submission_id: jic062-dpo-v1-4-nemo_v3
developer_uid: chace9580
best_of: 8
celo_rating: 1272.1
display_name: jic062-dpo-v1-4-nemo_v3
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': 1.0, 'top_p': 0.9, 'min_p': 0.05, 'top_k': 80, 'presence_penalty': 0.0, 'frequency_penalty': 0.0, 'stopping_words': ['\n', '/s', '[/INST]'], 'max_input_tokens': 1024, 'best_of': 8, 'max_output_tokens': 64}
gpu_counts: {'NVIDIA RTX A5000': 1}
is_internal_developer: False
language_model: jic062/dpo-v1.4-Nemo
latencies: [{'batch_size': 1, 'throughput': 0.61435386736416, 'latency_mean': 1.6276296544075013, 'latency_p50': 1.61992609500885, 'latency_p90': 1.7931848526000977}, {'batch_size': 3, 'throughput': 1.079684960054848, 'latency_mean': 2.7676103460788726, 'latency_p50': 2.784646153450012, 'latency_p90': 3.0163015604019163}, {'batch_size': 5, 'throughput': 1.2346955034286031, 'latency_mean': 4.033249439001083, 'latency_p50': 4.036059141159058, 'latency_p90': 4.610881614685058}, {'batch_size': 6, 'throughput': 1.2707817307874634, 'latency_mean': 4.700254266262054, 'latency_p50': 4.6685097217559814, 'latency_p90': 5.306362676620483}, {'batch_size': 8, 'throughput': 1.2522308821348949, 'latency_mean': 6.358148107528686, 'latency_p50': 6.412910580635071, 'latency_p90': 7.145080280303955}, {'batch_size': 10, 'throughput': 1.2215308257296889, 'latency_mean': 8.140109139680863, 'latency_p50': 8.188225507736206, 'latency_p90': 9.250119423866272}]
max_input_tokens: 1024
max_output_tokens: 64
model_architecture: MistralForCausalLM
model_group: jic062/dpo-v1.4-Nemo
model_name: jic062-dpo-v1-4-nemo_v3
model_num_parameters: 12772070400.0
model_repo: jic062/dpo-v1.4-Nemo
model_size: 13B
num_battles: 13006
num_wins: 6879
ranking_group: single
status: torndown
submission_type: basic
throughput_3p7s: 1.21
timestamp: 2024-09-20T05:03:44+00:00
us_pacific_date: 2024-09-19
win_ratio: 0.5289097339689374
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 jic062-dpo-v1-4-nemo-v3-mkmlizer
Waiting for job on jic062-dpo-v1-4-nemo-v3-mkmlizer to finish
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
jic062-dpo-v1-4-nemo-v3-mkmlizer: ╔═════════════════════════════════════════════════════════════════════╗
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ _____ __ __ ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ / _/ /_ ___ __/ / ___ ___ / / ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ / _/ / // / |/|/ / _ \/ -_) -_) / ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ /_//_/\_, /|__,__/_//_/\__/\__/_/ ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ /___/ ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ Version: 0.10.1 ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ Copyright 2023 MK ONE TECHNOLOGIES Inc. ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ https://mk1.ai ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ The license key for the current software has been verified as ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ belonging to: ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ Chai Research Corp. ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ Account ID: 7997a29f-0ceb-4cc7-9adf-840c57b4ae6f ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ Expiration: 2024-10-15 23:59:59 ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ║ ║
jic062-dpo-v1-4-nemo-v3-mkmlizer: ╚═════════════════════════════════════════════════════════════════════╝
Connection pool is full, discarding connection: %s. Connection pool size: %s
jic062-dpo-v1-4-nemo-v3-mkmlizer: Downloaded to shared memory in 52.138s
jic062-dpo-v1-4-nemo-v3-mkmlizer: quantizing model to /dev/shm/model_cache, profile:s0, folder:/tmp/tmpdnnhdbmf, device:0
jic062-dpo-v1-4-nemo-v3-mkmlizer: Saving flywheel model at /dev/shm/model_cache
jic062-dpo-v1-4-nemo-v3-mkmlizer: quantized model in 35.526s
jic062-dpo-v1-4-nemo-v3-mkmlizer: Processed model jic062/dpo-v1.4-Nemo in 87.665s
jic062-dpo-v1-4-nemo-v3-mkmlizer: creating bucket guanaco-mkml-models
jic062-dpo-v1-4-nemo-v3-mkmlizer: Bucket 's3://guanaco-mkml-models/' created
jic062-dpo-v1-4-nemo-v3-mkmlizer: uploading /dev/shm/model_cache to s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-v3
jic062-dpo-v1-4-nemo-v3-mkmlizer: cp /dev/shm/model_cache/special_tokens_map.json s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-v3/special_tokens_map.json
jic062-dpo-v1-4-nemo-v3-mkmlizer: cp /dev/shm/model_cache/tokenizer_config.json s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-v3/tokenizer_config.json
jic062-dpo-v1-4-nemo-v3-mkmlizer: cp /dev/shm/model_cache/tokenizer.json s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-v3/tokenizer.json
jic062-dpo-v1-4-nemo-v3-mkmlizer: cp /dev/shm/model_cache/flywheel_model.0.safetensors s3://guanaco-mkml-models/jic062-dpo-v1-4-nemo-v3/flywheel_model.0.safetensors
jic062-dpo-v1-4-nemo-v3-mkmlizer: Loading 0: 0%| | 0/363 [00:00<?, ?it/s] Loading 0: 1%|▏ | 5/363 [00:00<00:10, 34.06it/s] Loading 0: 4%|▎ | 13/363 [00:00<00:06, 55.49it/s] Loading 0: 5%|▌ | 19/363 [00:00<00:07, 48.83it/s] Loading 0: 7%|▋ | 25/363 [00:00<00:06, 49.51it/s] Loading 0: 9%|▊ | 31/363 [00:00<00:06, 52.41it/s] Loading 0: 10%|█ | 37/363 [00:00<00:06, 49.44it/s] Loading 0: 12%|█▏ | 43/363 [00:00<00:06, 48.31it/s] Loading 0: 13%|█▎ | 49/363 [00:00<00:06, 51.28it/s] Loading 0: 15%|█▌ | 55/363 [00:01<00:06, 49.62it/s] Loading 0: 17%|█▋ | 61/363 [00:01<00:08, 36.91it/s] Loading 0: 18%|█▊ | 66/363 [00:01<00:07, 37.29it/s] Loading 0: 20%|█▉ | 71/363 [00:01<00:07, 39.98it/s] Loading 0: 21%|██ | 76/363 [00:01<00:07, 40.56it/s] Loading 0: 23%|██▎ | 82/363 [00:01<00:07, 38.82it/s] Loading 0: 25%|██▍ | 90/363 [00:02<00:05, 46.24it/s] Loading 0: 26%|██▌ | 95/363 [00:02<00:05, 46.37it/s] Loading 0: 28%|██▊ | 100/363 [00:02<00:06, 39.04it/s] Loading 0: 30%|███ | 109/363 [00:02<00:05, 50.36it/s] Loading 0: 32%|███▏ | 115/363 [00:02<00:05, 47.33it/s] Loading 0: 33%|███▎ | 121/363 [00:02<00:05, 45.11it/s] Loading 0: 35%|███▍ | 126/363 [00:02<00:05, 44.47it/s] Loading 0: 36%|███▋ | 132/363 [00:02<00:05, 42.57it/s] Loading 0: 38%|███▊ | 137/363 [00:03<00:05, 41.27it/s] Loading 0: 39%|███▉ | 142/363 [00:03<00:06, 34.54it/s] Loading 0: 40%|████ | 146/363 [00:03<00:06, 35.01it/s] Loading 0: 41%|████▏ | 150/363 [00:03<00:06, 33.19it/s] Loading 0: 43%|████▎ | 156/363 [00:03<00:05, 39.06it/s] Loading 0: 44%|████▍ | 161/363 [00:03<00:05, 39.70it/s] Loading 0: 46%|████▌ | 166/363 [00:03<00:04, 41.35it/s] Loading 0: 47%|████▋ | 171/363 [00:03<00:04, 42.22it/s] Loading 0: 48%|████▊ | 176/363 [00:04<00:05, 36.91it/s] Loading 0: 51%|█████ | 184/363 [00:04<00:03, 45.36it/s] Loading 0: 52%|█████▏ | 189/363 [00:04<00:03, 46.49it/s] Loading 0: 53%|█████▎ | 194/363 [00:04<00:04, 38.25it/s] Loading 0: 55%|█████▌ | 201/363 [00:04<00:03, 43.28it/s] Loading 0: 57%|█████▋ | 206/363 [00:04<00:03, 41.55it/s] Loading 0: 58%|█████▊ | 211/363 [00:04<00:03, 40.59it/s] Loading 0: 60%|█████▉ | 216/363 [00:05<00:03, 41.97it/s] Loading 0: 61%|██████ | 222/363 [00:05<00:03, 41.57it/s] Loading 0: 63%|██████▎ | 227/363 [00:05<00:04, 32.07it/s] Loading 0: 64%|██████▎ | 231/363 [00:05<00:04, 31.79it/s] Loading 0: 66%|██████▌ | 238/363 [00:05<00:03, 38.24it/s] Loading 0: 67%|██████▋ | 243/363 [00:05<00:03, 39.75it/s] Loading 0: 68%|██████▊ | 248/363 [00:06<00:03, 35.54it/s] Loading 0: 70%|███████ | 255/363 [00:06<00:02, 41.90it/s] Loading 0: 72%|███████▏ | 260/363 [00:06<00:02, 42.38it/s] Loading 0: 73%|███████▎ | 266/363 [00:06<00:02, 40.19it/s] Loading 0: 75%|███████▌ | 274/363 [00:06<00:01, 47.44it/s] Loading 0: 77%|███████▋ | 279/363 [00:06<00:01, 46.84it/s] Loading 0: 78%|███████▊ | 284/363 [00:06<00:01, 40.62it/s] Loading 0: 80%|████████ | 291/363 [00:06<00:01, 45.66it/s] Loading 0: 82%|████████▏ | 296/363 [00:07<00:01, 46.04it/s] Loading 0: 83%|████████▎ | 302/363 [00:07<00:01, 49.57it/s] Loading 0: 85%|████████▍ | 308/363 [00:13<00:20, 2.75it/s] Loading 0: 86%|████████▌ | 312/363 [00:14<00:14, 3.47it/s] Loading 0: 88%|████████▊ | 320/363 [00:14<00:07, 5.54it/s] Loading 0: 90%|████████▉ | 326/363 [00:14<00:04, 7.45it/s] Loading 0: 91%|█████████ | 331/363 [00:14<00:03, 9.47it/s] Loading 0: 93%|█████████▎| 339/363 [00:14<00:01, 13.36it/s] Loading 0: 96%|█████████▌| 347/363 [00:14<00:00, 18.60it/s] Loading 0: 97%|█████████▋| 353/363 [00:14<00:00, 22.25it/s] Loading 0: 99%|█████████▉| 359/363 [00:15<00:00, 26.42it/s]
Job jic062-dpo-v1-4-nemo-v3-mkmlizer completed after 108.32s with status: succeeded
Stopping job with name jic062-dpo-v1-4-nemo-v3-mkmlizer
Pipeline stage MKMLizer completed in 110.08s
run pipeline stage %s
Running pipeline stage MKMLTemplater
Pipeline stage MKMLTemplater completed in 0.20s
run pipeline stage %s
Running pipeline stage MKMLDeployer
Creating inference service jic062-dpo-v1-4-nemo-v3
Waiting for inference service jic062-dpo-v1-4-nemo-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
Failed to get response for submission chaiml-lexical-nemo-v4-1k1e5_v3: ('http://chaiml-lexical-nemo-v4-1k1e5-v3-predictor.tenant-chaiml-guanaco.k2.chaiverse.com/v1/models/GPT-J-6B-lit-v2:predict', 'read tcp 127.0.0.1:41022->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
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
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-v3 ready after 211.54591393470764s
Pipeline stage MKMLDeployer completed in 212.62s
run pipeline stage %s
Running pipeline stage StressChecker
Received healthy response to inference request in 2.321613073348999s
Failed to get response for submission blend_hahit_2024-09-20: ('http://zonemercy-virgo-edit-v1-1e5-v13-predictor.tenant-chaiml-guanaco.k2.chaiverse.com/v1/models/GPT-J-6B-lit-v2:predict', 'read tcp 127.0.0.1:43160->127.0.0.1:8080: read: connection reset by peer\n')
Received healthy response to inference request in 3.0731418132781982s
Received healthy response to inference request in 1.785426139831543s
Received healthy response to inference request in 1.9041829109191895s
Received healthy response to inference request in 2.6333110332489014s
5 requests
0 failed requests
5th percentile: 1.8091774940490724
10th percentile: 1.8329288482666015
20th percentile: 1.88043155670166
30th percentile: 1.9876689434051513
40th percentile: 2.1546410083770753
50th percentile: 2.321613073348999
60th percentile: 2.44629225730896
70th percentile: 2.5709714412689206
80th percentile: 2.721277189254761
90th percentile: 2.8972095012664796
95th percentile: 2.9851756572723387
99th percentile: 3.0555485820770265
mean time: 2.343534994125366
Pipeline stage StressChecker completed in 12.70s
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 5.79s
Shutdown handler de-registered
jic062-dpo-v1-4-nemo_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 0.16s
run pipeline stage %s
Running pipeline stage MKMLProfilerTemplater
Pipeline stage MKMLProfilerTemplater completed in 0.15s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeployer
Creating inference service jic062-dpo-v1-4-nemo-v3-profiler
Waiting for inference service jic062-dpo-v1-4-nemo-v3-profiler to be ready
Inference service jic062-dpo-v1-4-nemo-v3-profiler ready after 210.4875328540802s
Pipeline stage MKMLProfilerDeployer completed in 210.90s
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-nemo-v3-profiler-predictor-00001-deploymenpfbd8:/code/chaiverse_profiler_1726809224 --namespace tenant-chaiml-guanaco
kubectl exec -it jic062-dpo-v1-4-nemo-v3-profiler-predictor-00001-deploymenpfbd8 --namespace tenant-chaiml-guanaco -- sh -c 'cd /code/chaiverse_profiler_1726809224 && 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 1024 --output_tokens 64 --summary /code/chaiverse_profiler_1726809224/summary.json'
kubectl exec -it jic062-dpo-v1-4-nemo-v3-profiler-predictor-00001-deploymenpfbd8 --namespace tenant-chaiml-guanaco -- bash -c 'cat /code/chaiverse_profiler_1726809224/summary.json'
Pipeline stage MKMLProfilerRunner completed in 1158.81s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeleter
Checking if service jic062-dpo-v1-4-nemo-v3-profiler is running
Tearing down inference service jic062-dpo-v1-4-nemo-v3-profiler
Service jic062-dpo-v1-4-nemo-v3-profiler has been torndown
Pipeline stage MKMLProfilerDeleter completed in 2.14s
Shutdown handler de-registered
jic062-dpo-v1-4-nemo_v3 status is now inactive due to auto deactivation removed underperforming models
jic062-dpo-v1-4-nemo_v3 status is now torndown due to DeploymentManager action