Model benchmarks

hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL local LLM performance

As of September 2026, hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL runs at up to 79.9 tok/s for local inference (best of 4 community benchmark runs across 1 GPU).

OllamaQ4_0
ShareRedditX

Model size

7.46B

Peak speed

79.9 tok/s

Average speed

79.7 tok/s

Avg PP

3111.7 tok/s

Min memory

2.8 GB

Max context

131,072 tokens

Avg output / run

9,239 tokens

Avg runtime / run

2m 13s

Avg quality

64.8

Benchmark runs

4

GPUs tested

1

Quality by task

Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 4 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.

TaskP5 (low)AvgP95 (high)
Overall60.664.869.8
Agent Workflow63.674.180.8
Code Generation43.252.960.0
Role Play & Narrative59.371.581.7
Research & Analysis55.660.864.4

Performance by hardware and tool

Every hardware/tool/quantization combination hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
NVIDIA GeForce RTX 2070 SUPEROllamaQ4_079.9 tok/s79.7 tok/s2.8 GB131,072 tokens64.84

Benchmark runs

All 4 hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL good for coding?
In our benchmarks, hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL scores 52.9/100 for coding. It runs at about 79.7 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MLX-oQ4e is faster (~107.9 tok/s) and still scores well for coding (81.5/100).
Is hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL good for agentic (tool-using) tasks?
In our benchmarks, hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL scores 74.1/100 for agentic workflows. It runs at about 79.7 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MTP-UD-IQ3_XXS is faster (~101.8 tok/s) and still scores well for agentic workflows (88.4/100).
How fast is hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL for local inference?
Across 4 community benchmark runs, hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL reaches up to 79.9 tok/s and averages 79.7 tok/s, with the fastest results on NVIDIA GeForce RTX 2070 SUPER.
How much memory does hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL need?
The leanest observed configuration used about 2.8 GB of memory (quantizations tested: Q4_0).
Which tools have been used to run hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:UD-Q4_K_XL?
Benchmarks were submitted using Ollama. Results are community-contributed and updated as new runs arrive.