Upload QUANTIZATION.md with huggingface_hub

QUANTIZATION.md

@@ -1,11 +1,12 @@
 # Confucius4-R2T2 — Quantization Technical Report & Research Log
 
 > This report documents the empirical quantization research conducted for the Confucius4-R2T2 model in [transcribe.cpp](https://github.com/NairoDorian/transcribe.cpp), leading to the creation of the `r2t2-q4_k_m.gguf` quantization.
+> Companion to [`PORTING.md`](PORTING.md) (which covers porting and parity, not quantization) and to [`QUANTIZATION_ARMS.md`](QUANTIZATION_ARMS.md) (which covers the method generally).
 
 # Confucius4-R2T2 — quantization
 
-Companion to `docs/porting/families/confucius4_r2t2.md` (which covers porting and
-parity, not quantization) and to `docs/tools/quantization-arms.md` (which covers
+Companion to [`PORTING.md`](PORTING.md) (which covers porting and
+parity, not quantization) and to [`QUANTIZATION_ARMS.md`](QUANTIZATION_ARMS.md) (which covers
 the method generally). This file records what was actually measured on R2T2.
 
 **Result: the floor is 1.187 GB — 52% below the 2.478 GB Q8_0 reference.**
@@ -39,7 +40,7 @@ the method generally). This file records what was actually measured on R2T2.
 
 `transcribe-quantize` could not build these arms: it is preset-only and offers no
 per-tensor control, and its preset menu has no Q2_K or Q3_K. See
-`docs/tools/quantization-arms.md` for the `--keep-type` mechanics, the name
+[`QUANTIZATION_ARMS.md`](QUANTIZATION_ARMS.md) for the `--keep-type` mechanics, the name
 matching rules, and the traps.
 
 ## The arm inventory