---

base_model: inclusionAI/Ling-3.0-flash

tags: [reap, expert-pruning, moe, bailingmoe, gguf]

---

Ling-3.0-flash REAP288 (71B total / 5.1B active) — GGUF

[288 of 512 routed experts kept per layer — 44% of experts pruned]

from inclusionAI/Ling-3.0-flash

(124B total / 5.1B active).

Method: one-shot REAP (Router-weighted Expert Activation Pruning) —

experts scored by router-gate-value × output-L2-norm over calibration data, lowest-scoring deleted.

No fine-tuning, no recovery training.

Calibration: 1M tokens, 50/25/25 ultrachat / wikitext / code

Compatibility

bailingmoe3 support is pending merge upstream in

PR #26608. Once merged, any llama.cpp build

from that point onward loads these files directly:

Until then, use the fork the PR was developed on:

https://github.com/aetherbird/llama.cpp/tree/bailingmoe3-support

Serving with experts in CPU RAM (attention on GPU, experts streamed from RAM):

llama-server -m Ling-3.0-flash-REAP288-71B-A5B-MXFP4.gguf   -ngl 99 -ot "ffn_.*_exps\.weight=CPU" --no-mmap -c 65536 --flash-attn on --jinja

Quants in this repo are all cut from a full-precision master: MXFP4 · Q4_K_M · Q3_K_M · Q2_K

- MXFP4 (experts MXFP4 / rest Q8_0) is the pick for CPU-offload serving.