FEAT Add JailbreakV_28k dataset from HF #1098
Open
+317
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
romanlutz
reviewed
Sep 23, 2025
romanlutz
reviewed
Sep 26, 2025
| "fetch_jbb_behaviors_dataset", | ||
| "fetch_jbb_behaviors_by_harm_category", | ||
| "fetch_jbb_behaviors_by_jbb_category", | ||
| "fetch_jailbreakv_28k_dataset", |
There was a problem hiding this comment.
mind keeping these alphabetical? I realize we missed out on that before but no better time to fix it than now 🙂
Comment on lines
+62
to
+63
| The dataset license: mit | ||
| authors: Weidi Luo, Siyuan Ma, Xiaogeng Liu, Chaowei Xiao, Xiaoyu Guo |
There was a problem hiding this comment.
Suggested change
| The dataset license: mit | |
| authors: Weidi Luo, Siyuan Ma, Xiaogeng Liu, Chaowei Xiao, Xiaoyu Guo | |
| The dataset license: MIT | |
| Authors: Weidi Luo, Siyuan Ma, Xiaogeng Liu, Chaowei Xiao, Xiaoyu Guo |
Comment on lines
+115
to
+134
| if image_abs_path: | ||
| group_id = uuid.uuid4() | ||
| text_seed_prompt = SeedPrompt( | ||
| value=item.get(text_field, ""), | ||
| harm_categories=[policy], | ||
| prompt_group_id=group_id, | ||
| data_type="text", | ||
| **common_metadata, # type: ignore[arg-type] | ||
| ) | ||
| image_seed_prompt = SeedPrompt( | ||
| value=image_abs_path, | ||
| harm_categories=[policy], | ||
| prompt_group_id=group_id, | ||
| data_type="image_path", | ||
| **common_metadata, # type: ignore[arg-type] | ||
| ) | ||
| seed_prompts.append(text_seed_prompt) | ||
| seed_prompts.append(image_seed_prompt) | ||
| else: | ||
| missing_images += 1 |
There was a problem hiding this comment.
Suggested change
| if image_abs_path: | |
| group_id = uuid.uuid4() | |
| text_seed_prompt = SeedPrompt( | |
| value=item.get(text_field, ""), | |
| harm_categories=[policy], | |
| prompt_group_id=group_id, | |
| data_type="text", | |
| **common_metadata, # type: ignore[arg-type] | |
| ) | |
| image_seed_prompt = SeedPrompt( | |
| value=image_abs_path, | |
| harm_categories=[policy], | |
| prompt_group_id=group_id, | |
| data_type="image_path", | |
| **common_metadata, # type: ignore[arg-type] | |
| ) | |
| seed_prompts.append(text_seed_prompt) | |
| seed_prompts.append(image_seed_prompt) | |
| else: | |
| missing_images += 1 | |
| if not image_abs_path: | |
| missing_images += 1 | |
| continue | |
| group_id = uuid.uuid4() | |
| text_seed_prompt = SeedPrompt( | |
| value=item.get(text_field, ""), | |
| harm_categories=[policy], | |
| prompt_group_id=group_id, | |
| data_type="text", | |
| **common_metadata, # type: ignore[arg-type] | |
| ) | |
| image_seed_prompt = SeedPrompt( | |
| value=image_abs_path, | |
| harm_categories=[policy], | |
| prompt_group_id=group_id, | |
| data_type="image_path", | |
| **common_metadata, # type: ignore[arg-type] | |
| ) | |
| seed_prompts.append(text_seed_prompt) | |
| seed_prompts.append(image_seed_prompt) | |
There was a problem hiding this comment.
i.e., move the second case to first and remove indentation
Comment on lines
+140
to
+144
| if not seed_prompts: | ||
| raise ValueError( | ||
| "JailBreakV-28K fetch produced 0 prompts. " | ||
| "Likely caused by all items returned after filtering having invalid image paths." | ||
| ) |
There was a problem hiding this comment.
how many are currently missing? Should we have a cutoff (>0) at which point it should error out?
There was a problem hiding this comment.
Currently, the vast majority are missing, making this dataset not as useful as previously expected. I have started a discussion on HF about adding the full images folder which currently is only contained a zip file held in a separate cloud drive.
| assert sum(p.data_type == "text" for p in jailbreakv_28k.prompts) == len(jailbreakv_28k.prompts) / 2 | ||
| assert sum(p.data_type == "image_path" for p in jailbreakv_28k.prompts) == len(jailbreakv_28k.prompts) / 2 | ||
| except Exception as e: | ||
| pytest.skip(f"Integration test skipped due to: {e}") |
There was a problem hiding this comment.
Good point, I am not sure why I was simply going off of what I thought was the convention based on previously merged custom integration tests (jbb dataset integration)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This PR adds support for the JailbreakV_28k dataset to PyRIT.
Addresses #1007
Changes Made:
Files Added/Modified:
Tests and Documentation