Datasets:

Modalities:
Tabular
Text
Formats:
csv
Size:
< 1K
Libraries:
Datasets
pandas
License:
kartikey-aa commited on
Commit
1ad5c90
·
verified ·
1 Parent(s): c8cab57

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +24 -0
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: