1 of 7

Red Hen Meeting

Zhongheng Cheng

Jul 03

2 of 7

Developed JSON reader for llama-index

  • Improved upon the foundation of Rohan’s JSON parser
    • Leave out the irrelevant information and formats
    • Keep core FrameNet data skeleton
  • Applied JSON Reader in llama-index

3 of 7

Developed multi-conversation function

  • Multi-round conversation
  • Store memory of previous conversation

4 of 7

Prompt Examples

In the FrameNet dataset, what frames are related to selling and buying?

5 of 7

Completion of analysis and sentence generation

6 of 7

Attempts towards CoT

Step 1: Think of Frames Involved:

* Color Frame

This involves concepts related to the visual properties of objects, such as hue, saturation, brightness, etc.

* Time Frame

This involves concepts related to the passage of time, such as hours, minutes, schedules, deadlines, etc.

Step 2: Analyze input spaces

Color: The source frame includes elements like red, blue, green, etc.

Time: The source frame includes elements like seconds, hours, schedules, deadlines, etc.

Step 3: Analyze Cross-Space Mapping

The elements of the "color" frame are mapped onto the "time" frame.

For instance, "red time" is analogous to "spending time," and "bright future" is akin to "investing time."

Step 4: Blended Space

In the blended space, color is conceptualized as a temporal concept, with different hues representing different time periods. For example, "red time" could represent the present, while "blue time" could represent the past.

Step 5: Emergent Structure

This blend creates a new understanding where the visual properties of objects are seen through the lens of time.

Step 6: Generate sentence

"The red time was spent wisely."

7 of 7

Next Week

  • Continue reading books and CoT papers
  • Work on prompt generating algorithm