Update README.md
Browse files
README.md
CHANGED
@@ -127,6 +127,30 @@ Examples of other types of questions include:
|
|
127 |
- **Temporal and Conditional Logic Analysis:** Track time-series trends, implement conditional decision rules, and determine threshold-based alerts or actions
|
128 |
- **Research and Classification:** Analyze patterns, classify and identify relevant documents to recall specific information
|
129 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
## Scoring Approach
|
131 |
|
132 |
We use an LLM-based equality checker to evaluate responses:
|
|
|
127 |
- **Temporal and Conditional Logic Analysis:** Track time-series trends, implement conditional decision rules, and determine threshold-based alerts or actions
|
128 |
- **Research and Classification:** Analyze patterns, classify and identify relevant documents to recall specific information
|
129 |
|
130 |
+
**Prompt Template:**
|
131 |
+
|
132 |
+
We load the relevant documents for each question into context in the same prompt as the question text.
|
133 |
+
|
134 |
+
```python
|
135 |
+
documents_text = "\n\n".join(f"BEGIN DOCUMENT {i + 1}:\n{doc}\nEND DOCUMENT {i + 1}" for i, doc in enumerate(docs))
|
136 |
+
prompt = """BEGIN INPUT DOCUMENTS
|
137 |
+
|
138 |
+
{documents_text}
|
139 |
+
|
140 |
+
END INPUT DOCUMENTS
|
141 |
+
|
142 |
+
Answer the following question using the input documents provided above.
|
143 |
+
|
144 |
+
START QUESTION
|
145 |
+
|
146 |
+
{question}
|
147 |
+
|
148 |
+
END QUESTION
|
149 |
+
"""
|
150 |
+
```
|
151 |
+
|
152 |
+
Reported token counts per question are based on the completed prompt, using the `cl100k_base` tokenizer from `tiktoken`.
|
153 |
+
|
154 |
## Scoring Approach
|
155 |
|
156 |
We use an LLM-based equality checker to evaluate responses:
|