Back to search results
Qwen3-4B-Thinking-2507 · IQ2_XS
Apple M1llama.cpp v0.44L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (IQ2_XS) on Apple M1 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
75.23 tok/s
Decode
Decode speed
381.59 tok/s
Prefill
Prefill speed
0.67 s
TTFT
Time to first token
6.13 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
2 days ago
Last verified
Performance
- llama.cpp · IQ2_XS (current)Decode 75.23 · Prefill 381.59 ·
- llama.cpp · Q4_K_M Decode 59.36 · Prefill 212.58 ·
- vLLM · FP16 Decode 30.09 · Prefill 100.41 ·
- vLLM · IQ2_XS Decode 79.87 · Prefill 436.11 ·
Core figures
- Decode (avg)
- 75.23 tok/s
- Prefill (avg)
- 381.59 tok/s
- TTFT (avg)
- 0.67 s
- VRAM (avg)
- 6.13 GB
- MTP acceptance rate
- —
- TTFB
- 0.6 GB
- Power draw
- 51.2 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-4B-Thinking-2507
- Quantization
- IQ2_XS
- Framework
- llama.cpp
- Version
- v0.44
- Context length
- 2048 tokens
- Batch size
- 512
- GPU layers
- 99
- Flash Attention
- Off
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- Apple M1
- Measured VRAM (avg)
- 6.13 GB
- OS
- macOS 15.0
- Driver
- Apple M1
- CUDA
- N/A
- Power draw
- 51.2 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.44 · macOS 15.0 · CUDA N/A · 2048 ctx
75.23 tok/s
Decode
381.59 tok/s
Prefill
0.67 s
TTFT
6.13 GB
VRAM
—
MTP
51.2 W
Power draw
L2 batch seed 2026-10-16: llama.cpp on hw_id=99 running model_id=57 at IQ2_XS