2024-02-09 10:02:57 Publish script update

app.py

@@ -9,9 +9,9 @@ from pipeline.config import LLMBoardConfig, QueriesConfig
 
 README = """
 Projects compares different large language models and their providers for real time applications and mass data processing. 
-While other benchmarks compare LLMs on different human intelligence tasks this bencchmark focus on features related to business and engineering aspects such as response times, pricing and data streaming capabilities. 
+While other benchmarks compare LLMs on different human intelligence tasks this benchmark focus on features related to business and engineering aspects such as response times, pricing and data streaming capabilities. 
 
-To preform evaluation we chose a task of newspaper articles summarization from DATASET_LINK/NAME as it represents a very standard type of task where model has to understand unstructured natural language text, process it and output text in a specified format.
+To preform evaluation we chose a task of newspaper articles summarization from [GEM/xlsum](https://huggingface.co/datasets/GEM/xlsum) dataset as it represents a very standard type of task where model has to understand unstructured natural language text, process it and output text in a specified format.
 For this version we chose English, Polish and Japanese languages, with Japanese representing languages using logographic alphabets. This enable us also validate the effectiveness of the LLM for different language groups.
 
 Each of the models was asked to summarize the text using the following prompt:
@@ -20,7 +20,7 @@ Each of the models was asked to summarize the text using the following prompt:
 {}
 ```
 
-Where \{language\} stands for original language of the text as we wanted to avoid the model translating the text to English during summarization.
+Where {{language}} stands for original language of the text as we wanted to avoid the model translating the text to English during summarization.
 
 LLM was asked to return the output in three formats: markdown, json and function call. Note that currently function calls are only supported by Open AI API.
 To do that we added following text to the query: