Parser Boundary

Parser Boundary

Coordination issues #8 (GGUF / engram-parser) and #9 (Safetensors).

Layer ownership

Layer	Canonical owner	In this crate today
GGUF magic + v3 header, KV, tensor directory	engram-parser	moe/checkpoint.rs (parse_checkpoint_layout) — transitional
GGML / GGUF value type constants	engram-parser	moe/gguf.rs — transitional
Per-expert raw weight extraction	engram-parser	not implemented here
mmap'd tensor access for router	cortex-tensor	probe_and_map_checkpoint
Dequant to f32	cortex-tensor	moe/dequant.rs
Routing math + family adapters	cortex-tensor	routing.rs, adapter.rs

Freeze policy

While engram-parser#7 lands:

• No new parser paths, dtypes, or GGUF format enhancements in checkpoint.rs, gguf.rs, or widening dequant dtype set
• Known gaps (BF16, Q6_K, IQ3_*, ggml_type_label, wire-type-31 ≠ IQ3_M discipline) park on engram-parser
• Planning tracked with Linear LIM-88 under LIM-9

Safetensors (#9)

Planned dedicated safetensors-parser crate (header + manifest + MoE candidate discovery), one-way copy from corinth-canal reference. No implementation or dependency in cortex-tensor yet.

Related

• Architecture
• Ecosystem

---

Last updated: July 29, 2026 Updated by: Grok Build: Grok 4.5 Package tip reference: 37c5a21 (main)