REAP-pruned DeepSeek V4 Flash at K128 (128 routed experts per MoE layer). Prunes 50% of routed experts via Cerebras REAP (Router-weighted Expert Activation Pruning), preserving all attention, embeddings, shared experts, router, and MTP components.

Drop-in compatible with the standard ds4-engine runtime — uniform IQ2_XXS/Q2_K expert quantization throughout all layers. No per-layer quant dispatch required.

At a Glance


Base model	DeepSeek V4 Flash
Donor GGUF	antirez IQ2XXS-w2Q2K-AProjQ8 (80.8 GiB)
Pruning method	REAP (Cerebras Research)
Routed experts	128 per layer (down from 256)
Kept slots	5,888 / 11,008
Hash-preserved	Layers 0-2 (256 experts each)
Pruned	Layers 3-42 (128 experts each)
Format	ds4-compact-v1 GGUF
File size	46.98 GiB
Quantization	Uniform IQ2_XXS / Q2_K experts in all layers

This is the recommended variant for most users. Uniform expert quantization means it works out of the box with any ds4-engine build. The mixed-precision variant (52 GiB) has Q4_K in layers 37-42 but requires runtime per-layer quant dispatch.

Domain Split (Calibration)

8,000 prompts · 5.0M tokens · 1.3B routed expert observations

Domain	Share
Coding & development	35-40%
Agentic tool-calling	16%
Research & knowledge	15-20%
Math & science	10-15%
Design & planning	5-10%
Trivia & general QA	3-5%

Calibration used the REAP activation_energy_sum2 score metric with 4,096 token context per prompt. Top-to-bottom expert score gap in layer 3: 2,200x (strong pruning signal).

How to Run

Requires eouya2/ds4-for-reaped (ds4 engine with compact GGUF support):

bash

git clone https://github.com/eouya2/ds4-for-reaped
cd ds4-for-reaped
make cuda-spark -j$(nproc)  # DGX Spark / CUDA
# or: make                    # Metal / macOS

./ds4 --cuda -m DeepSeek-V4-Flash-REAP-K128-uniform.gguf --ctx 131072

API server mode:

bash

./ds4-server --cuda -m DeepSeek-V4-Flash-REAP-K128-uniform.gguf \
  --host 0.0.0.0 --port 17777 --ctx 131072

How It Was Built

Donor GGUF: Downloaded antirez IQ2XXS-w2Q2K-AProjQ8 variant (80.8 GiB) — uniform IQ2_XXS/Q2_K experts throughout
Calibration: 8,000 prompts collected and run through ds4's imatrix collector on a DGX Spark (NVIDIA GB10) at 4,096 token context
REAP scoring: Imatrix activation data converted to per-expert REAP scores using activation_energy_sum2 (same calibration used for the mixed-precision variant — REAP scores are quantization-independent)
Pruning: 50% expert removal via ds4_prune_gguf.py from eouya2/reap-for-ds4. Layers 0-2 (hash-routed) preserved. Expert tensors copied byte-for-byte — no dequant/requant.
Output: ds4-compact-v1 GGUF

No fine-tuning. Purely structural expert removal. Weights are unmodified — a subset of the original experts.

Comparison with Mixed-Precision Variant

	Uniform (this)	Mixed Precision
File size	46.98 GiB	52.04 GiB
Expert quants	IQ2_XXS/Q2_K all layers	Q4_K in layers 37-42
Compatibility	Drop-in	Needs quant-aware runtime
Quality	Baseline	Slightly higher in deep layers

Acknowledgments

DeepSeek — DeepSeek V4 Flash base model
antirez — ds4 engine and GGUF quants
eouya2 — ds4-for-reaped and reap-for-ds4
Cerebras Research — REAP (code)
NVIDIA — DGX Spark hardware

sleepyeldrazi/deepseek-v4-flash-reap-k128-Q2-GGUF

DeepSeek V4 Flash — REAP K128 (Uniform)