--- tags: - custom_code - pytorch - causal-lm - transformer-alternative - o1-memory - hfp - physics-informed license: agpl-3.0 language: - en --- # Hyper Flux Projection (HFP) - O(1) Memory Causal LM **HFP** is a groundbreaking, physics-inspired causal language model architecture that completely replaces the traditional KV-Cache with thermodynamic context compression, dual-masked attention, and quantum-inspired quantized learning rate schedulers. ## 🚀 Key Innovation: O(1) Memory Scaling Unlike standard Transformers whose VRAM usage grows quadratically `O(N^2)` with sequence length, **HFP maintains a strictly constant `O(1)` memory footprint**, enabling infinite context windows on consumer hardware.  ## Usage Because this model uses a completely novel architecture, you must use `trust_remote_code=True` to load it. ```python import torch from transformers import AutoModelForCausalLM, AutoConfig # Load the architecture model = AutoModelForCausalLM.from_pretrained( "kayrahan35/HFP-O1-Memory-Model", trust_remote_code=True ) print(model) Theory & Paper The theoretical physics foundation of this model can be found here: Hyper Flux Projection Theory ( http://osf.io/xc7e4 ) License GNU AGPL v3.0
