dahus/gemma-4-e2b-it-Q4_K_S-GGUF

Gemma 4 E2B it — Q4_K_S GGUF

4-bit small quantized GGUF version of google/gemma-4-e2b-it.
Slightly smaller and faster than Q4_K_M with near-identical output quality.

Other quantizations in this series:
Q2_K · Q3_K_S · Q3_K_M · Q4_K_M · Q5_K_S · Q5_K_M · Q6_K · Q8

---

File Info

Property	Value
Format	GGUF Q4_K_S
File size	3.37 GB
Bits per weight	~4
Size vs F16	2.8× smaller

---

Benchmark Results

Tested across 4 categories (Math, Logic, Code, Science), 3 prompts each.
Greedy decoding, 200 max new tokens. Metrics compare logit distributions vs F16 baseline.

Results by Category

Category	Speed (tok/s)	SQNR	Top-1 Agreement	KL Divergence
🔢 Math	25.0	19.0 dB	84.1%	0.2770
🧠 Logic	25.0	19.2 dB	78.7%	0.4684
💻 Code	25.2	19.4 dB	81.2%	0.3213
🔬 Science	24.9	18.8 dB	79.4%	0.3157
Overall	25.0	19.10 dB	80.9%	0.3456

---

Quantization Comparison

Model	Size	Speed (tok/s)	vs F16 speed	SQNR	Top-1 Agree	KL Div
F16 (baseline)	8.67 GB	5.7	1.0×	baseline	baseline	baseline
Q2_K	2.99 GB	31.6	5.6×	5.85 dB	32.0%	4.1149
Q3_K_S	3.11 GB	28.9	5.1×	10.12 dB	63.2%	1.2605
Q3_K_M	2.98 GB	27.4	4.8×	13.93 dB	63.2%	1.6747
Q4_K_S (this)	3.37 GB	25.0	4.4×	19.10 dB	80.9%	0.3456
Q4_K_M	3.43 GB	24.0	4.2×	20.33 dB	82.4%	0.3356
Q5_K_S	3.6 GB	21.9	3.9x	23.32 dB	87.7%	0.1547
Q5_K_M	3.63 GB	22.0	3.9×	23.25 dB	86.9%	0.1248
Q8	4.97 GB	16.2	2.9×	37.11 dB	96.0%	0.0171

---

Key Findings

• Quality: 80.9% Top-1 agreement — strong and coherent outputs across all task types
• Speed: 25.0 tok/s — slightly faster than Q4_K_M (24.0 tok/s)
• Size: 3.13 GB — 60 MB smaller than Q4_K_M
• vs Q4_K_M: Marginally lower quality metrics (1.2 dB SQNR, 1.5% Top-1), but faster and smaller; the difference is imperceptible in practice
• Best for: Same use case as Q4_K_M; prefer Q4_K_S if you want a touch more speed or are tight on the 4 GB boundary

---

Usage

# llama.cpp CLI
./llama-cli -m gemma-4-e2b-q4ks.gguf -p "Write a Python function for binary search." -n 200

# llama-cpp-python
from llama_cpp import Llama

llm = Llama(model_path="gemma-4-e2b-q4ks.gguf", n_ctx=2048)
output = llm("Write a Python function for binary search.", max_tokens=200)
print(output["choices"][0]["text"])

---

Hardware

Tested on: CPU inference (llama.cpp)
Context: 2048 tokens | Greedy decoding