rica40325-filter-mulsft-_2205

developer_uid: rica40325

submission_id: rica40325-filter-mulsft-_2205_v4

model_name: rica40325-filter-mulsft-_2205_v4

model_group: rica40325/filter-mulsft-

status: torndown

timestamp: 2025-06-21T15:35:26+00:00

num_battles: 8178

num_wins: 3653

celo_rating: 1243.55

family_friendly_score: 0.5358

family_friendly_standard_error: 0.007052919395541112

submission_type: basic

model_repo: rica40325/filter-mulsft-v5-600

model_architecture: MistralForCausalLM

model_num_parameters: 12772070400.0

best_of: 8

max_input_tokens: 1024

max_output_tokens: 64

reward_model: default

latencies: [{'batch_size': 1, 'throughput': 0.5896374608188018, 'latency_mean': 1.6958374297618866, 'latency_p50': 1.684547781944275, 'latency_p90': 1.864600419998169}, {'batch_size': 3, 'throughput': 1.058547724659918, 'latency_mean': 2.8273551762104034, 'latency_p50': 2.8221545219421387, 'latency_p90': 3.093483591079712}, {'batch_size': 5, 'throughput': 1.2676794195433159, 'latency_mean': 3.9313712739944457, 'latency_p50': 3.916387438774109, 'latency_p90': 4.4344443559646605}, {'batch_size': 6, 'throughput': 1.3174497846388002, 'latency_mean': 4.529395688772201, 'latency_p50': 4.525589346885681, 'latency_p90': 5.0642801284790036}, {'batch_size': 8, 'throughput': 1.3846726373215377, 'latency_mean': 5.737730449438095, 'latency_p50': 5.8005088567733765, 'latency_p90': 6.339632940292359}, {'batch_size': 10, 'throughput': 1.4084897788084985, 'latency_mean': 7.054127885103226, 'latency_p50': 7.106122255325317, 'latency_p90': 7.895165991783141}]

gpu_counts: {'NVIDIA RTX A5000': 1}

display_name: rica40325-filter-mulsft-_2205_v4

is_internal_developer: False

language_model: rica40325/filter-mulsft-v5-600

model_size: 13B

ranking_group: single

throughput_3p7s: 1.24

us_pacific_date: 2025-06-21

win_ratio: 0.4466862313524089

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': 1024, 'best_of': 8, 'max_output_tokens': 64}

formatter: {'memory_template': '{memory}', 'prompt_template': '[INST]{prompt}[/INST]', 'bot_template': '{bot_name}: {message}', 'user_template': '[INST]{user_name}: {message}[/INST]', 'response_template': '{bot_name}:', 'truncate_by_message': False}

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 rica40325-filter-mulsft-2205-v4-mkmlizer
Waiting for job on rica40325-filter-mulsft-2205-v4-mkmlizer to finish
rica40325-filter-mulsft-2205-v4-mkmlizer: Downloaded to shared memory in 67.212s
rica40325-filter-mulsft-2205-v4-mkmlizer: Checking if rica40325/filter-mulsft-v5-600 already exists in ChaiML
rica40325-filter-mulsft-2205-v4-mkmlizer: Creating repo ChaiML/filter-mulsft-v5-600 and uploading /tmp/tmp7ov2bpno to it
rica40325-filter-mulsft-2205-v4-mkmlizer: 
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:04<00:47,  4.28s/it]
 17%|█▋        | 2/12 [00:12<01:06,  6.60s/it]
 25%|██▌       | 3/12 [00:20<01:05,  7.27s/it]
 33%|███▎      | 4/12 [00:28<01:01,  7.71s/it]
 42%|████▏     | 5/12 [00:37<00:55,  7.89s/it]
 50%|█████     | 6/12 [00:44<00:45,  7.54s/it]
 58%|█████▊    | 7/12 [00:47<00:31,  6.33s/it]
 67%|██████▋   | 8/12 [00:51<00:22,  5.53s/it]
 75%|███████▌  | 9/12 [00:56<00:15,  5.17s/it]
 83%|████████▎ | 10/12 [01:02<00:11,  5.68s/it]
 92%|█████████▏| 11/12 [01:04<00:04,  4.30s/it]
100%|██████████| 12/12 [01:04<00:00,  3.21s/it]
100%|██████████| 12/12 [01:04<00:00,  5.40s/it]
rica40325-filter-mulsft-2205-v4-mkmlizer: quantizing model to /dev/shm/model_cache, profile:s0, folder:/tmp/tmp7ov2bpno, device:0
rica40325-filter-mulsft-2205-v4-mkmlizer: Saving flywheel model at /dev/shm/model_cache
rica40325-filter-mulsft-2205-v4-mkmlizer: /opt/conda/lib/python3.10/site-packages/mk1/flywheel/functional/loader.py:55: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
rica40325-filter-mulsft-2205-v4-mkmlizer:   tensors = torch.load(model_shard_filename, map_location=torch.device(self.device), mmap=True)
rica40325-filter-mulsft-2205-v4-mkmlizer: quantized model in 38.504s
rica40325-filter-mulsft-2205-v4-mkmlizer: Processed model rica40325/filter-mulsft-v5-600 in 218.646s
rica40325-filter-mulsft-2205-v4-mkmlizer: creating bucket guanaco-mkml-models
rica40325-filter-mulsft-2205-v4-mkmlizer: Bucket 's3://guanaco-mkml-models/' created
rica40325-filter-mulsft-2205-v4-mkmlizer: uploading /dev/shm/model_cache to s3://guanaco-mkml-models/rica40325-filter-mulsft-2205-v4
rica40325-filter-mulsft-2205-v4-mkmlizer: cp /dev/shm/model_cache/config.json s3://guanaco-mkml-models/rica40325-filter-mulsft-2205-v4/config.json
rica40325-filter-mulsft-2205-v4-mkmlizer: cp /dev/shm/model_cache/special_tokens_map.json s3://guanaco-mkml-models/rica40325-filter-mulsft-2205-v4/special_tokens_map.json
rica40325-filter-mulsft-2205-v4-mkmlizer: cp /dev/shm/model_cache/tokenizer_config.json s3://guanaco-mkml-models/rica40325-filter-mulsft-2205-v4/tokenizer_config.json
rica40325-filter-mulsft-2205-v4-mkmlizer: cp /dev/shm/model_cache/tokenizer.json s3://guanaco-mkml-models/rica40325-filter-mulsft-2205-v4/tokenizer.json
rica40325-filter-mulsft-2205-v4-mkmlizer: cp /dev/shm/model_cache/flywheel_model.0.safetensors s3://guanaco-mkml-models/rica40325-filter-mulsft-2205-v4/flywheel_model.0.safetensors
rica40325-filter-mulsft-2205-v4-mkmlizer: 
Loading 0:   0%|          | 0/363 [00:00<?, ?it/s]
Loading 0:   2%|▏         | 6/363 [00:00<00:05, 59.78it/s]
Loading 0:   3%|▎         | 12/363 [00:00<00:06, 56.78it/s]
Loading 0:   5%|▍         | 18/363 [00:00<00:06, 54.73it/s]
Loading 0:   7%|▋         | 24/363 [00:01<00:27, 12.19it/s]
Loading 0:   8%|▊         | 28/363 [00:01<00:23, 14.52it/s]
Loading 0:  10%|▉         | 35/363 [00:01<00:15, 21.02it/s]
Loading 0:  12%|█▏        | 42/363 [00:01<00:11, 27.82it/s]
Loading 0:  13%|█▎        | 49/363 [00:02<00:25, 12.44it/s]
Loading 0:  15%|█▌        | 56/363 [00:03<00:18, 16.74it/s]
Loading 0:  17%|█▋        | 63/363 [00:03<00:13, 21.92it/s]
Loading 0:  19%|█▉        | 70/363 [00:03<00:10, 26.85it/s]
Loading 0:  21%|██        | 77/363 [00:03<00:09, 31.33it/s]
Loading 0:  23%|██▎       | 83/363 [00:04<00:20, 13.43it/s]
Loading 0:  24%|██▍       | 88/363 [00:04<00:16, 16.29it/s]
Loading 0:  26%|██▌       | 95/363 [00:04<00:12, 21.19it/s]
Loading 0:  28%|██▊       | 102/363 [00:04<00:09, 26.25it/s]
Loading 0:  30%|███       | 109/363 [00:05<00:08, 31.42it/s]
Loading 0:  31%|███▏      | 114/363 [00:06<00:19, 12.97it/s]
Loading 0:  33%|███▎      | 119/363 [00:06<00:15, 15.83it/s]
Loading 0:  35%|███▍      | 126/363 [00:06<00:11, 21.10it/s]
Loading 0:  37%|███▋      | 133/363 [00:06<00:08, 26.77it/s]
Loading 0:  39%|███▊      | 140/363 [00:06<00:06, 32.33it/s]
Loading 0:  40%|████      | 146/363 [00:07<00:15, 13.78it/s]
Loading 0:  42%|████▏     | 151/363 [00:07<00:12, 16.62it/s]
Loading 0:  44%|████▎     | 158/363 [00:07<00:09, 21.95it/s]
Loading 0:  45%|████▌     | 165/363 [00:08<00:07, 26.74it/s]
Loading 0:  47%|████▋     | 172/363 [00:08<00:06, 31.08it/s]
Loading 0:  49%|████▉     | 177/363 [00:09<00:14, 12.99it/s]
Loading 0:  50%|█████     | 182/363 [00:09<00:11, 15.70it/s]
Loading 0:  52%|█████▏    | 189/363 [00:09<00:08, 20.87it/s]
Loading 0:  54%|█████▍    | 196/363 [00:09<00:06, 25.96it/s]
Loading 0:  56%|█████▌    | 203/363 [00:09<00:04, 32.12it/s]
Loading 0:  58%|█████▊    | 209/363 [00:10<00:11, 13.57it/s]
Loading 0:  59%|█████▉    | 214/363 [00:10<00:09, 16.14it/s]
Loading 0:  61%|██████    | 221/363 [00:11<00:06, 21.31it/s]
Loading 0:  63%|██████▎   | 228/363 [00:11<00:05, 26.95it/s]
Loading 0:  65%|██████▍   | 235/363 [00:11<00:04, 31.59it/s]
Loading 0:  66%|██████▌   | 240/363 [00:12<00:09, 12.99it/s]
Loading 0:  67%|██████▋   | 245/363 [00:12<00:07, 15.80it/s]
Loading 0:  69%|██████▉   | 252/363 [00:12<00:05, 21.04it/s]
Loading 0:  71%|███████▏  | 259/363 [00:12<00:03, 26.11it/s]
Loading 0:  73%|███████▎  | 266/363 [00:12<00:03, 31.46it/s]
Loading 0:  75%|███████▍  | 271/363 [00:14<00:07, 13.00it/s]
Loading 0:  76%|███████▋  | 277/363 [00:14<00:05, 16.14it/s]
                                                            
Job rica40325-filter-mulsft-2205-v4-mkmlizer completed after 256.4s with status: succeeded
Stopping job with name rica40325-filter-mulsft-2205-v4-mkmlizer
Pipeline stage MKMLizer completed in 256.87s
run pipeline stage %s
Running pipeline stage MKMLTemplater
Pipeline stage MKMLTemplater completed in 0.14s
run pipeline stage %s
Running pipeline stage MKMLDeployer
Creating inference service rica40325-filter-mulsft-2205-v4
Waiting for inference service rica40325-filter-mulsft-2205-v4 to be ready
Inference service rica40325-filter-mulsft-2205-v4 ready after 120.43140578269958s
Pipeline stage MKMLDeployer completed in 120.94s
run pipeline stage %s
Running pipeline stage StressChecker
Received healthy response to inference request in 2.3868091106414795s
Received healthy response to inference request in 1.522552490234375s
Received healthy response to inference request in 2.0631847381591797s
Received healthy response to inference request in 1.679926872253418s
Received healthy response to inference request in 1.8946197032928467s
5 requests
0 failed requests
5th percentile: 1.5540273666381836
10th percentile: 1.5855022430419923
20th percentile: 1.6484519958496093
30th percentile: 1.7228654384613038
40th percentile: 1.8087425708770752
50th percentile: 1.8946197032928467
60th percentile: 1.96204571723938
70th percentile: 2.029471731185913
80th percentile: 2.12790961265564
90th percentile: 2.2573593616485597
95th percentile: 2.3220842361450194
99th percentile: 2.3738641357421875
mean time: 1.9094185829162598
Pipeline stage StressChecker completed in 10.78s
run pipeline stage %s
Running pipeline stage OfflineFamilyFriendlyTriggerPipeline
run_pipeline:run_in_cloud %s
starting trigger_guanaco_pipeline args=%s
triggered trigger_guanaco_pipeline args=%s
Pipeline stage OfflineFamilyFriendlyTriggerPipeline completed in 0.69s
run pipeline stage %s
Running pipeline stage TriggerMKMLProfilingPipeline
run_pipeline:run_in_cloud %s
starting trigger_guanaco_pipeline args=%s
triggered trigger_guanaco_pipeline args=%s
Pipeline stage TriggerMKMLProfilingPipeline completed in 0.84s
Shutdown handler de-registered
rica40325-filter-mulsft-_2205_v4 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 0.10s
run pipeline stage %s
Running pipeline stage MKMLProfilerTemplater
Pipeline stage MKMLProfilerTemplater completed in 0.10s
run pipeline stage %s
Running pipeline stage MKMLProfilerDeployer
Creating inference service rica40325-filter-mulsft-2205-v4-profiler
Waiting for inference service rica40325-filter-mulsft-2205-v4-profiler to be ready
Shutdown handler registered
run pipeline %s
run pipeline stage %s
Running pipeline stage OfflineFamilyFriendlyScorer
Evaluating %s Family Friendly Score with %s threads
%s, retrying in %s seconds...
Evaluating %s Family Friendly Score with %s threads
Pipeline stage OfflineFamilyFriendlyScorer completed in 3923.08s
Shutdown handler de-registered
rica40325-filter-mulsft-_2205_v4 status is now inactive due to auto deactivation removed underperforming models
rica40325-filter-mulsft-_2205_v4 status is now torndown due to DeploymentManager action