long-t5 and onnx

evaluating using optimum with ONNX runtime + long-t5 models for summarization
will be updated with new logs over time

examples
with ONNX conversion
2023-04-29 22:14:01,759 - INFO - Loaded model pszemraj/long-t5-tglobal-base-sci-simplify-elife to cpu with ONNX Runtime
2023-04-29 22:14:01,760 - INFO - input parameters: {'length_penalty': 0.5, 'repetition_penalty': 1.0, 'no_repeat_ngram_size': 3, 'encoder_no_repeat_ngram_size': 4, 'num_beams': 4, 'min_length': 4, 'max_length': 256, 'early_stopping': True, 'do_sample': False}, batch_length=1024, batch_stride=16

  0%|          | 0/3 [00:00<?, ?it/s]2023-04-29 22:16:27,057 - INFO - 	The Eldar have a great lore about the Eldar and the Valar; and they have many of the same names as the Elves. The Valar and the Maiar are a group of Elves, and they live in a world called Arda. The Elves have many names in their tongues, and many of them are called gods. In the Elvish language they are called Manwe, Melkor, and Ulmo. Manwe is the first of the seven Kings of Arda; and the Queen Varda lives in the Stars. The Queens are seven also, and the Lords are also seven. The Ainur have many other names, but their names are the same as those of Manwe and the Melkor. The name of the Queen is not spoken on Earth, but it is spoken in the tongues of the Elders, as it was in Valino. The Lord of the Winds and the Seas, Manwe; the Queen of the Waters, Varra; the King of the Mountains, Sulimo; the Lord of Winds, Orome; Mando; Lorien; and Tulks.
Score:	-20.2915

 33%|███▎      | 1/3 [02:25<04:50, 145.28s/it]2023-04-29 22:19:10,363 - INFO - 	The Elves are a people of great strength, and they have a great power over all the things that are made in the world. They are the masters of all the substances that make up the world, and the Feanturi are the lords of the spirits that are found in the Earth. The Feantsuri are bred of gold and silver, and are the jewels that are fair in their hand. They also have the horns that make music in the deep waters of the Earth; and they speak to those who long for the sea. Melkor is jealous of them, for he is most like him in thought and powers; and the Elves have long fought and fought against him, but they have never lost their friendship. The Fanturi also speak to Elves and men with voices that they hear only as the sounds of water. They make music that are heard in all the seas and lakes, rivers and fountains, and their voices are heard as the water of the ocean.
Score:	-17.5264

 67%|██████▋   | 2/3 [05:08<02:35, 155.88s/it]2023-04-29 22:22:05,762 - INFO - 	Dough Dough is a Dough-duck, and Doughducks are a dreed; and Ducks Doodles are Duck-deeddddos. Duckdeeds Daughtedda Duck Doodle Doodled DuckDeed DoodleDddded Dock Deed Deedd Dowed Doweddd Ddeed DowedDd DeedDdDdde Ddded DdeddDdedDeeddDedDidds DeedsDeedsdeedDeedDde Deeddedsddeeded ddreddd, Ddred DddeDdsDdeed dowed Ddderdd sdered Dddenddhddvdd. Dddondddes Dddesddhedd'dddon'd'ded'Dddesded teeps dreads sneezes pity spouts lamentation screams flutters
Score:	-33.3491

100%|██████████| 3/3 [08:03<00:00, 164.79s/it]
100%|██████████| 3/3 [08:03<00:00, 161.33s/it]
2023-04-29 22:22:05,784 - INFO - Runtime: 9.15 minutes

two
2023-04-29 22:40:59,634 - INFO - Loaded model pszemraj/long-t5-tglobal-base-sci-simplify-elife to cpu with ONNX Runtime
2023-04-29 22:40:59,634 - INFO - input parameters: {'length_penalty': 0.5, 'repetition_penalty': 1.0, 'no_repeat_ngram_size': 3, 'encoder_no_repeat_ngram_size': 4, 'num_beams': 4, 'min_length': 4, 'max_length': 256, 'early_stopping': True, 'do_sample': False}, batch_length=1024, batch_stride=16

  0%|          | 0/4 [00:00<?, ?it/s]2023-04-29 22:43:31,562 - INFO - 	Reinforcement learning is an artificial intelligence framework that can help us to learn how to make decisions. It is a way to build artificial intelligence agents that can learn optimal behavior by interacting and receiving rewards. The goal is to learn the optimal behavior. The agent collects data from other agents and human demonstrations and then uses this data to update its policy. This process is called offline reinforcement learning. But it's not as simple as it is in online reinforcement learning: the agent only interacts with the environment, and it does not have the data. This means that it can be expensive and insecure, and the agent can exploit it if it provides a competitive edge. The Decision Transformer is an algorithm that can be used to build decision transformers. The algorithm works by generating a sequence of actions that will achieve the desired reward. The model predicts the future actions that the agent will take to achieve this reward.
Score:	-14.2222

 25%|██▌       | 1/4 [02:31<07:35, 151.92s/it]2023-04-29 22:46:02,771 - INFO - 	The Decision Transformer is a computer model that predicts the future actions of a user. The model is based on a model called GPT-2, which is able to predict future actions by predicting future actions using a self-attention mask. This mask can be used to predict the future behavior of the user. Using this mask, the model predicts future behaviors by calculating how much time it takes to complete a task, and then predicting the next time-step in the model. This model is also available as a part of the Huggingface library, which will soon be made available in the Atari environment.
Score:	-11.8279

 50%|█████     | 2/4 [05:03<05:03, 151.50s/it]2023-04-29 22:46:43,159 - INFO - 	Decision Transformers are a set of tools that can be used to control the behavior of a game, but they are not the only tools that we can use to control game behavior. Instead, we can also use these tools to control how the game behaves.
Score:	-8.0473

 75%|███████▌  | 3/4 [05:43<01:40, 100.76s/it]2023-04-29 22:49:41,384 - INFO - 	DD Dd DdD D DddDD DDdDddd DDDDd D D Dvdd dddeddcddsddfddvdseddeedsdeedddpdd-ddadddesddledddedsDdeedDds DdeededDvdDed DdedDsdDeedsDeeddDpdDfd DfdDcdDad Dp DddeDdvDdfvd Dcd DvDDvD Df DdfD Dpd Dsd Dn DdpD Dv Ddc Ddvsdseeddeed Dd,ddecteddderedddakedddéeddiseddd,Ddected Ddecteded, datedddredddatedDdredsd, Ddredded DreddDatedd DateddD dredDdated Ddered Ddised Ddateded edected
Score:	-36.0197

100%|██████████| 4/4 [08:41<00:00, 131.35s/it]
100%|██████████| 4/4 [08:41<00:00, 130.44s/it]
2023-04-29 22:49:41,406 - INFO - Runtime: 9.82 minutes