TPIPS — Activation Distance Text conditioned perceptual image similarity, built on Qwen/Qwen3 VL Embedding 8B. This repo holds the activation dist checkpoint (one of three TPIPS models, each in its own repo — see the table at the bottom). Code and full docs: https://github.com/PeterWang512/TPIPS. Activation distance. A multi layer feature distance over the VLM's per layer hidden states, weighted by a small text conditioned MLP. The pairwise score is dist(a, b) (higher = more similar). Odd one out probabilities are a softmax over the three "other pair" scores divided by the temperature; 2AFC compares the two reference candidate scores. The pairwise score is the model's raw output (temperature is applied at the probability step). Property Value Base model Qwen/Qwen3 VL Embedding 8B Pairwise score dist(a, b) Fine tuning channel lora (LoRA r=16, α=32) + per layer text MLP head Probe layers 20 evenly spaced LLM layers Temperature 0.0025 (applied at the probability step) Prompt X Represent the similarity of the image based on X. Usage flash attn is required (the model runs with flash attention 2); see the Installation guide for an easy matching wheel install. The TPIPS models Model Repo…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy