Upload 3 files

README.md

@@ -1,3 +1,35 @@
 ---
 license: mit
+language:
+- en
+pipeline_tag: text-classification
+tags:
+- dristi
+- router
+- system-one
+- distilbert
+- classification
+- active-learning
 ---
+
+# ⚡ DRISTI-Router (V1 Speed Demon)
+
+**Dristi** is an ultra-fast, strictly-typed, non-autoregressive probabilistic decision engine. Built as a "System One" router, it handles routine AI routing tasks in ~40 milliseconds, safely handing off uncertain or "Out-of-Distribution" (OOD) queries to heavy System 2 LLMs.
+
+## Model Details
+* **Architecture:** DistilBERT (Base, Uncased) with Custom Routing Heads (Binary, Choice, Ordinal)
+* **Parameters:** ~66 Million
+* **Latency:** ~40ms (RTX 3050)
+* **License:** MIT
+
+## Features
+* **Mathematical Self-Awareness (OOD Detection):** Rejects alien or unsafe queries using Cosine Similarity.
+* **Calibrated Confidence:** Uses L-BFGS Temperature Scaling.
+* **Self-Healing API:** Designed to actively learn from human feedback in production.
+
+## How to use
+This model requires the custom `DristiModelV03` architecture from the main GitHub repository.
+
+1. Clone the architecture: `git clone https://github.com/SWARAJCHATTARAJ/DRISTI.git`
+2. Place this `pytorch_model.bin` and `calibration.json` into the `checkpoints/` folder.
+3. Run the blazing-fast API!

calibration.json

@@ -0,0 +1,5 @@
+{
+    "binary_temperature": 2.400517702102661,
+    "choice_temperature": 2.7279388904571533,
+    "note": "Temperatures are applied as: logits / temperature before softmax."
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de7a5f033dd730bc34d4575e54091994180e8e5d6af80cd71303be9b59bfbc4c
+size 267864483