Chakragram/core-llm-e2b-seed

CORE.llm E2B Seed

E2B (Extract-to-Bridge) extraction of nucleus-imprint-v1 f16. MatFormer slice — 30/35 layers (skipping 20-24).

Purpose: Knowledge seed for CORE.llm fine-tuning. Contains extracted kernel knowledge from the Nucleus AI OS coding-dev orbit.

Contents:

• core-llm-v1-e2b-f16.gguf — 12 GB f16 GGUF (847 tensors, 35 layers)
• core_kernel.jsonl — 3,294 lines, 13 immutable kernel docs (~206 KB)
• train.py — Unsloth QLoRA fine-tune script (rank=64)

Training target: core-llm-v1-e2b-core-q4_k.gguf (QLoRA on kernel corpus)

Pipeline: f16 → Q4_K quantize → QLoRA (1 epoch, LR=2e-4, rank=64) → export GGUF