Speed test matrix
LFM2.5-1.2B-Instruct 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/LFM2.5-1.2B-Instruct --served-model-name lfm --trust-remote-code --tp 1 --tool-call-parser lfm2 --context-length 128000 --mem-fraction-static 0.865179 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted>
Prefill
prompt tok/s by context length8K context91.6K90 ms TTFT
16K context110.6K147 ms TTFT
32K context114.6K282 ms TTFT
64K context97.8K659 ms TTFT
120K context75.8K1,594 ms TTFT
125K context73.9K1,702 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 | 515.440 ms TTFT | 981.259 ms TTFT | 1,867.051 ms TTFT | 3,514.359 ms TTFT | 6,132.7122 ms TTFT | 9,708.7296 ms TTFT | 14.1K580 ms TTFT | 18.4K1,256 ms TTFT |
| 16K | 504.194 ms TTFT | 570.2122 ms TTFT | 1,774.0149 ms TTFT | 3,198.7192 ms TTFT | 5,300.4327 ms TTFT | 7,956.1385 ms TTFT | 8,926.1910 ms TTFT | 9,599.02,303 ms TTFT |
| 32K | 488.3137 ms TTFT | 656.3165 ms TTFT | 1,612.8Requests queued230 ms TTFT | 2,784.5334 ms TTFT | 4,330.4169 ms TTFT | 5,421.3937 ms TTFT | 6,222.92,646 ms TTFT | 8,341.54,625 ms TTFT |
| 64K | 462.4203 ms TTFT | 290.9255 ms TTFT | 1,326.4377 ms TTFT | 2,168.8701 ms TTFT | 2,455.61,365 ms TTFT | 3,088.72,710 ms TTFT | 4,405.65,194 ms TTFT | Over capacity |
| 120K | 420.5319 ms TTFT | 260.4432 ms TTFT | 1,102.4688 ms TTFT | 1,386.11,292 ms TTFT | 1,543.82,705 ms TTFT | 2,392.85,196 ms TTFT | Over capacity | Over capacity |
Requests queued: the server could not run all requests at once, so some waited in a queue. 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 | 91.6K | — | 90 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 110.6K | — | 147 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 114.6K | — | 282 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 97.8K | — | 659 ms | — | — | 1 | — | — |
| Prefill | 120K | 1 | Valid | 75.8K | — | 1,594 ms | — | — | 1 | — | — |
| Prefill | 125K | 1 | Valid | 73.9K | — | 1,702 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 515.4 | 518.1 | 40 ms | 1.9 ms | 1.0 | 4 | 15,461 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 981.2 | 493.6 | 59 ms | 2.0 ms | 2.0 | 8 | 29,435 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 1,867.0 | 471.8 | 51 ms | 2.1 ms | 4.1 | 12 | 56,011 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 3,514.3 | 442.9 | 59 ms | 2.3 ms | 7.9 | 24 | 105,430 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 6,132.7 | 388.2 | 122 ms | 2.6 ms | 16.0 | 48 | 183,977 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 9,708.7 | 313.5 | 296 ms | 3.2 ms | 33.3 | 64 | 291,262 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 14.1K | 220.3 | 580 ms | 4.5 ms | 67.1 | 64 | 423,077 | 30.0 s |
| Sustained decode | 8K | 128 | Valid | 18.4K | 150.6 | 1,256 ms | 6.6 ms | 126.6 | 128 | 552,209 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 504.1 | 510.4 | 94 ms | 2.0 ms | 1.0 | 4 | 15,123 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 570.2 | 476.2 | 122 ms | 2.1 ms | 73.9 | 4 | 17,105 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 1,774.0 | 451.3 | 149 ms | 2.2 ms | 4.1 | 12 | 53,221 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 3,198.7 | 408.2 | 192 ms | 2.4 ms | 8.0 | 24 | 95,960 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 5,300.4 | 340.8 | 327 ms | 2.9 ms | 16.1 | 32 | 159,015 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 7,956.1 | 254.0 | 385 ms | 3.9 ms | 31.2 | 64 | 238,678 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 8,926.1 | 148.4 | 910 ms | 6.8 ms | 84.2 | 64 | 267,786 | 30.0 s |
| Sustained decode | 16K | 128 | Valid | 9,599.0 | 76.4 | 2,303 ms | 13.1 ms | 192.0 | 0 | 288,001 | 30.0 s |
| Sustained decode | 32K | 1 | Valid | 488.3 | 495.0 | 137 ms | 2.0 ms | 1.0 | 3 | 14,649 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 656.3 | 457.9 | 165 ms | 2.2 ms | 48.7 | 4 | 19,688 | 30.0 s |
| Sustained decode | 32K | 4 | Requests queued | 1,612.8 | 410.9 | 230 ms | 2.4 ms | 3.7 | 12 | 48,385 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 2,784.5 | 356.7 | 334 ms | 2.8 ms | 8.2 | 16 | 83,536 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 4,330.4 | 275.3 | 169 ms | 3.6 ms | 15.8 | 32 | 129,908 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 5,421.3 | 176.6 | 937 ms | 5.7 ms | 37.6 | 32 | 162,650 | 30.0 s |
| Sustained decode | 32K | 64 | Valid | 6,222.9 | 98.6 | 2,646 ms | 10.1 ms | 83.0 | 0 | 186,689 | 30.0 s |
| Sustained decode | 32K | 128 | Valid | 8,341.5 | 65.6 | 4,625 ms | 15.2 ms | 135.2 | 0 | 250,240 | 30.0 s |
| Sustained decode | 64K | 1 | Valid | 462.4 | 471.8 | 203 ms | 2.1 ms | 1.0 | 3 | 13,871 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 290.9 | 262.6 | 255 ms | 5.3 ms | 98.3 | 2 | 8,728 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 1,326.4 | 360.2 | 377 ms | 2.8 ms | 4.3 | 8 | 39,788 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 2,168.8 | 279.3 | 701 ms | 3.6 ms | 7.6 | 16 | 65,064 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 2,455.6 | 176.3 | 1,365 ms | 5.8 ms | 19.6 | 16 | 73,672 | 30.0 s |
| Sustained decode | 64K | 32 | Valid | 3,088.7 | 97.3 | 2,710 ms | 10.3 ms | 44.8 | 0 | 92,652 | 30.0 s |
| Sustained decode | 64K | 64 | Valid | 4,405.6 | 68.3 | 5,194 ms | 14.6 ms | 68.8 | 0 | 132,175 | 30.0 s |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 1 | Valid | 420.5 | 434.5 | 319 ms | 2.3 ms | 0.9 | 3 | 12,615 | 30.0 s |
| Sustained decode | 120K | 2 | Valid | 260.4 | 123.5 | 432 ms | 8.1 ms | 50.8 | 1 | 7,813 | 30.0 s |
| Sustained decode | 120K | 4 | Valid | 1,102.4 | 291.9 | 688 ms | 3.4 ms | 4.1 | 8 | 33,073 | 30.0 s |
| Sustained decode | 120K | 8 | Valid | 1,386.1 | 186.2 | 1,292 ms | 5.4 ms | 9.4 | 8 | 41,582 | 30.0 s |
| Sustained decode | 120K | 16 | Valid | 1,543.8 | 95.4 | 2,705 ms | 10.5 ms | 21.2 | 0 | 46,307 | 30.0 s |
| Sustained decode | 120K | 32 | Valid | 2,392.8 | 73.5 | 5,196 ms | 13.6 ms | 34.5 | 0 | 71,764 | 30.0 s |
| Sustained decode | 120K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.