Add MoE architecture docs

moe/README.md

@@ -0,0 +1,34 @@
+# MoE Architecture
+
+The MoE model is the main scaling direction for code completion. The strongest
+current structure is sparse FFN experts with shared attention and optional
+multi-token prediction heads.
+
+## Recommended Structure
+
+- Decoder-only Transformer.
+- RMSNorm before attention and FFN.
+- RoPE positional encoding.
+- Grouped-query attention.
+- SwiGLU experts.
+- Top-2 routing during training.
+- Shared expert for common syntax and indentation.
+- Router z-loss and auxiliary load-balancing loss.
+
+## Routing Metrics
+
+Log:
+
+- Per-layer expert load.
+- Per-expert token fraction.
+- Dropped-token count.
+- Router entropy.
+- Auxiliary loss.
+- Domain-to-expert correlation.
+
+## Failure Modes
+
+- Expert collapse.
+- Router churn.
+- Synthetic Python overfitting.
+- MTP loss overpowering next-token loss.

moe/TRAINING_PLAN.md

@@ -0,0 +1,22 @@
+# MoE Training Plan
+
+## Data Mix
+
+Train only on manifest-verified JSONL checkpoints. Do not point the trainer at
+raw source directories.
+
+## Curriculum
+
+1. Warm up on short and medium FIM examples.
+2. Add long repository-context examples.
+3. Add code generation as an auxiliary continuation task.
+4. Fine-tune with real-code FIM weighted higher than synthetic continuation.
+
+## Defaults
+
+- Context: start at 2048, then ablate 4096.
+- Precision: bf16 on H100.
+- Optimizer: AdamW.
+- Grad clipping: enabled.
+- MoE aux loss: start small and tune from expert load.
+- Top-k: top-2 for training, top-1 as inference ablation only.