Format-blended SFT: chat fluency + preserved benchmark formats

README.md

@@ -101,3 +101,15 @@ Chimera is the synthesis of a controlled series of mixer swaps (Kuramoto → Qua
   note   = {Research artifact; custom PyTorch architecture (Quazim0t0 / SpikeWhale family)}
 }
 ```
+
+## Update: format-blended SFT
+
+Short instruction/format SFT on a 60/25/15 blend of HuggingFaceTB/smoltalk,
+GSM8K-train ('#### N' reasoning), and MMLU-style ('Answer: <letter>') examples,
+so chat fluency improves while the benchmark output-formats are preserved.
+Held-out (test-split) before->after:
+
+  held-out 2.23->1.34, MMLU acc 0.035->0.250, format 0.140->1.000
+
+Note: fluency + output-format gains. Benchmark *accuracy* remains near the floor
+for a model this size -- the SFT does not add reasoning ability.

chkpt/quazimoto_sft.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba384eef836bfcdebfc87458b40c3c30bfdc7a85542838716f09c8c224cffa0f
-size 768990661
+oid sha256:de6e4755b180edfb9acf30dce5749fe39cdead5196c05ca9547f53391297e73c
+size 768972805