vllm-sr

/

Decision-1.0-Route-0.6B

Like 4

Follow

vLLM Semantic Router 462

Updates the fine-tuned weights, with the card, training data list and evaluation pages to match. What changed in training:

• Noul trains for one epoch instead of three.
• Choice options are trained as KEY: description, the form the Decision runtime sends.
• Jailbreak adds real attack messages from the SaTML 2024 LLM CTF, with its utility-chat openings and prompts.chat persona prompts as benign rows.
• PII adds Few-NERD sentences with and without a person name.
• Training uses the suite's training split only; the previous weights also trained on its in-distribution test split.