Commit
•
63cfe59
1
Parent(s):
65d8c96
Upload prompt template grounding_nli_json_alt.yaml
Browse files- grounding_nli_json_alt.yaml +59 -23
grounding_nli_json_alt.yaml
CHANGED
@@ -1,27 +1,63 @@
|
|
1 |
prompt:
|
2 |
-
template:
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
**Input
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
template_variables:
|
26 |
- user_request
|
27 |
- context_document
|
|
|
1 |
prompt:
|
2 |
+
template: |-
|
3 |
+
You are a helpful and harmless AI assistant. You will be provided with a textual context and a model-generated response.
|
4 |
+
Your task is to analyze the response sentence by sentence and classify each sentence according to its relationship with the provided context.
|
5 |
+
|
6 |
+
**Instructions:**
|
7 |
+
|
8 |
+
1. **Decompose the response into individual sentences.**
|
9 |
+
2. **For each sentence, assign one of the following labels:**
|
10 |
+
* **`supported`**: The sentence is entailed by the given context. Provide a supporting excerpt from the context.
|
11 |
+
* **`unsupported`**: The sentence is not entailed by the given context. Provide an excerpt that is close but does not fully support the sentence.
|
12 |
+
* **`contradictory`**: The sentence is falsified by the given context. Provide a contradicting excerpt from the context.
|
13 |
+
* **`no_rad`**: The sentence does not require factual attribution (e.g., opinions, greetings, questions, disclaimers). No excerpt is needed for this label.
|
14 |
+
|
15 |
+
3. **For each label, provide a short rationale explaining your decision.** The rationale should be separate from the excerpt.
|
16 |
+
|
17 |
+
**Input Format:**
|
18 |
+
|
19 |
+
The input will consist of two parts, clearly separated:
|
20 |
+
|
21 |
+
* **Context:** The textual context used to generate the response.
|
22 |
+
* **Response:** The model-generated response to be analyzed.
|
23 |
+
|
24 |
+
**Output Format:**
|
25 |
+
|
26 |
+
For each sentence in the response, output a JSON object with the following fields:
|
27 |
+
|
28 |
+
* `"sentence"`: The sentence being analyzed.
|
29 |
+
* `"label"`: One of `supported`, `unsupported`, `contradictory`, or `no_rad`.
|
30 |
+
* `"rationale"`: A brief explanation for the assigned label.
|
31 |
+
* `"excerpt"`: A relevant excerpt from the context. Only required for `supported`, `unsupported`, and `contradictory` labels.
|
32 |
+
|
33 |
+
Output each JSON object on a new line.
|
34 |
+
|
35 |
+
**Example:**
|
36 |
+
|
37 |
+
**Input:**
|
38 |
+
|
39 |
+
```
|
40 |
+
Context: Apples are red fruits. Bananas are yellow fruits.
|
41 |
+
|
42 |
+
Response: Apples are red. Bananas are green. Enjoy your fruit!
|
43 |
+
```
|
44 |
+
|
45 |
+
**Output:**
|
46 |
+
|
47 |
+
{"sentence": "Apples are red.", "label": "supported", "rationale": "The context explicitly states that apples are red.", "excerpt": "Apples are red fruits."}
|
48 |
+
{"sentence": "Bananas are green.", "label": "contradictory", "rationale": "The context states that bananas are yellow, not green.", "excerpt": "Bananas are yellow fruits."}
|
49 |
+
{"sentence": "Enjoy your fruit!", "label": "no_rad", "rationale": "This is a general expression and does not require factual attribution.", "excerpt": null}
|
50 |
+
|
51 |
+
**Now, please analyze the following context and response:**
|
52 |
+
|
53 |
+
**User Query:**
|
54 |
+
{{user_request}}
|
55 |
+
|
56 |
+
**Context:**
|
57 |
+
{{context_document}}
|
58 |
+
|
59 |
+
**Response:**
|
60 |
+
{{response}}
|
61 |
template_variables:
|
62 |
- user_request
|
63 |
- context_document
|