Hugging Face's logo
Tarun Reddi's picture
Open to Work
3 3 19

Tarun Reddi PRO

Teen-Different
charvi020's profile picture branikita's profile picture blazej-mako's profile picture
·
https://redditarun.github.io/

AI & ML interests

Generative AI, Modular AI Systems, Reinforcement Learning

Recent Activity

posted an update 13 days ago
Adaptive Attention at Inference Time: Does It Actually Work? A hypernetwork that rewires GPT's value heads on every forward pass. The answer: not a clean win — but not a failure either. Blog post: https://teendifferent.substack.com/p/adaptive-attention-at-inference-time Code: https://github.com/REDDITARUN/a-gpt Weights: https://huggingface.co/Teen-Different/adaptive-gpts What This Is Five small language model variants trained for 12k steps on a 300M token mixed corpus, answering one question: can the residual stream be used to slightly rewrite the model's own computation while it's running? Instead of a fixed W_v for every context, a TinyHeadTransformer hypernetwork generates low-rank (LoRA-style) updates to the value projection of each attention head — conditioned on the current residual stream. Each token gets a dynamically adapted value transformation. The Five Models Base GPT — 28.9M params, 139 tok/s, val loss ~3.82 Matched GPT (+2 layers) — 30.5M params, 204 tok/s, val loss ~3.80 Adaptive GPT — 30.5M params, 38.7 tok/s, val loss ~3.88–3.92 Diffusion GPT — 28.9M params, 110 tok/s, val loss ~5.0–5.2 Adaptive Diffusion GPT — 30.5M params, 40.4 tok/s, val loss ~5.0–5.2 Architecture: 4 layers, 4 heads, d_model=256, context=256, RoPE, GPT-2 tokenizer. How the Hypernetwork Works For each attention head, a TinyHeadTransformer encodes the head's residual stream slice, mean-pools it to a conditioning vector, then projects into low-rank factors A (d×r) and B (r×d) at rank=8. The dynamic value update follows LoRA conventions with alpha/r scaling. B is zero-initialized so the adaptive path starts inert and the model begins as a vanilla GPT — critical for training stability. The diffusion variant uses bidirectional attention, RMSNorm, squared ReLU, and a learned timestep embedding.
updated a model 13 days ago
Teen-Different/adaptive-gpts
published a model 15 days ago
Teen-Different/adaptive-gpts
View all activity

Organizations

Hugging Face Discord Community's profile picture Makora's profile picture Pi4Wear's profile picture Mistral Hack-a-ton 2026's profile picture OpenEnv Community's profile picture