text
stringlengths 0
366
|
|---|
1. Load the Model: You need the model before you can use it.
|
2. Set Prompt/Context: Providing a system prompt helps guide the model's behavior.
|
3. Initial Inference: This is where the model generates the primary response to the user's query.
|
4. Self-Evaluation: This step takes the original query and the model's response and formats them
|
into a new prompt designed to evaluate the response (in this case, for accuracy).
|
Keep in mind that the files currently contain conceptual descriptions and placeholders ({here your logic...}, Example (conceptual):) rather than complete, runnable code for the LLM operations themselves. However, the sequence of steps and the purpose of each cell are well-defined for a prototype aiming to test this self-evaluation approach.
|
This prototype is designed so you can simply paste your current scripts into the coder-model of your choice—whether it's an agent, API, web interface, or one loaded on your own hardware.
|
You can then just ask the model to adapt using the proposed prototype.
|
This first option is a really simple approach that repeats the query after the initial processing, using the same model for self-evaluation.
|
Many corporations already have similar systems deployed, but since companies do not display their auxiliary system scripts, at least for now, I am sharing mine.
|
Ronni Ross
|
2025
|
Φ
|
https://github.com/ronniross
|
Approach 2: Confidence Estimation via a Second, Separate-cloned Model
|
This approach involves using a completely independent machine learning model specifically tasked with analyzing the output of the primary LLM and providing a confidence score.
|
Here's a breakdown of the steps implemented in the cells:
|
1. **Load Models:** You need to load the weights for *both* your primary LLM and the separate confidence estimation model. This is done in Cell 1.
|
2. **Set Prompts/Context:** This involves setting up any necessary system prompts or context for both the primary model and potentially the confidence estimation model.
|
3. **Initial Inference:** The primary LLM generates its initial response to the user's query.
|
4. **Confidence Estimation:** The core of this approach. The original query and the primary model's response are passed to the separate confidence estimation model. This model is prompted to evaluate the response and output a confidence score, ideally as a numerical percentage. This process, including running the second model in parallel and extracting the score.
|
Keep in mind that the files currently contain conceptual descriptions and placeholders ({here your logic...}, Example (conceptual):) rather than complete, runnable code for the LLM operations themselves.
|
However, the sequence of steps and the purpose of each cell are well-defined for a prototype aiming to test this separate-model evaluation approach.
|
This prototype is designed so you can simply paste your current scripts into the coder-model of your choice—whether it's an agent, API, web interface, or one loaded on your own hardware. You can then just ask the model to adapt using the proposed prototype.
|
This second option introduces a separate component with its own processing pipeline, decoupling the confidence estimation from the primary generation process and performed by a model dedicated solely to evaluation. [cite: 2] It does not require internal access to the primary LLM.
|
Ronni Ross
|
2025
|
Φ
|
https://github.com/ronniross
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.