Conversation
- Create metric_hps.py using HPSv2 library - Uses ViT-H-14 model with laion2B-s32B-b79K - Add hpsv2 to evaluation dependencies - Register HPSMetric in __init__.py
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 3 potential issues.
Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.
Comment @cursor review or bugbot run to trigger another review on this PR
| # Process the prompt | ||
| text = self.tokenizer([prompt]).to(device=self.device, non_blocking=True) | ||
| # Calculate the HPS | ||
| with torch.amp.autocast(device_type=self.device.type): |
There was a problem hiding this comment.
String device has no .type attribute, causing crash
High Severity
self.device is assigned from set_to_best_available_device(), which returns a str (e.g., "cuda:0"). At line 172, self.device.type is called, but Python strings have no .type attribute — this raises an AttributeError every time update is invoked. It likely needs torch.device(self.device).type or a string split to extract the device type.
Additional Locations (1)
| import hpsv2 | ||
| except ImportError: | ||
| pruna_logger.error("hpsv2 is not installed. Install with: pip install hpsv2") | ||
| raise |
There was a problem hiding this comment.
Redundant try-import after top-level hpsv2 imports
Low Severity
The try: import hpsv2 guard inside __init__ is dead code. The module already imports from hpsv2.src.open_clip import ... and from hpsv2.utils import ... at the top level, so a missing hpsv2 package would cause an ImportError at module load time, long before __init__ is reached.
Additional Locations (1)
| text = self.tokenizer([prompt]).to(device=self.device, non_blocking=True) | ||
| # Calculate the HPS | ||
| with torch.amp.autocast(device_type=self.device.type): | ||
| outputs = model(image_tensor, text) |
There was a problem hiding this comment.
Parameter outputs shadowed by model result in loop
Low Severity
The local variable outputs on line 173 reuses the name of the method parameter outputs. While it doesn't cause a functional bug currently (since the parameter isn't read after this point), the shadowing is confusing and fragile — any future refactoring that accesses the original outputs after the loop would silently get the wrong value.
davidberenstein1957
force-pushed
the
feat/metric-hps
branch
from
63cb3b6 to
caf9f3b
Compare
…ore References) Co-authored-by: Cursor <cursoragent@cursor.com>
1 participant
Add HPSMetric using hpsv2 for human preference scoring.