riverise-0910-2005-sft

submission_id: riverise-0910-2005-sft_v1

developer_uid: Riverise

best_of: 16

celo_rating: 1250.57

display_name: riverise-0910-2005-sft_v1

family_friendly_score: 0.0

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': 1.15, 'top_p': 0.95, 'min_p': 0.0, 'top_k': 40, 'presence_penalty': 0.0, 'frequency_penalty': 0.0, 'stopping_words': ['\n'], 'max_input_tokens': 512, 'best_of': 16, 'max_output_tokens': 64}

gpu_counts: {'NVIDIA RTX A5000': 1}

is_internal_developer: False

language_model: Riverise/0910_2005_sft

latencies: [{'batch_size': 1, 'throughput': 0.9126851073418217, 'latency_mean': 1.095582489967346, 'latency_p50': 1.0996582508087158, 'latency_p90': 1.2246589660644531}, {'batch_size': 4, 'throughput': 1.8124047146727709, 'latency_mean': 2.1968948543071747, 'latency_p50': 2.1941298246383667, 'latency_p90': 2.4720942497253415}, {'batch_size': 5, 'throughput': 1.869959903234761, 'latency_mean': 2.667214788198471, 'latency_p50': 2.685762405395508, 'latency_p90': 2.9580538988113405}, {'batch_size': 8, 'throughput': 1.9962490229636836, 'latency_mean': 3.9810542571544647, 'latency_p50': 3.9866278171539307, 'latency_p90': 4.4572797298431395}, {'batch_size': 10, 'throughput': 2.039232490176434, 'latency_mean': 4.85510537981987, 'latency_p50': 4.805861115455627, 'latency_p90': 5.621788549423218}, {'batch_size': 12, 'throughput': 2.024820540596493, 'latency_mean': 5.85373482465744, 'latency_p50': 5.958730697631836, 'latency_p90': 6.688148880004883}, {'batch_size': 15, 'throughput': 2.045176503018188, 'latency_mean': 7.185724745988846, 'latency_p50': 7.270509362220764, 'latency_p90': 8.067982029914855}]

max_input_tokens: 512

max_output_tokens: 64

model_architecture: LlamaForCausalLM

model_group: Riverise/0910_2005_sft

model_name: riverise-0910-2005-sft_v1

model_num_parameters: 8030261248.0

model_repo: Riverise/0910_2005_sft

model_size: 8B

num_battles: 10350

num_wins: 5314

ranking_group: single

status: torndown

submission_type: basic

throughput_3p7s: 1.99

timestamp: 2024-09-11T05:04:03+00:00

us_pacific_date: 2024-09-10

win_ratio: 0.5134299516908213

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 riverise-0910-2005-sft-v1-mkmlizer
Waiting for job on riverise-0910-2005-sft-v1-mkmlizer to finish
riverise-0910-2005-sft-v1-mkmlizer: ╔═════════════════════════════════════════════════════════════════════╗
riverise-0910-2005-sft-v1-mkmlizer: ║     _____            __           __                                ║
riverise-0910-2005-sft-v1-mkmlizer: ║    / _/ /_ ___    __/ /  ___ ___ / /                                ║
riverise-0910-2005-sft-v1-mkmlizer: ║   / _/ / // / |/|/ / _ \/ -_) -_) /                                 ║
riverise-0910-2005-sft-v1-mkmlizer: ║  /_//_/\_, /|__,__/_//_/\__/\__/_/                                  ║
riverise-0910-2005-sft-v1-mkmlizer: ║       /___/                                                         ║
riverise-0910-2005-sft-v1-mkmlizer: ║                                                                     ║
riverise-0910-2005-sft-v1-mkmlizer: ║  Version: 0.10.1                                                    ║
riverise-0910-2005-sft-v1-mkmlizer: ║  Copyright 2023 MK ONE TECHNOLOGIES Inc.                            ║
riverise-0910-2005-sft-v1-mkmlizer: ║  https://mk1.ai                                                     ║
riverise-0910-2005-sft-v1-mkmlizer: ║                                                                     ║
riverise-0910-2005-sft-v1-mkmlizer: ║  The license key for the current software has been verified as      ║
riverise-0910-2005-sft-v1-mkmlizer: ║  belonging to:                                                      ║
riverise-0910-2005-sft-v1-mkmlizer: ║                                                                     ║
riverise-0910-2005-sft-v1-mkmlizer: ║  Chai Research Corp.                                                ║
riverise-0910-2005-sft-v1-mkmlizer: ║  Account ID: 7997a29f-0ceb-4cc7-9adf-840c57b4ae6f                   ║
riverise-0910-2005-sft-v1-mkmlizer: ║  Expiration: 2024-10-15 23:59:59                                    ║
riverise-0910-2005-sft-v1-mkmlizer: ║                                                                     ║
riverise-0910-2005-sft-v1-mkmlizer: ╚═════════════════════════════════════════════════════════════════════╝
riverise-0910-2005-sft-v1-mkmlizer: Downloaded to shared memory in 34.147s
riverise-0910-2005-sft-v1-mkmlizer: quantizing model to /dev/shm/model_cache, profile:s0, folder:/tmp/tmp7bkc0yfn, device:0
riverise-0910-2005-sft-v1-mkmlizer: Saving flywheel model at /dev/shm/model_cache
riverise-0910-2005-sft-v1-mkmlizer: quantized model in 25.683s
riverise-0910-2005-sft-v1-mkmlizer: Processed model Riverise/0910_2005_sft in 59.830s
riverise-0910-2005-sft-v1-mkmlizer: creating bucket guanaco-mkml-models
riverise-0910-2005-sft-v1-mkmlizer: Bucket 's3://guanaco-mkml-models/' created
riverise-0910-2005-sft-v1-mkmlizer: uploading /dev/shm/model_cache to s3://guanaco-mkml-models/riverise-0910-2005-sft-v1
riverise-0910-2005-sft-v1-mkmlizer: cp /dev/shm/model_cache/special_tokens_map.json s3://guanaco-mkml-models/riverise-0910-2005-sft-v1/special_tokens_map.json
riverise-0910-2005-sft-v1-mkmlizer: cp /dev/shm/model_cache/config.json s3://guanaco-mkml-models/riverise-0910-2005-sft-v1/config.json
riverise-0910-2005-sft-v1-mkmlizer: cp /dev/shm/model_cache/tokenizer_config.json s3://guanaco-mkml-models/riverise-0910-2005-sft-v1/tokenizer_config.json
riverise-0910-2005-sft-v1-mkmlizer: cp /dev/shm/model_cache/tokenizer.json s3://guanaco-mkml-models/riverise-0910-2005-sft-v1/tokenizer.json
Connection pool is full, discarding connection: %s. Connection pool size: %s
riverise-0910-2005-sft-v1-mkmlizer: cp /dev/shm/model_cache/flywheel_model.0.safetensors s3://guanaco-mkml-models/riverise-0910-2005-sft-v1/flywheel_model.0.safetensors
riverise-0910-2005-sft-v1-mkmlizer: 
Loading 0:   0%|          | 0/291 [00:00<?, ?it/s]
Loading 0:   2%|▏         | 7/291 [00:00<00:05, 54.50it/s]
Loading 0:   8%|▊         | 22/291 [00:00<00:03, 87.58it/s]
Loading 0:  11%|█         | 31/291 [00:00<00:02, 86.91it/s]
Loading 0:  14%|█▍        | 42/291 [00:00<00:02, 94.94it/s]
Loading 0:  18%|█▊        | 52/291 [00:00<00:03, 76.03it/s]
Loading 0:  21%|██        | 61/291 [00:00<00:02, 77.99it/s]
Loading 0:  24%|██▍       | 70/291 [00:00<00:02, 80.83it/s]
Loading 0:  27%|██▋       | 80/291 [00:00<00:02, 85.92it/s]
Loading 0:  31%|███       | 89/291 [00:02<00:09, 20.29it/s]
Loading 0:  33%|███▎      | 97/291 [00:02<00:07, 25.44it/s]
Loading 0:  36%|███▋      | 106/291 [00:02<00:05, 32.52it/s]
Loading 0:  42%|████▏     | 121/291 [00:02<00:03, 45.97it/s]
Loading 0:  46%|████▌     | 133/291 [00:02<00:02, 53.91it/s]
Loading 0:  49%|████▉     | 142/291 [00:02<00:02, 58.36it/s]
Loading 0:  52%|█████▏    | 151/291 [00:02<00:02, 61.46it/s]
Loading 0:  55%|█████▍    | 160/291 [00:03<00:01, 66.70it/s]
Loading 0:  58%|█████▊    | 170/291 [00:03<00:01, 73.72it/s]
Loading 0:  62%|██████▏   | 179/291 [00:03<00:01, 75.63it/s]
Loading 0:  65%|██████▍   | 188/291 [00:04<00:04, 22.10it/s]
Loading 0:  69%|██████▉   | 202/291 [00:04<00:02, 31.46it/s]
Loading 0:  73%|███████▎  | 211/291 [00:04<00:02, 37.43it/s]
Loading 0:  76%|███████▌  | 220/291 [00:04<00:01, 44.18it/s]
Loading 0:  79%|███████▊  | 229/291 [00:04<00:01, 51.50it/s]
Loading 0:  82%|████████▏ | 238/291 [00:04<00:00, 57.17it/s]
Loading 0:  85%|████████▍ | 247/291 [00:05<00:00, 61.63it/s]
Loading 0:  88%|████████▊ | 256/291 [00:05<00:00, 62.86it/s]
Loading 0:  92%|█████████▏| 267/291 [00:05<00:00, 73.20it/s]
Loading 0:  95%|█████████▌| 277/291 [00:05<00:00, 73.46it/s]
Loading 0:  99%|█████████▊| 287/291 [00:05<00:00, 45.03it/s]
                                                            
Job riverise-0910-2005-sft-v1-mkmlizer completed after 83.45s with status: succeeded
Stopping job with name riverise-0910-2005-sft-v1-mkmlizer
Pipeline stage MKMLizer completed in 84.70s
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 riverise-0910-2005-sft-v1
Waiting for inference service riverise-0910-2005-sft-v1 to be ready
Failed to get response for submission blend_siken_2024-09-09: ('http://chaiml-llama-8b-pairwis-8189-v19-predictor.tenant-chaiml-guanaco.k.chaiverse.com/v1/models/GPT-J-6B-lit-v2:predict', 'read tcp 127.0.0.1:42102->127.0.0.1:8080: read: connection reset by peer\n')
Inference service riverise-0910-2005-sft-v1 ready after 161.21022057533264s
Pipeline stage MKMLDeployer completed in 162.88s
run pipeline stage %s
Running pipeline stage StressChecker
Received healthy response to inference request in 2.639464855194092s
Failed to get response for submission chaiml-elo-alignment-run-3_v41: ('http://chaiml-llama-8b-pairwis-8189-v19-predictor.tenant-chaiml-guanaco.k.chaiverse.com/v1/models/GPT-J-6B-lit-v2:predict', 'read tcp 127.0.0.1:45426->127.0.0.1:8080: read: connection reset by peer\n')
Received healthy response to inference request in 2.1718075275421143s
Received healthy response to inference request in 1.8709888458251953s
Received healthy response to inference request in 3.0572893619537354s
Received healthy response to inference request in 1.2721972465515137s
5 requests
0 failed requests
5th percentile: 1.39195556640625
10th percentile: 1.5117138862609862
20th percentile: 1.751230525970459
30th percentile: 1.931152582168579
40th percentile: 2.0514800548553467
50th percentile: 2.1718075275421143
60th percentile: 2.3588704586029055
70th percentile: 2.545933389663696
80th percentile: 2.7230297565460204
90th percentile: 2.890159559249878
95th percentile: 2.9737244606018067
99th percentile: 3.0405763816833495
mean time: 2.20234956741333
Pipeline stage StressChecker completed in 11.66s
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 8.10s
Shutdown handler de-registered
riverise-0910-2005-sft_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 riverise-0910-2005-sft-v1-profiler
Waiting for inference service riverise-0910-2005-sft-v1-profiler to be ready
Inference service riverise-0910-2005-sft-v1-profiler ready after 160.37507915496826s
Pipeline stage MKMLProfilerDeployer completed in 160.72s
run pipeline stage %s
Running pipeline stage MKMLProfilerRunner
kubectl cp /code/guanaco/guanaco_inference_services/src/inference_scripts tenant-chaiml-guanaco/riverise-0910-2005-sft-v1-profiler-predictor-00001-deploymjmgtw:/code/chaiverse_profiler_1726031513 --namespace tenant-chaiml-guanaco
kubectl exec -it riverise-0910-2005-sft-v1-profiler-predictor-00001-deploymjmgtw --namespace tenant-chaiml-guanaco -- sh -c 'cd /code/chaiverse_profiler_1726031513 && python profiles.py profile --best_of_n 16 --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_1726031513/summary.json'
kubectl exec -it riverise-0910-2005-sft-v1-profiler-predictor-00001-deploymjmgtw --namespace tenant-chaiml-guanaco -- bash -c 'cat /code/chaiverse_profiler_1726031513/summary.json'
Pipeline stage MKMLProfilerRunner completed in 836.26s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeleter
Checking if service riverise-0910-2005-sft-v1-profiler is running
Tearing down inference service riverise-0910-2005-sft-v1-profiler
Service riverise-0910-2005-sft-v1-profiler has been torndown
Pipeline stage MKMLProfilerDeleter completed in 1.97s
Shutdown handler de-registered
riverise-0910-2005-sft_v1 status is now inactive due to auto deactivation removed underperforming models
riverise-0910-2005-sft_v1 status is now torndown due to DeploymentManager action