Back to search results
Qwen3.5-4B · Q4_K_M
NVIDIA H2097GBvLLM v0.50L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3.5-4B (Q4_K_M) on NVIDIA H20 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
68.71 tok/s
Decode
Decode speed
315.97 tok/s
Prefill
Prefill speed
2.49 s
TTFT
Time to first token
8.99 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
16 days ago
Last verified
Performance
- vLLM · Q3_K_S Decode 84.6 · Prefill 335.18 ·
- vLLM · Q4_K_M (current)Decode 68.71 · Prefill 315.97 ·
- Strata · IQ4_XS Decode 97.77 · Prefill 362.6 ·
Core figures
- Decode (avg)
- 68.71 tok/s
- Prefill (avg)
- 315.97 tok/s
- TTFT (avg)
- 2.49 s
- VRAM (avg)
- 8.99 GB
- MTP acceptance rate
- —
- TTFB
- 0.55 GB
- Power draw
- 344.3 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.5-4B
- Quantization
- Q4_K_M
- Framework
- vLLM
- Version
- v0.50
- Context length
- 16384 tokens
- Batch size
- 8
- GPU layers
- 999
- Flash Attention
- Off
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA H20
- Nominal VRAM
- 97 GB
- Measured VRAM (avg)
- 8.99 GB
- OS
- Ubuntu 22.04
- Driver
- 570.86
- CUDA
- 12.4
- Power draw
- 344.3 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.50 · Ubuntu 22.04 · CUDA 12.4 · 16384 ctx
68.71 tok/s
Decode
315.97 tok/s
Prefill
2.49 s
TTFT
8.99 GB
VRAM
—
MTP
344.3 W
Power draw
L2 batch seed 2026-10-19: vLLM on hw_id=56 running model_id=48 at Q4_K_M