anthropic model (Sonet tested) and a asset file added #323
Conversation
| return { | ||
| "author": "Arabic Language Technologies, QCRI, HBKU", | ||
| "model": "claude-3-5-sonnet-20240620", | ||
| "description": "Anthropic model", |
There was a problem hiding this comment.
I think a more informative description is needed?
There was a problem hiding this comment.
I think the asset name isn't consistent with what we had before, e.g., "Adult_GPT4_ZeroShot"
There was a problem hiding this comment.
this is one of the previous asset name for subjectivity, same convention here, right ?
CT23Subjectivity_Jais13b_ZeroShot
There was a problem hiding this comment.
yeah but your asset name has "Sonet" in it, not sure why when the mode is Claude
| prompt = f""" | ||
| Classify the following Arabic 'sentence' as subjective or objective. Provide only the label. | ||
| Provide your response in the following JSON format: {{"label": "your label"}}. | ||
| Please provide JSON output only. No additional text. |
There was a problem hiding this comment.
Why json? I think adding this part made the prompt too long.
There was a problem hiding this comment.
For original implementation, we also asked to provide explanation, score etc. For which json structure is important to get them right.
|
|
||
| def post_process(response): | ||
| data = response["content"][0]["text"].lower() | ||
| data = json.loads(data) |
There was a problem hiding this comment.
From my experience, loading model output as json directly might break,
No description provided.