST]\n', 'response_template': '[INST]assistant\n{bot_name}:', 'truncate_by_message': False}
generation_params: {'temperature': 0.75, 'top_p': 1.0, 'min_p': 0.1, 'top_k': 40, 'presence_penalty': 0.0, 'frequency_penalty': 0.0, 'stopping_words': ['\n', '[/INST]'], 'max_input_tokens': 512, 'best_of': 8, 'max_output_tokens': 64}
gpu_counts: {'NVIDIA RTX A5000': 1}
is_internal_developer: False
language_model: jic062/dpo-v1.4-Nemo-c1000
latencies: [{'batch_size': 1, 'throughput': 0.698830034959821, 'latency_mean': 1.4308720886707307, 'latency_p50': 1.4335582256317139, 'latency_p90': 1.5987879991531373}, {'batch_size': 3, 'throughput': 1.3473500716661029, 'latency_mean': 2.222927585840225, 'latency_p50': 2.2189888954162598, 'latency_p90': 2.515632939338684}, {'batch_size': 5, 'throughput': 1.5758910503240149, 'latency_mean': 3.1568307888507845, 'latency_p50': 3.1903942823410034, 'latency_p90': 3.558679485321045}, {'batch_size': 6, 'throughput': 1.6419792710796557, 'latency_mean': 3.628674541711807, 'latency_p50': 3.6411460638046265, 'latency_p90': 4.125023198127747}, {'batch_size': 8, 'throughput': 1.6063265745441317, 'latency_mean': 4.945366599559784, 'latency_p50': 4.978110909461975, 'latency_p90': 5.5822220087051395}, {'batch_size': 10, 'throughput': 1.5551052577938866, 'latency_mean': 6.374917727708817, 'latency_p50': 6.3816728591918945, 'latency_p90': 7.306980586051941}]
max_input_tokens: 512
max_output_tokens: 64
model_architecture: MistralForCausalLM
model_group: jic062/dpo-v1.4-Nemo-c10
model_name: jic062-dpo-v1-4-nemo-c1000_v2
model_num_parameters: 12772070400.0
model_repo: jic062/dpo-v1.4-Nemo-c1000
model_size: 13B
num_battles: 12065
num_wins: 6237
ranking_group: single
status: torndown
submission_type: basic
throughput_3p7s: 1.65
timestamp: 2024-09-13T04:11:57+00:00
us_pacific_date: 2024-09-12
win_ratio: 0.5169498549523415
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 jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer
Waiting for job on jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer to finish
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ╔═════════════════════════════════════════════════════════════════════╗
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ _____ __ __ ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ / _/ /_ ___ __/ / ___ ___ / / ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ / _/ / // / |/|/ / _ \/ -_) -_) / ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ /_//_/\_, /|__,__/_//_/\__/\__/_/ ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ /___/ ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ Version: 0.10.1 ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ Copyright 2023 MK ONE TECHNOLOGIES Inc. ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ https://mk1.ai ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ The license key for the current software has been verified as ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ belonging to: ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ Chai Research Corp. ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ Account ID: 7997a29f-0ceb-4cc7-9adf-840c57b4ae6f ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ Expiration: 2024-10-15 23:59:59 ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ║ ║
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: ╚═════════════════════════════════════════════════════════════════════╝
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: Downloaded to shared memory in 34.827s
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: quantizing model to /dev/shm/model_cache, profile:s0, folder:/tmp/tmpl2rjq8aw, device:0
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: Saving flywheel model at /dev/shm/model_cache
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: quantized model in 36.806s
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: Processed model jic062/dpo-v1.4-Nemo-c1000 in 71.633s
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: creating bucket guanaco-mkml-models
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: Bucket 's3://guanaco-mkml-models/' created
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: uploading /dev/shm/model_cache to s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-c1000-v2
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: cp /dev/shm/model_cache/config.json s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-c1000-v2/config.json
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: cp /dev/shm/model_cache/special_tokens_map.json s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-c1000-v2/special_tokens_map.json
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: cp /dev/shm/model_cache/tokenizer_config.json s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-c1000-v2/tokenizer_config.json
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: cp /dev/shm/model_cache/tokenizer.json s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-c1000-v2/tokenizer.json
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: cp /dev/shm/model_cache/flywheel_model.0.safetensors s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-c1000-v2/flywheel_model.0.safetensors
jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer: Loading 0: 0%| | 0/363 [00:00<?, ?it/s] Loading 0: 1%|▏ | 5/363 [00:00<00:11, 31.14it/s] Loading 0: 4%|▎ | 13/363 [00:00<00:06, 50.86it/s] Loading 0: 5%|▌ | 19/363 [00:00<00:07, 45.69it/s] Loading 0: 7%|▋ | 24/363 [00:00<00:07, 43.95it/s] Loading 0: 9%|▊ | 31/363 [00:00<00:06, 50.09it/s] Loading 0: 10%|█ | 37/363 [00:00<00:07, 45.94it/s] Loading 0: 12%|█▏ | 42/363 [00:00<00:07, 43.14it/s] Loading 0: 13%|█▎ | 48/363 [00:01<00:06, 45.21it/s] Loading 0: 15%|█▍ | 53/363 [00:01<00:07, 44.27it/s] Loading 0: 17%|█▋ | 60/363 [00:01<00:06, 46.13it/s] Loading 0: 18%|█▊ | 65/363 [00:01<00:09, 31.09it/s] Loading 0: 20%|█▉ | 72/363 [00:01<00:07, 38.08it/s] Loading 0: 21%|██▏ | 78/363 [00:01<00:07, 39.43it/s] Loading 0: 23%|██▎ | 83/363 [00:02<00:07, 38.19it/s] Loading 0: 25%|██▍ | 89/363 [00:02<00:06, 42.37it/s] Loading 0: 26%|██▌ | 94/363 [00:02<00:06, 40.10it/s] Loading 0: 27%|██▋ | 99/363 [00:02<00:06, 39.37it/s] Loading 0: 29%|██▊ | 104/363 [00:02<00:06, 41.05it/s] Loading 0: 30%|███ | 110/363 [00:02<00:06, 39.80it/s] Loading 0: 32%|███▏ | 115/363 [00:02<00:06, 40.49it/s] Loading 0: 33%|███▎ | 120/363 [00:02<00:06, 38.24it/s] Loading 0: 34%|███▍ | 125/363 [00:03<00:05, 40.60it/s] Loading 0: 36%|███▌ | 130/363 [00:03<00:05, 39.94it/s] Loading 0: 37%|███▋ | 135/363 [00:03<00:05, 38.81it/s] Loading 0: 39%|███▊ | 140/363 [00:03<00:05, 41.14it/s] Loading 0: 40%|███▉ | 145/363 [00:03<00:08, 26.19it/s] Loading 0: 41%|████ | 149/363 [00:03<00:08, 26.33it/s] Loading 0: 43%|████▎ | 156/363 [00:04<00:06, 33.39it/s] Loading 0: 44%|████▍ | 160/363 [00:04<00:06, 33.79it/s] Loading 0: 45%|████▌ | 165/363 [00:04<00:05, 37.09it/s] Loading 0: 47%|████▋ | 170/363 [00:04<00:05, 37.27it/s] Loading 0: 48%|████▊ | 175/363 [00:04<00:04, 38.65it/s] Loading 0: 50%|████▉ | 180/363 [00:04<00:04, 39.39it/s] Loading 0: 51%|█████ | 185/363 [00:04<00:05, 32.30it/s] Loading 0: 53%|█████▎ | 192/363 [00:04<00:04, 38.63it/s] Loading 0: 54%|█████▍ | 197/363 [00:05<00:04, 39.51it/s] Loading 0: 56%|█████▌ | 202/363 [00:05<00:03, 41.24it/s] Loading 0: 57%|█████▋ | 207/363 [00:05<00:03, 41.71it/s] Loading 0: 58%|█████▊ | 212/363 [00:05<00:04, 33.70it/s] Loading 0: 61%|██████ | 220/363 [00:05<00:03, 43.42it/s] Loading 0: 62%|██████▏ | 225/363 [00:05<00:04, 28.23it/s] Loading 0: 63%|██████▎ | 230/363 [00:06<00:04, 29.48it/s] Loading 0: 65%|██████▌ | 237/363 [00:06<00:03, 35.61it/s] Loading 0: 67%|██████▋ | 242/363 [00:06<00:03, 36.98it/s] Loading 0: 68%|██████▊ | 247/363 [00:06<00:03, 36.83it/s] Loading 0: 69%|██████▉ | 252/363 [00:06<00:02, 38.74it/s] Loading 0: 71%|███████ | 257/363 [00:06<00:03, 32.82it/s] Loading 0: 73%|███████▎ | 264/363 [00:06<00:02, 40.49it/s] Loading 0: 74%|███████▍ | 269/363 [00:07<00:02, 39.61it/s] Loading 0: 75%|███████▌ | 274/363 [00:07<00:02, 40.20it/s] Loading 0: 77%|███████▋ | 279/363 [00:07<00:01, 42.57it/s] Loading 0: 78%|███████▊ | 284/363 [00:07<00:02, 35.94it/s] Loading 0: 80%|████████ | 291/363 [00:07<00:01, 41.22it/s] Loading 0: 82%|████████▏ | 296/363 [00:07<00:01, 40.72it/s] Loading 0: 83%|████████▎ | 301/363 [00:07<00:01, 42.69it/s] Loading 0: 84%|████████▍ | 306/363 [00:14<00:23, 2.45it/s] Loading 0: 85%|████████▌ | 310/363 [00:14<00:16, 3.17it/s] Loading 0: 87%|████████▋ | 314/363 [00:14<00:11, 4.15it/s] Loading 0: 88%|████████▊ | 319/363 [00:15<00:07, 5.84it/s] Loading 0: 89%|████████▉ | 323/363 [00:15<00:05, 7.54it/s] Loading 0: 91%|█████████ | 329/363 [00:15<00:03, 11.01it/s] Loading 0: 92%|█████████▏| 334/363 [00:15<00:02, 14.29it/s] Loading 0: 93%|█████████▎| 339/363 [00:15<00:01, 16.63it/s] Loading 0: 96%|█████████▌| 347/363 [00:15<00:00, 24.18it/s] Loading 0: 97%|█████████▋| 353/363 [00:15<00:00, 26.78it/s] Loading 0: 99%|█████████▊| 358/363 [00:15<00:00, 28.89it/s]
Job jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer completed after 93.93s with status: succeeded
Stopping job with name jic062-dpo-v1-4-nemo-c1000-v2-mkmlizer
Pipeline stage MKMLizer completed in 95.29s
run pipeline stage %s
Running pipeline stage MKMLTemplater
Pipeline stage MKMLTemplater completed in 0.08s
run pipeline stage %s
Running pipeline stage MKMLDeployer
Creating inference service jic062-dpo-v1-4-nemo-c1000-v2
Waiting for inference service jic062-dpo-v1-4-nemo-c1000-v2 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
Inference service jic062-dpo-v1-4-nemo-c1000-v2 ready after 181.60020351409912s
Pipeline stage MKMLDeployer completed in 181.94s
run pipeline stage %s
Running pipeline stage StressChecker
Received healthy response to inference request in 3.8354203701019287s
Received healthy response to inference request in 1.9880354404449463s
Received healthy response to inference request in 2.5826878547668457s
Received healthy response to inference request in 1.8155813217163086s
Received healthy response to inference request in 2.22796630859375s
5 requests
0 failed requests
5th percentile: 1.8500721454620361
10th percentile: 1.8845629692077637
20th percentile: 1.9535446166992188
30th percentile: 2.036021614074707
40th percentile: 2.1319939613342287
50th percentile: 2.22796630859375
60th percentile: 2.3698549270629883
70th percentile: 2.5117435455322266
80th percentile: 2.8332343578338626
90th percentile: 3.334327363967896
95th percentile: 3.584873867034912
99th percentile: 3.7853110694885252
mean time: 2.489938259124756
Pipeline stage StressChecker completed in 13.23s
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 5.32s
Shutdown handler de-registered
jic062-dpo-v1-4-nemo-c1000_v2 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 0.18s
run pipeline stage %s
Running pipeline stage MKMLProfilerTemplater
Pipeline stage MKMLProfilerTemplater completed in 0.11s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeployer
Creating inference service jic062-dpo-v1-4-nemo-c1000-v2-profiler
Waiting for inference service jic062-dpo-v1-4-nemo-c1000-v2-profiler to be ready
Inference service jic062-dpo-v1-4-nemo-c1000-v2-profiler ready after 180.43208241462708s
Pipeline stage MKMLProfilerDeployer completed in 182.05s
run pipeline stage %s
Running pipeline stage MKMLProfilerRunner
kubectl cp /code/guanaco/guanaco_inference_services/src/inference_scripts tenant-chaiml-guanaco/jic062-dpo-v1-4-nemo11bc1fd75fab72ad1ab765f44f829a69-deplombgjk:/code/chaiverse_profiler_1726201234 --namespace tenant-chaiml-guanaco
kubectl exec -it jic062-dpo-v1-4-nemo11bc1fd75fab72ad1ab765f44f829a69-deplombgjk --namespace tenant-chaiml-guanaco -- sh -c 'cd /code/chaiverse_profiler_1726201234 && python profiles.py profile --best_of_n 8 --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 512 --output_tokens 64 --summary /code/chaiverse_profiler_1726201234/summary.json'
kubectl exec -it jic062-dpo-v1-4-nemo11bc1fd75fab72ad1ab765f44f829a69-deplombgjk --namespace tenant-chaiml-guanaco -- bash -c 'cat /code/chaiverse_profiler_1726201234/summary.json'
Pipeline stage MKMLProfilerRunner completed in 943.45s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeleter
Checking if service jic062-dpo-v1-4-nemo-c1000-v2-profiler is running
Tearing down inference service jic062-dpo-v1-4-nemo-c1000-v2-profiler
Service jic062-dpo-v1-4-nemo-c1000-v2-profiler has been torndown
Pipeline stage MKMLProfilerDeleter completed in 2.07s
Shutdown handler de-registered
jic062-dpo-v1-4-nemo-c1000_v2 status is now inactive due to auto deactivation removed underperforming models
jic062-dpo-v1-4-nemo-c1000_v2 status is now torndown due to DeploymentManager action