Hugging Face
Models
Datasets
Spaces
Buckets
Docs
Enterprise
Pricing
Website
Tasks
HuggingChat
Collections
Languages
Organizations
Community
Blog
Posts
Daily Papers
Hardware
Learn
Discord
Forum
GitHub
Solutions
Team & Enterprise
Hugging Face PRO
Enterprise Support
Inference Providers
Inference Endpoints
Storage Buckets
Log In
Sign Up
HenDav
/
value-steer-safety-head
Like
0
value-steer
value-filtered-decoding
safety
vllm
License:
apache-2.0
Model card
Files
Files and versions
xet
Community
Copy to bucket
main
value-steer-safety-head
Commit History
Aug 10
09:37
Replace single canonical head with 6 decode-matched paper heads (mistral+llama x hh-rlhf/beavertails/pku_saferlhf), per-backbone layout + conformal alpha-curve sidecars
HenDav
Verified signature
3515b78
Jul 4
14:17
Upload README.md with huggingface_hub
HenDav
Verified signature
8331bdc
10:45
Upload README.md with huggingface_hub
HenDav
Verified signature
c95d287
10:36
Upload value_head.bin.meta.json with huggingface_hub
HenDav
Verified signature
d1bb07e
10:36
Upload README.md with huggingface_hub
HenDav
Verified signature
70b7360
Jun 30
20:15
Canonical decode-matched safety value head (Mistral-7B, 17k prompts x4); held-out eval + model card
HenDav
Verified signature
f02885d
20:15
initial commit
HenDav
Verified signature
929fc7b
