AgentKernel 100M Paper QA BitNet Stage7947

Browser-loadable Model Stack BitNet export of the AgentKernel Lite 100M encoder-decoder paper-answer synthesizer.

This model was trained for retrieved-paper answer synthesis and uses the tokenizer-native <AK_END> stop token. The Stage7949 native generation audit passed with:

<AK_END> seen rate: 1.0
answer starts with Answer: rate: 1.0
[P1] citation rate: 1.0
punctuation ending rate: 1.0
internal token leak rate: 0.0
mean compact target F1: 0.7065

The bundle root contains manifest.json, tokenizer/, runtime/, layers/, and dense/ for browser loading.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support