Speed test matrix
LFM2.5-230M on H200 NVL 141GB
sglang · 0.5.6.post3.dev9265+g05c584c44 · BF165 contexts (8K–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K, 125K
Hardware
- Type
- Discrete GPU
- GPU
- H200 NVL
- VRAM
- 141 GB
- OS
- linux
Engine
- Engine
- sglang
- Version
- 0.5.6.post3.dev9265+g05c584c44
- Quantization
- BF16
- Backend
- cuda
- Repository
- github.com/sgl-project/sglang
- Build
- cuda-13.2
- Commit
05c584c44fb0
Run
- Served model
- lfm
- Revision
- main
- Benchmark tool
- llm-inference-bench v0.4.29
- Decode test
- Fixed duration
- Prefill mode
- Measured during decode warm-up
- Time per measurement
- 30 s
- Max output
- 4,096 tokens
Launch command
env SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 "$HOME/env_h200-liquidai-sglang/bin/sglang" serve --model-path LiquidAI/LFM2.5-230M --served-model-name lfm --trust-remote-code --tp 1 --tool-call-parser lfm2 --context-length 128000 --mem-fraction-static 0.865061 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted>
Prefill
prompt tok/s by context length8K context124.2K66 ms TTFT
16K context180K90 ms TTFT
32K context218.2K148 ms TTFT
64K context200.3K322 ms TTFT
120K context154.5K782 ms TTFT
125K context156.8K802 ms TTFT
Sustained decode
total tok/s · rows are context length, columns are simultaneous requests| Context | 1 | 2 | 4 | 8 | 16 | 32 | 64 | 128 |
|---|---|---|---|---|---|---|---|---|
| 8K | 672.841 ms TTFT | 1,263.240 ms TTFT | 2,462.443 ms TTFT | 4,778.256 ms TTFT | 7,991.733 ms TTFT | 11.9K189 ms TTFT | 15.2K519 ms TTFT | 20.5K1,063 ms TTFT |
| 16K | 655.364 ms TTFT | 907.788 ms TTFT | 2,282.590 ms TTFT | 4,230.8178 ms TTFT | 6,692.8197 ms TTFT | 9,388.2554 ms TTFT | 11.2K1,192 ms TTFT | 11.6K2,301 ms TTFT |
| 32K | 624.994 ms TTFT | 1,146.8113 ms TTFT | 2,065.9139 ms TTFT | 3,544.3186 ms TTFT | 5,131.4222 ms TTFT | 6,711.11,064 ms TTFT | 7,796.22,397 ms TTFT | 9,053.94,745 ms TTFT |
| 64K | 584.4148 ms TTFT | 504.8225 ms TTFT | 1,736.9236 ms TTFT | 2,680.6337 ms TTFT | 3,568.1856 ms TTFT | 4,132.41,672 ms TTFT | 3,332.35,190 ms TTFT | Over capacity |
| 120K | 520.5241 ms TTFT | 99.3377 ms TTFT | 1,144.6500 ms TTFT | 1,568.31,032 ms TTFT | 2,016.32,620 ms TTFT | 1,790.45,396 ms TTFT | Over capacity | Over capacity |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.
TTFT is measured after the benchmark warms the server's prompt cache, so it mostly excludes prefill.
All measurements
Latencies are medians. Total tok/s is output across all simultaneous requests.| Phase | Context | Requests at once | Status | Total tok/s | Per request tok/s | TTFT | Time between tokens | Average active requests | Completed requests | Output tokens | Measured for |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Prefill | 8K | 1 | Valid | 124.2K | — | 66 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 180K | — | 90 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 218.2K | — | 148 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 200.3K | — | 322 ms | — | — | 1 | — | — |
| Prefill | 120K | 1 | Valid | 154.5K | — | 782 ms | — | — | 1 | — | — |
| Prefill | 125K | 1 | Valid | 156.8K | — | 802 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 672.8 | 677.4 | 41 ms | 1.5 ms | 1.0 | 5 | 20,185 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 1,263.2 | 636.2 | 40 ms | 1.6 ms | 2.0 | 10 | 37,895 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 2,462.4 | 620.3 | 43 ms | 1.6 ms | 4.0 | 20 | 73,873 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 4,778.2 | 602.7 | 56 ms | 1.7 ms | 8.0 | 32 | 143,348 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 7,991.7 | 504.2 | 33 ms | 2.0 ms | 15.9 | 64 | 239,752 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 11.9K | 384.2 | 189 ms | 2.6 ms | 32.8 | 96 | 358,463 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 15.2K | 263.0 | 519 ms | 3.8 ms | 73.9 | 64 | 455,990 | 30.0 s |
| Sustained decode | 8K | 128 | Valid | 20.5K | 171.3 | 1,063 ms | 5.8 ms | 126.7 | 128 | 614,064 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 655.3 | 662.8 | 64 ms | 1.5 ms | 1.0 | 5 | 19,660 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 907.7 | 616.2 | 88 ms | 1.6 ms | 44.2 | 6 | 27,229 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 2,282.5 | 584.7 | 90 ms | 1.7 ms | 4.0 | 16 | 68,474 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 4,230.8 | 539.3 | 178 ms | 1.9 ms | 8.0 | 32 | 126,932 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 6,692.8 | 425.1 | 197 ms | 2.4 ms | 15.9 | 48 | 200,783 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 9,388.2 | 300.1 | 554 ms | 3.3 ms | 32.5 | 64 | 281,651 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 11.2K | 190.9 | 1,192 ms | 5.2 ms | 71.5 | 64 | 334,911 | 30.0 s |
| Sustained decode | 16K | 128 | Valid | 11.6K | 90.4 | 2,301 ms | 11.1 ms | 170.4 | 0 | 348,441 | 30.0 s |
| Sustained decode | 32K | 1 | Valid | 624.9 | 636.8 | 94 ms | 1.6 ms | 1.0 | 5 | 18,745 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 1,146.8 | 581.0 | 113 ms | 1.7 ms | 2.0 | 8 | 34,403 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 2,065.9 | 525.8 | 139 ms | 1.9 ms | 4.0 | 16 | 61,977 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 3,544.3 | 450.0 | 186 ms | 2.2 ms | 8.0 | 24 | 106,331 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 5,131.4 | 323.9 | 222 ms | 3.1 ms | 15.8 | 32 | 153,944 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 6,711.1 | 209.9 | 1,064 ms | 4.8 ms | 31.6 | 32 | 201,329 | 30.0 s |
| Sustained decode | 32K | 64 | Valid | 7,796.2 | 120.1 | 2,397 ms | 8.3 ms | 67.9 | 0 | 233,850 | 30.0 s |
| Sustained decode | 32K | 128 | Valid | 9,053.9 | 73.6 | 4,745 ms | 13.6 ms | 128.0 | 0 | 271,608 | 30.0 s |
| Sustained decode | 64K | 1 | Valid | 584.4 | 595.2 | 148 ms | 1.7 ms | 1.0 | 4 | 17,533 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 504.8 | 338.0 | 225 ms | 4.2 ms | 75.3 | 2 | 15,144 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 1,736.9 | 445.1 | 236 ms | 2.2 ms | 3.9 | 12 | 52,098 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 2,680.6 | 339.9 | 337 ms | 2.9 ms | 7.8 | 16 | 80,416 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 3,568.1 | 227.2 | 856 ms | 4.4 ms | 15.9 | 16 | 107,043 | 30.0 s |
| Sustained decode | 64K | 32 | Valid | 4,132.4 | 133.6 | 1,672 ms | 7.5 ms | 31.8 | 32 | 123,970 | 30.0 s |
| Sustained decode | 64K | 64 | Valid | 3,332.3 | 54.6 | 5,190 ms | 18.3 ms | 96.0 | 0 | 99,968 | 30.0 s |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 1 | Valid | 520.5 | 538.2 | 241 ms | 1.9 ms | 1.0 | 4 | 15,614 | 30.0 s |
| Sustained decode | 120K | 2 | Valid | 99.3 | 49.2 | 377 ms | 20.3 ms | 90.6 | 0 | 2,978 | 30.0 s |
| Sustained decode | 120K | 4 | Valid | 1,144.6 | 345.6 | 500 ms | 2.9 ms | 5.9 | 8 | 34,339 | 30.0 s |
| Sustained decode | 120K | 8 | Valid | 1,568.3 | 209.4 | 1,032 ms | 4.9 ms | 9.6 | 8 | 47,031 | 30.0 s |
| Sustained decode | 120K | 16 | Valid | 2,016.3 | 125.4 | 2,620 ms | 8.0 ms | 17.7 | 10 | 59,710 | 29.6 s |
| Sustained decode | 120K | 32 | Valid | 1,790.4 | 59.7 | 5,396 ms | 16.8 ms | 42.0 | 0 | 53,712 | 30.0 s |
| Sustained decode | 120K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.