alibaba / alibaba/conv-llava

Results about ScienceQA

Open
#2 5 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
129
Forks
7
PR merge metrics
No merged PRs in 30d

Description

你好,我将你的配置替换到了llava以后,跑llava的时候,发现prompt似乎对输出无法规范化,之前llava跑SQA,可以直接输出Answer,比如ABCD,但是换了这个baseline以后,原始的prompt无法使得模型直接输出选项,例如,我的结果例子如下:
`id": "5", "prompt": "\nContext: People can use the engineering-design process to develop solutions to problems. One step in the process is testing if a potential solution meets the requirements of the design.\nQuestion: Which of the following could Gordon's test show?\nOptions: (A) if the spacecraft was damaged when using a parachute with a 1 m vent going 200 km per hour (B) how steady a parachute with a 1 m vent was at 200 km per hour (C) whether a parachute with a 1 m vent would swing too much at 400 km per hour\nAnswer with the option's letter from the given choices directly.", "text": "(B) how steady a parachute with a 1 m vent was at 200 km per hour".`

`id": "11", "prompt": "\nContext: N/A\nQuestion: What is the name of the colony shown?\nOptions: (A) Maryland (B) New Hampshire (C) Rhode Island (D) Vermont\nAnswer with the option's letter from the given choices directly.", "text": "The colony shown in the image is New Hampshire. This can be determined by the distinctive shape of the state, which is elongated with a jagged eastern border that follows the coastline, a characteristic feature of New Hampshire's geography. The color green is used to highlight this particular state on the map, distinguishing it from the surrounding areas. Therefore, the correct answer to the question is (B) New Hampshire.",
`
之前llava的tex{}里面是直接选项例如ABCD,现在是各种都有,这是什么问题?

Contributor guide

No contributing guide indexed for this repository

Research direction

No file or test is named in the report. Start by tracing the ScienceQA/LLaVA prompt and generation or evaluation entry points, then compare the previous and replacement configurations. Done means ScienceQA responses reliably follow the requested option-letter format, or the output-format limitation is documented.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.