docs: clarify warning — Osaurus loader is Swift-only, sidecar is required

README.md

@@ -25,19 +25,17 @@ tags:
 > quantized weight shards, and the codebook + Hadamard-rotation sidecar
 > (`jangtq_runtime.safetensors`) was uploaded ~2.1h later.
 >
-> **If your local copy is missing `jangtq_runtime.safetensors`,** the Swift
-> JANGTQ loader (and any decoder that reads codebooks from a sidecar
-> rather than re-deriving them at load time) will produce incoherent output.
+> **OsaurusAI bundles are loaded by the Osaurus Swift app via the native
+> JANGTQ runtime — `jangtq_runtime.safetensors` is required.** Without it,
+> the loader cannot reconstruct the codebook used during quantization and
+> output is incoherent.
 >
 > **Action:** re-pull the bundle (or just download the single file):
 > ```bash
 > hf download OsaurusAI/MiniMax-M2.7-Small-JANGTQ jangtq_runtime.safetensors --local-dir <your-dir>
 > ```
-> Bundles without this file but with `model.safetensors.index.json` listing
-> only `*.tq_packed`/`.tq_norms`/`.tq_bits` triplets are also fine for the
-> **Python** runtime — codebooks are recomputed from `(in_features, bits)`
-> via `jang_tools.turboquant.codebook.compute_codebook` at load time. The
-> sidecar exists for the Swift native loader.
+
+
 
 
 <p align="center">