Update TR-HASH Vision card for v6

README.md

@@ -39,9 +39,14 @@ Our work combines model design, dataset engineering, evaluation and inference sy
 
 A deterministic token-routed mixture-of-experts language model, with its base checkpoint, instruction-tuned release, post-training corpus and interactive demonstration.
 
-### [TR-HASH Vision 0.8M](https://huggingface.co/AETHORIA-AI/TR-HASH-Vision-0.8M-VOC)
-
-A compact token-routed object detector with P2 small-object features, STAL-style assignment and one-to-one NMS-free inference. Pascal VOC training is in progress; weights and validated metrics will be published when the run completes.
+### [TR-HASH Vision v6](https://github.com/Complexity-ML/complexity-framework)
+
+A compact hierarchical token-routed vision architecture with shifted-window
+attention, native multi-stage features, lightweight PAN fusion, P2/STAL
+small-object supervision, and optional one-to-one NMS-free inference. The
+0.76M-parameter ImageNet-1K pretraining model transfers a 0.63M tower into an
+approximately 1M-parameter COCO detector. Pretraining is active; weights and
+same-protocol COCO metrics will be published only after validation.
 
 ## Join the community
 

index.html

@@ -206,16 +206,17 @@
     </a>
 
     <p class="label" style="margin-top: 22px;">Vision research</p>
-    <a class="project" href="https://huggingface.co/AETHORIA-AI/TR-HASH-Vision-0.8M-VOC" target="_blank" rel="noreferrer">
+    <a class="project" href="https://github.com/Complexity-ML/complexity-framework" target="_blank" rel="noreferrer">
       <div>
-        <h2>TR-HASH Vision 0.8M</h2>
-        <p>Compact token-routed object detector with small-object features and end-to-end NMS-free inference. Training on Pascal VOC is in progress.</p>
+        <h2>TR-HASH Vision v6</h2>
+        <p>Compact hierarchical token-routed vision model with native multi-scale features, shifted-window attention and a lightweight detector. ImageNet-1K pretraining is active; COCO evaluation comes next.</p>
         <div class="chips">
-          <span class="chip">0.83M parameters</span>
-          <span class="chip">TR-Hash MoE</span>
-          <span class="chip">P2 + STAL</span>
-          <span class="chip">NMS-free</span>
-          <span class="chip">Training in progress</span>
+          <span class="chip">0.76M pretrain</span>
+          <span class="chip">≈1M detector</span>
+          <span class="chip">Hierarchical TR-Hash</span>
+          <span class="chip">P2 + PAN + STAL</span>
+          <span class="chip">DDP + Triton</span>
+          <span class="chip">Evaluation pending</span>
         </div>
       </div>
       <span class="arrow" aria-hidden="true">→</span>