Back to search results
Qwen3-4B-Thinking-2507 · Q4_K_M
NVIDIA RTX 5070 SUPER18GBvLLM 0.7.3L0 Self-reported
This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (Q4_K_M) on NVIDIA RTX 5070 SUPER with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
317.3 tok/s
Decode
Decode speed
2,379.8 tok/s
Prefill
Prefill speed
0.04 s
TTFT
Time to first token
2.9 GB
VRAM
VRAM usage
L0 Self-reported
1
Measured runs
1
Independent sources
Other
Source platforms
Today
Last verified
Performance
- vLLM · Q5_K_M Decode 253.9 · Prefill 1904.3 ·
- vLLM · Q4_K_M (current)Decode 317.3 · Prefill 2379.8 ·
- llama.cpp · Q5_K_M Decode 233 · Prefill 1747.5 ·
- llama.cpp · Q4_K_M Decode 291.2 · Prefill 2184 ·
Core figures
- Decode (avg)
- 317.3 tok/s
- Prefill (avg)
- 2,379.8 tok/s
- TTFT (avg)
- 0.04 s
- VRAM (avg)
- 2.9 GB
- MTP acceptance rate
- —
- TTFB
- — GB
- Power draw
- — W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-4B-Thinking-2507
- Quantization
- Q4_K_M
- Framework
- vLLM
- Version
- 0.7.3
- Context length
- 4096 tokens
- Batch size
- 512
- GPU layers
- 99
- Flash Attention
- On
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA RTX 5070 SUPER
- Nominal VRAM
- 18 GB
- Measured VRAM (avg)
- 2.9 GB
- OS
- Linux
- Driver
- 570.65
- CUDA
- 12.6
Sources & evidence
1 measured records in total, each traceable to its original source
0.7.3 · Linux · CUDA 12.6 · 4096 ctx
317.3 tok/s
Decode
2,379.8 tok/s
Prefill
0.04 s
TTFT
2.9 GB
VRAM
—
MTP
— W
Power draw
L0 理论估算（add-500-records-batch2）