Beta
Explore
Marketplace
Neural Labs
Chat
Wallet
Docs
llama-2-7b-chat-lr5e-5-gsm8k-lr5e-5-cbwarp_4994_new – AI Model by wvnvwn | AlphaNeural AI
You can deploy this model and start earning money today!
wvnvwn
/
llama-2-7b-chat-lr5e-5-gsm8k-lr5e-5-cbwarp_4994_new
like
0
safetensors
llama
warp
circuit-breakers
llama-2
openai/gsm8k
kmseong/llama2_7b-chat-Safety-FT-lr5e-5
finetune
us
Views
No views yet
Model card
Files and Versions
Community
API
Deploy
Circuit-Breakers WaRP importance-data-size ablation
This model was trained from
kmseong/llama2_7b-chat-Safety-FT-lr5e-5
with the legacy WaRP pipeline.
Phase-1 basis: SVD over all 4,994 Circuit Breakers safety examples
Phase-2 importance:
grad_abs
over 4994 safety examples
Left basis: identity
Frozen coefficient ratio: 0.1
Downstream task: GSM8K, learning rate 5e-5, 3 epochs
Seed: 42; effective batch size: 16
See
cbwarp_experiment.json
for the captured local Phase 1/2/3 metadata.
