Dedicated Endpoints
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Get help setting up a custom Dedicated Endpoints.
Talk with our engineer to get a quote for reserved GPU instances with discounts.
README
Metrics
| Metric | Value |
|---|---|
| Baseline refusal | 95.8% |
| Edited refusal | 12.4% |
| Harmless KL | 0.133 |
| KL target | 0.060 |
| Preserve rank | 4 |
| Preserve source | none |
| Direction layer | 24 |
| Elapsed | 522.6 sec |
Reproduction
bash
apostate ablate --model google/gemma-4-E4B-it --out C:\Users\Levit\OneDrive\Desktop\apostatehfmodels\gemma-4-e4b-it-apostate --resume --activation-cache-dir C:\Users\Levit\OneDrive\Desktop\apostatehfmodels\gemma-4-e4b-it-apostate\activation_cache
Measurement
| field | value |
|---|---|
| edit type | weight projection |
| refusal judge | classifier + hard refusal guard |
| preservation metric | harmless kl |
Model provider
heterodoxin
Model tree
Base
google/gemma-4-E4B-it
Fine-tuned
this model
Modalities
Input
Text, Image
Output
Text
Pricing
Dedicated Endpoints
View detailsSupported Functionality
Model APIs
Dedicated Endpoints
Container
More information