elephant-1 · FP16
This page aggregates 1 real-world runs of elephant-1 (FP16) on NVIDIA H20 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
25.76 tok/s
Decode
Decode speed
119.08 tok/s
Prefill
Prefill speed
0.5 s
TTFT
Time to first token
37.06 GB
VRAM
VRAM usage
1
Measured runs
1
Independent sources
Other
Source platforms
13 days ago
Last verified
Performance
- vLLM · FP16 (current)Decode 25.76 · Prefill 119.08 ·
- vLLM · IQ2_XS Decode 68.56 · Prefill 374.49 ·
- Ollama · Q5_K_M Decode 36.62 · Prefill 154.85 ·
Core figures
- Decode (avg)
- 25.76 tok/s
- Prefill (avg)
- 119.08 tok/s
- TTFT (avg)
- 0.5 s
- VRAM (avg)
- 37.06 GB
- MTP acceptance rate
- —
- TTFB
- 0.21 GB
- Power draw
- 304.2 W
Configuration
Member-level fields are taken from the most recent run
- Model
- elephant-1
- Quantization
- FP16
- Framework
- vLLM
- Version
- v0.39
- Context length
- 8192 tokens
- Batch size
- 16
- 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)
- 37.06 GB
- OS
- Windows 11 24H2
- Driver
- 572.83
- CUDA
- 12.8
- Power draw
- 304.2 W
Get started
Original model resources
Base model resources may not match this quantization. Use the verified deployment weights above when available.
Sources & evidence
1 measured records in total, each traceable to its original source
v0.39 · Windows 11 24H2 · CUDA 12.8 · 8192 ctx
25.76 tok/s
Decode
119.08 tok/s
Prefill
0.5 s
TTFT
37.06 GB
VRAM
—
MTP
304.2 W
Power draw
L2 batch seed 2026-10-22: vLLM on hw_id=56 running model_id=95 at FP16