skblv/gemma-3-27b-it-lora-cholect50-verbs

Gemma 3 27B-it LoRA — CholecT50 verb recognition

LoRA adapter plus a 10-way linear classification head on google/gemma-3-27b-it, fine-tuned for multi-label surgical-action (verb) presence on CholecT50 frames.

This is the “Gemma 3 27B fine-tuned” row on the Recommendations tab of the SDSC × Chicago Booth surgical video understanding leaderboard.

Prompt example

This closed-set example mirrors the leaderboard format, not a text-input API for this checkpoint.

[surgical frame]

Which surgical actions are being performed in this cholecystectomy frame?
Select every matching label.

- grasp
- retract
- dissect
- coagulate
- clip
- cut
- aspirate
- irrigate
- pack
- idle

Files

• adapter_model.safetensors / adapter_config.json — LoRA (r=128, alpha=256) on q/k/v/o_proj and out_proj
• classifier.pt — linear head over the 5376-d hidden state (10 actions)
• model_config.json — action vocabulary and training hyperparameters

You need the adapter and the classification head; this is not a generative JSON adapter.

Training

• 10 epochs, lr 5e-6, effective batch size 2, seed 42
• 80,940 training frames / 19,923 validation frames

Evaluation

Full CholecT50 verbs validation split (95% bootstrap CI):

Metric	Value
Exact match	60.6% (60.0–61.3)
Micro-averaged F1	78.8% (78.4–79.2)

References

• Skobelev, K., Fithian, E., Baranovski, Y., et al. A Comparative Study in Surgical AI: Potential and Limitations of Data, Compute, and Scaling. arXiv:2603.27341, 2026.
• Dataset: Nwoye, C. I., Yu, T., Gonzalez, C., et al. Rendezvous: Attention Mechanisms for the Recognition of Surgical Action Triplets in Endoscopic Videos. Medical Image Analysis, 78, 102433 (2022).

Limitations

Research baseline only. Not a medical device.