Model-specific PRA runtime mappings, profiles, components, compatibility metadata, and qualification evidence; not replacement models or fine-tunes.
AI & ML interests
- Progressive Retrieval Attention - Cognitive CoProcessors
Recent Activity
View all activity
Organization Card
Research and runtime infrastructure
EInnovator
Efficient, addressable context for language models.
Progressive Retrieval Attention
PRA separates logical context, retrieval, and model-visible execution. Bundles package structural mappings, optional learned components, profiles, and qualification evidence without duplicating base-model weights.
Current qualified bundle
Qwen3 0.6B
models 16
EInnovator/pra-qwen2-5-1-5b-instruct-bnb-8bit
Updated
EInnovator/pra-gemma3-1b-mlx-8bit
Updated
EInnovator/pra-llama3-2-1b-mlx-8bit
Updated
EInnovator/pra-qwen3-8b-mlx-6bit
Updated
EInnovator/pra-qwen3-14b-mlx-8bit
Updated
EInnovator/pra-qwen3-8b-mlx-8bit
Updated
EInnovator/pra-qwen3-4b-mlx-8bit
Updated
EInnovator/pra-gemma3-1b-mlx-4bit
Updated
EInnovator/pra-llama3-1-8b-mlx-4bit
Updated
EInnovator/pra-qwen3-32b-mlx-4bit
Updated
datasets 0
None public yet