M2-2B: 30K-step byte-level checkpoint (val PPL 2.73, bf16)

README.md

@@ -16,16 +16,18 @@ pipeline_tag: text-generation
 the M2 experimental line's 2b configuration trained via `m2_training`.
 Byte-level tokenizer (UTF-8 bytes, vocab 256).
 
-**Status at upload:** research milestone — first complete 30,000-step run.
+**Status at upload:** research milestone — 60,000-step checkpoint.
 
 ## Measured
 
 | Milestone | val PPL (byte-level, WikiText-103) |
 |---|---|
 | 10K steps (feasibility) | 4.53 |
-| **30K steps (this checkpoint)** | **2.73** |
+| 30K steps | 2.73 |
+| **60K steps (this checkpoint)** | **2.556** |
 
-≈ 1.45 BPB. Training: bf16 model, AdamW8bit, lr 1e-4, seq 128,
+Context-flat: PPL 2.72 / 2.72 / 2.68 at seq 128 / 256 / 512 (no length degradation).
+Training: bf16 model, AdamW8bit, lr 1e-4, seq 128,
 batch 8 × grad_accum 4, on a single A100-PCIE-40GB.
 
 ## Boundaries (honest)