submission_id: jellywibble-lora-30k-pre_9052_v2
developer_uid: Jellywibble
status: inactive
model_repo: Jellywibble/lora_30k_pref_data_ep1
reward_repo: ChaiML/reward_gpt2_medium_preference_24m_e2
generation_params: {'temperature': 0.95, 'top_p': 1.0, 'min_p': 0.08, 'top_k': 40, 'presence_penalty': 0.0, 'frequency_penalty': 0.0, 'stopping_words': ['\n', '<|eot_id|>'], 'max_input_tokens': 512, 'best_of': 16, 'max_output_tokens': 64}
formatter: {'memory_template': "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n{bot_name}'s Persona: {memory}\n\n", 'prompt_template': '{prompt}<|eot_id|>', 'bot_template': '<|start_header_id|>assistant<|end_header_id|>\n\n{bot_name}: {message}<|eot_id|>', 'user_template': '<|start_header_id|>user<|end_header_id|>\n\n{user_name}: {message}<|eot_id|>', 'response_template': '<|start_header_id|>assistant<|end_header_id|>\n\n{bot_name}:', 'truncate_by_message': False}
reward_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}
timestamp: 2024-07-04T04:23:06+00:00
model_name: nitral-ai-hathor-l3-8b-v-01_v1
model_group: Jellywibble/lora_30k_pre
num_battles: 25000
num_wins: 13209
celo_rating: 1207.02
propriety_score: 0.7044901384809065
propriety_total_count: 11915.0
submission_type: basic
model_architecture: LlamaForCausalLM
model_num_parameters: 8030261248.0
best_of: 16
max_input_tokens: 512
max_output_tokens: 64
display_name: nitral-ai-hathor-l3-8b-v-01_v1
ineligible_reason: None
language_model: Jellywibble/lora_30k_pref_data_ep1
model_size: 8B
reward_model: ChaiML/reward_gpt2_medium_preference_24m_e2
us_pacific_date: 2024-07-03
win_ratio: 0.52836
Resubmit model
Running pipeline stage MKMLizer
Starting job with name jellywibble-lora-30k-pre-9052-v2-mkmlizer
Waiting for job on jellywibble-lora-30k-pre-9052-v2-mkmlizer to finish
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ╔═════════════════════════════════════════════════════════════════════╗
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ _____ __ __ ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ / _/ /_ ___ __/ / ___ ___ / / ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ / _/ / // / |/|/ / _ \/ -_) -_) / ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ /_//_/\_, /|__,__/_//_/\__/\__/_/ ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ /___/ ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ Version: 0.8.14 ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ Copyright 2023 MK ONE TECHNOLOGIES Inc. ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ https://mk1.ai ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ The license key for the current software has been verified as ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ belonging to: ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ Chai Research Corp. ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ Account ID: 7997a29f-0ceb-4cc7-9adf-840c57b4ae6f ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ Expiration: 2024-07-15 23:59:59 ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ║ ║
jellywibble-lora-30k-pre-9052-v2-mkmlizer: ╚═════════════════════════════════════════════════════════════════════╝
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Downloaded to shared memory in 75.011s
jellywibble-lora-30k-pre-9052-v2-mkmlizer: quantizing model to /dev/shm/model_cache
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Saving flywheel model at /dev/shm/model_cache
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Loading 0: 0%| | 0/291 [00:00<?, ?it/s] Loading 0: 2%|▏ | 5/291 [00:00<00:07, 39.97it/s] Loading 0: 4%|▍ | 13/291 [00:00<00:04, 58.53it/s] Loading 0: 7%|▋ | 21/291 [00:00<00:04, 65.04it/s] Loading 0: 10%|▉ | 28/291 [00:00<00:04, 64.59it/s] Loading 0: 12%|█▏ | 35/291 [00:00<00:07, 32.93it/s] Loading 0: 14%|█▎ | 40/291 [00:00<00:07, 34.98it/s] Loading 0: 16%|█▋ | 48/291 [00:01<00:05, 43.77it/s] Loading 0: 19%|█▊ | 54/291 [00:01<00:05, 46.91it/s] Loading 0: 21%|██ | 60/291 [00:01<00:04, 47.52it/s] Loading 0: 23%|██▎ | 67/291 [00:01<00:04, 52.39it/s] Loading 0: 25%|██▌ | 74/291 [00:01<00:03, 55.94it/s] Loading 0: 28%|██▊ | 81/291 [00:01<00:05, 35.47it/s] Loading 0: 30%|██▉ | 87/291 [00:01<00:05, 39.73it/s] Loading 0: 32%|███▏ | 93/291 [00:02<00:04, 42.09it/s] Loading 0: 34%|███▍ | 100/291 [00:02<00:04, 47.45it/s] Loading 0: 37%|███▋ | 108/291 [00:02<00:03, 54.15it/s] Loading 0: 40%|███▉ | 115/291 [00:02<00:03, 56.92it/s] Loading 0: 42%|████▏ | 122/291 [00:02<00:02, 59.33it/s] Loading 0: 44%|████▍ | 129/291 [00:02<00:02, 58.80it/s] Loading 0: 47%|████▋ | 136/291 [00:03<00:04, 36.95it/s] Loading 0: 48%|████▊ | 141/291 [00:03<00:03, 38.08it/s] Loading 0: 51%|█████ | 148/291 [00:03<00:03, 43.87it/s] Loading 0: 54%|█████▎ | 156/291 [00:03<00:02, 50.98it/s] Loading 0: 56%|█████▌ | 162/291 [00:03<00:02, 52.61it/s] Loading 0: 58%|█████▊ | 168/291 [00:03<00:02, 51.79it/s] Loading 0: 60%|██████ | 175/291 [00:03<00:02, 55.80it/s] Loading 0: 64%|██████▍ | 186/291 [00:04<00:02, 43.20it/s] Loading 0: 66%|██████▌ | 192/291 [00:04<00:02, 41.88it/s] Loading 0: 68%|██████▊ | 198/291 [00:04<00:02, 44.98it/s] Loading 0: 70%|███████ | 204/291 [00:04<00:01, 46.04it/s] Loading 0: 73%|███████▎ | 211/291 [00:04<00:01, 50.74it/s] Loading 0: 75%|███████▌ | 219/291 [00:04<00:01, 56.74it/s] Loading 0: 78%|███████▊ | 226/291 [00:04<00:01, 58.34it/s] Loading 0: 80%|████████ | 233/291 [00:05<00:01, 37.80it/s] Loading 0: 82%|████████▏ | 238/291 [00:05<00:01, 38.49it/s] Loading 0: 85%|████████▍ | 246/291 [00:05<00:00, 46.00it/s] Loading 0: 87%|████████▋ | 252/291 [00:05<00:00, 48.74it/s] Loading 0: 89%|████████▊ | 258/291 [00:05<00:00, 48.96it/s] Loading 0: 91%|█████████ | 265/291 [00:05<00:00, 54.08it/s] Loading 0: 94%|█████████▍| 273/291 [00:05<00:00, 60.25it/s] Loading 0: 96%|█████████▌| 280/291 [00:05<00:00, 62.27it/s] Loading 0: 99%|█████████▊| 287/291 [00:11<00:01, 3.74it/s] Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
jellywibble-lora-30k-pre-9052-v2-mkmlizer: quantized model in 27.055s
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Processed model Jellywibble/lora_30k_pref_data_ep1 in 102.066s
jellywibble-lora-30k-pre-9052-v2-mkmlizer: creating bucket guanaco-mkml-models
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Bucket 's3://guanaco-mkml-models/' created
jellywibble-lora-30k-pre-9052-v2-mkmlizer: uploading /dev/shm/model_cache to s3://guanaco-mkml-models/jellywibble-lora-30k-pre-9052-v2
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /dev/shm/model_cache/tokenizer_config.json s3://guanaco-mkml-models/jellywibble-lora-30k-pre-9052-v2/tokenizer_config.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /dev/shm/model_cache/special_tokens_map.json s3://guanaco-mkml-models/jellywibble-lora-30k-pre-9052-v2/special_tokens_map.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /dev/shm/model_cache/config.json s3://guanaco-mkml-models/jellywibble-lora-30k-pre-9052-v2/config.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /dev/shm/model_cache/tokenizer.json s3://guanaco-mkml-models/jellywibble-lora-30k-pre-9052-v2/tokenizer.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /dev/shm/model_cache/flywheel_model.0.safetensors s3://guanaco-mkml-models/jellywibble-lora-30k-pre-9052-v2/flywheel_model.0.safetensors
jellywibble-lora-30k-pre-9052-v2-mkmlizer: loading reward model from ChaiML/reward_gpt2_medium_preference_24m_e2
jellywibble-lora-30k-pre-9052-v2-mkmlizer: /opt/conda/lib/python3.10/site-packages/transformers/models/auto/configuration_auto.py:919: FutureWarning: The `use_auth_token` argument is deprecated and will be removed in v5 of Transformers. Please use `token` instead.
jellywibble-lora-30k-pre-9052-v2-mkmlizer: warnings.warn(
jellywibble-lora-30k-pre-9052-v2-mkmlizer: /opt/conda/lib/python3.10/site-packages/huggingface_hub/file_download.py:1132: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
jellywibble-lora-30k-pre-9052-v2-mkmlizer: warnings.warn(
jellywibble-lora-30k-pre-9052-v2-mkmlizer: /opt/conda/lib/python3.10/site-packages/transformers/models/auto/tokenization_auto.py:769: FutureWarning: The `use_auth_token` argument is deprecated and will be removed in v5 of Transformers. Please use `token` instead.
jellywibble-lora-30k-pre-9052-v2-mkmlizer: warnings.warn(
jellywibble-lora-30k-pre-9052-v2-mkmlizer: /opt/conda/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py:468: FutureWarning: The `use_auth_token` argument is deprecated and will be removed in v5 of Transformers. Please use `token` instead.
jellywibble-lora-30k-pre-9052-v2-mkmlizer: warnings.warn(
jellywibble-lora-30k-pre-9052-v2-mkmlizer: /opt/conda/lib/python3.10/site-packages/torch/_utils.py:831: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly. To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
jellywibble-lora-30k-pre-9052-v2-mkmlizer: return self.fget.__get__(instance, owner)()
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Saving model to /tmp/reward_cache/reward.tensors
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Saving duration: 0.390s
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Processed model ChaiML/reward_gpt2_medium_preference_24m_e2 in 12.672s
jellywibble-lora-30k-pre-9052-v2-mkmlizer: creating bucket guanaco-reward-models
jellywibble-lora-30k-pre-9052-v2-mkmlizer: Bucket 's3://guanaco-reward-models/' created
jellywibble-lora-30k-pre-9052-v2-mkmlizer: uploading /tmp/reward_cache to s3://guanaco-reward-models/jellywibble-lora-30k-pre-9052-v2_reward
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /tmp/reward_cache/config.json s3://guanaco-reward-models/jellywibble-lora-30k-pre-9052-v2_reward/config.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /tmp/reward_cache/tokenizer_config.json s3://guanaco-reward-models/jellywibble-lora-30k-pre-9052-v2_reward/tokenizer_config.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /tmp/reward_cache/special_tokens_map.json s3://guanaco-reward-models/jellywibble-lora-30k-pre-9052-v2_reward/special_tokens_map.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /tmp/reward_cache/merges.txt s3://guanaco-reward-models/jellywibble-lora-30k-pre-9052-v2_reward/merges.txt
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /tmp/reward_cache/vocab.json s3://guanaco-reward-models/jellywibble-lora-30k-pre-9052-v2_reward/vocab.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /tmp/reward_cache/tokenizer.json s3://guanaco-reward-models/jellywibble-lora-30k-pre-9052-v2_reward/tokenizer.json
jellywibble-lora-30k-pre-9052-v2-mkmlizer: cp /tmp/reward_cache/reward.tensors s3://guanaco-reward-models/jellywibble-lora-30k-pre-9052-v2_reward/reward.tensors
Job jellywibble-lora-30k-pre-9052-v2-mkmlizer completed after 145.55s with status: succeeded
Stopping job with name jellywibble-lora-30k-pre-9052-v2-mkmlizer
Pipeline stage MKMLizer completed in 146.54s
Running pipeline stage MKMLKubeTemplater
Pipeline stage MKMLKubeTemplater completed in 0.11s
Running pipeline stage ISVCDeployer
Creating inference service jellywibble-lora-30k-pre-9052-v2
Waiting for inference service jellywibble-lora-30k-pre-9052-v2 to be ready
Inference service jellywibble-lora-30k-pre-9052-v2 ready after 50.361170291900635s
Pipeline stage ISVCDeployer completed in 57.78s
Running pipeline stage StressChecker
Received healthy response to inference request in 2.1474835872650146s
Received healthy response to inference request in 1.4021847248077393s
Received healthy response to inference request in 1.3852262496948242s
Received healthy response to inference request in 1.353020191192627s
Received healthy response to inference request in 1.4253201484680176s
5 requests
0 failed requests
5th percentile: 1.3594614028930665
10th percentile: 1.3659026145935058
20th percentile: 1.3787850379943847
30th percentile: 1.3886179447174072
40th percentile: 1.3954013347625733
50th percentile: 1.4021847248077393
60th percentile: 1.4114388942718505
70th percentile: 1.4206930637359618
80th percentile: 1.5697528362274171
90th percentile: 1.858618211746216
95th percentile: 2.003050899505615
99th percentile: 2.118597049713135
mean time: 1.5426469802856446
Pipeline stage StressChecker completed in 8.77s
jellywibble-lora-30k-pre_9052_v2 status is now deployed due to DeploymentManager action
jellywibble-lora-30k-pre_9052_v2 status is now inactive due to auto deactivation removed underperforming models

Usage Metrics

Latency Metrics