Is your feature request related to a problem? Please describe.
The HSTU inference workflows are tightly coupled to the current HSTU dense backbone. Users who want to evaluate a transformer-style decoder with the same recommendation inputs, embedding stack, KV-cache manager, export flow, and serving path must build and maintain a separate integration.
Describe the solution you'd like
A clear and concise description of what you want to happen.
Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
Additional context
Add any other context, code examples, or references to existing implementations about the feature request here.
By submitting this issue, you agree to follow our code of conduct and our contributing guidelines.
Is your feature request related to a problem? Please describe.
The HSTU inference workflows are tightly coupled to the current HSTU dense backbone. Users who want to evaluate a transformer-style decoder with the same recommendation inputs, embedding stack, KV-cache manager, export flow, and serving path must build and maintain a separate integration.
Describe the solution you'd like
A clear and concise description of what you want to happen.
Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
Additional context
Add any other context, code examples, or references to existing implementations about the feature request here.
By submitting this issue, you agree to follow our code of conduct and our contributing guidelines.