Skip to content

Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

C

fix(plan): match DeepSeek V4 expert tensor naming (layers.N.ffn.experts.M)

dcutugnod44b09631ce3
diff --git a/c/resource_plan.py b/c/resource_plan.py
index 7a599bfd..ae8a62f3 100644
--- a/c/resource_plan.py
+++ b/c/resource_plan.py
@@ -12,7 +12,7 @@ from pathlib import Path
GB = 1_000_000_000
-EXPERT_RE = re.compile(r"model\.layers\.(\d+)\.mlp\.experts\.(\d+)\.")
+EXPERT_RE = re.compile(r"(?:model\.)?layers\.(\d+)\.(?:mlp|ffn)\.experts\.(\d+)\.")
def _tensor_sizes(path):