Motivation
I’like to have a model generated summaries about all my articles and am experimenting generating with different models. To be able to ingest a long text i took led-large-16384 and have got nearly a trash output. At the same time, the same source text is summarizied by ChatGPT3.5 quite well.
Summarized article
Result of led-large-16384
Before precoessing the source text content i’ve cleaned it up removing tables, images and code blocks. It helped a bit improving the very first absolute trash result.
The best output i’ve got from the model yet:
summarizer = pipeline("summarization", model="allenai/led-base-16384")
result = summarizer(content, max_length=256, min_length=128).
Output:
Explore and visualize the parser events and performance. Pandas. Seaborn. KLib. ADTK First, i was going to start writing an article about researching the parser events and performance. I decided to start with a few examples of the parser events and performance. I started with a few examples of the parser events and data.seaborn. In the first example, i was going to write a few examples of the parser events and performance. In the second example, i was going to write a few examples of the parser events and performance. I was going to write a few examples of the parser events and performance. I was going to write a few examples of the parser events and performance. I was going to write
As you see above, the summary is mostly a rephrasing of the sentence was going to write a few examples of the parser events and performance from the source text. The different values max_length, min_legth were tested.
Result of ChatGPT 3.5
Prompt:
Summarize the text above.
Output:
- Logs Overview: Introduces the log records generated by the parser, which are JSON lines produced by the structlog logging library.
- Continuous Running Index: Examines a continuous running index to ensure uninterrupted parser execution.
- Events Frequency: Analyzes the frequency of different events produced by the parser, grouping them by target and thread. The focus is on the task.get event and its distribution across targets.
- Network Latency Time Series: Studies network latency by calculating time deltas between specific events (e.g., data.received and data.get). The data is sorted to ensure correct event order.
- Latency Overall: Computes and visualizes the distribution of network latency, highlighting distinct latency groups.
- Per Thread Latency: Compares network latency patterns between different thread types (producers and consumers), noting differences in latency stability.
- Per Thread Type Distribution and Statistics: Breaks down latency distribution and statistics separately for producer and consumer threads, revealing significant performance disparities.
- Disk Writes Time Series: Analyzes disk write times by calculating time deltas between data.received and data.saved events, showing variations in write workload and identifying patterns.
- Anomalies Detection: Uses quantiles (99%, 95%, 75%) to identify potential anomalies in write times, using the ADTK library for visualization. Focuses on the 95% quantile, concluding that the parser’s performance remains satisfactory.
The analysis highlights the parser’s behavior, performance characteristics, and potential anomalies, providing insights into how the multithreaded parser handles tasks such as data retrieval, processing, and disk writes.
Question
Can i configure led-large-16384 to extract a similar to the ChatGPT result?