[Cookbook] Kimi-K3: add measured B300 1x8 Unified 8k/1k speed numbers (#37878)
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
zijiexia
co-authored by
Claude Opus 5
Claude Opus 5
parent
a44bb397a9
commit
3b64169f9d
@@ -81,6 +81,16 @@ import { KimiK3MambaRatioCalculator } from "/src/snippets/_kimi_k3_mamba_ratio_c
|
||||
|
||||
<Deployment config={config} benchmarks={benchmarks} />
|
||||
|
||||
<Note>
|
||||
B300 1×8 Unified speed numbers are measured on `v0.5.18 @ 71de97b2` with `--random-range-ratio 1.0`,
|
||||
`--warmup-requests 64`, `--flush-cache`, at ISL 8192 / OSL 1024. DSPARK cells pin the acceptance
|
||||
length via the serve env `SGLANG_SIMULATE_ACC_LEN=4.5` — they report what block size 7 delivers at
|
||||
that acceptance, not a measured acceptance rate for this workload. Balanced DSPARK adds
|
||||
`--max-running-requests 256`; without it speculation resets the cap to 48. The KDA state pool still
|
||||
clamps admission below that (101 / 68 / 91 / 60 concurrent requests for MXFP4 NOSPEC / MXFP4 DSPARK /
|
||||
NVFP4 NOSPEC / NVFP4 DSPARK), which is why no point past concurrency 64 is published for Balanced.
|
||||
</Note>
|
||||
|
||||
### Mamba ratio calculator
|
||||
|
||||
<KimiK3MambaRatioCalculator />
|
||||
@@ -124,9 +134,10 @@ are scheduled to release by July 27, 2026**. The recipes on this page were valid
|
||||
repository (`moonshotai/Kimi-K3`) and a public `lmsysorg/sglang` image with K3 support will be
|
||||
available at launch.
|
||||
|
||||
Every cell in the Deploy panel above is currently marked **Final Verification In Progress**: the
|
||||
recipe runs, but its serving round on the final weights and current code is still open. Re-measure
|
||||
throughput and accuracy before you rely on any of them.
|
||||
The B300 1×8 `Unified` Low-Latency and Balanced cells are **Verified** — a speed round on the final
|
||||
weights is published below. Every other cell is still marked **Final Verification In Progress**: the
|
||||
recipe runs, but its serving round on the final weights and current code is still open. Accuracy has
|
||||
not been re-measured on any cell — re-measure before you rely on one.
|
||||
</Note>
|
||||
|
||||
**Recommended generation:** `temperature=1.0`, `top_p=0.95`, `presence_penalty=0`, `frequency_penalty=0` (fixed by the model; informational — do not hardcode in sample code).
|
||||
@@ -155,7 +166,7 @@ Speculation: DSPARK holds block size + 1 (= 8) intermediate states per request
|
||||
|
||||
**Context length.** `--context-length` bounds the longest accepted request plus some context-scaled buffers; it does not size the KV pool. For long context the lever that adds capacity is `fp8_e4m3` KV.
|
||||
|
||||
**DSPARK.** Adds `--speculative-algorithm DSPARK` plus the draft checkpoint on top of the showing strategy. Leave `--speculative-draft-attention-backend` unset. No serving round on the final draft checkpoint has landed — measure against the same recipe running NOSPEC before adopting.
|
||||
**DSPARK.** Adds `--speculative-algorithm DSPARK` plus the draft checkpoint on top of the showing strategy. Leave `--speculative-draft-attention-backend` unset. The published B300 DSPARK numbers pin the acceptance length with `SGLANG_SIMULATE_ACC_LEN`, so no measured acceptance rate exists for a real workload yet — measure against the same recipe running NOSPEC before adopting.
|
||||
|
||||
**Per-platform notes:**
|
||||
|
||||
@@ -179,7 +190,7 @@ Speculation: DSPARK holds block size + 1 (= 8) intermediate states per request
|
||||
- Calculator ratios run well above 1 here (`r > 1` is legal): `bfloat16` state buys admission, `fp8` KV buys context.
|
||||
- Don't use EP with an a2a backend: a2a buffers reclaim the KV that DCP buys. Compose only to measure. a2a backend is set when `--moe-a2a-backend` is set.
|
||||
|
||||
No cell has a serving round in this exact shape — treat them as starting points to verify.
|
||||
Outside the two verified B300 1×8 `Unified` cells, no cell has a serving round in this exact shape — treat those as starting points to verify.
|
||||
|
||||
<a id="amd-env" />
|
||||
|
||||
|
||||
@@ -19,4 +19,76 @@ export const benchmarks = [
|
||||
{ match: { hw: "gb200", pdMode: "unified", strategy: "low-latency" } },
|
||||
{ match: { hw: "gb200", pdMode: "unified", strategy: "balanced" } },
|
||||
{ match: { hw: "gb200", pdMode: "unified", strategy: "high-throughput" } },
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "low-latency", quant: "mxfp4", spec: "none" },
|
||||
sglang_version: "v0.5.18 @ 71de97b2",
|
||||
speed: [
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 },
|
||||
ttft_ms: 378, tpot_ms: 8.51, tokens_per_sec_per_gpu: 127 },
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 },
|
||||
ttft_ms: 3539, tpot_ms: 19.47, tokens_per_sec_per_gpu: 785 },
|
||||
],
|
||||
},
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "low-latency", quant: "mxfp4", spec: "dspark" },
|
||||
sglang_version: "v0.5.18 @ 71de97b2",
|
||||
speed: [
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 },
|
||||
ttft_ms: 389, tpot_ms: 2.84, tokens_per_sec_per_gpu: 351 },
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 },
|
||||
ttft_ms: 3942, tpot_ms: 9.88, tokens_per_sec_per_gpu: 1319 },
|
||||
],
|
||||
},
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "low-latency", quant: "nvfp4", spec: "none" },
|
||||
sglang_version: "v0.5.18 @ 71de97b2",
|
||||
speed: [
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 },
|
||||
ttft_ms: 369, tpot_ms: 10.12, tokens_per_sec_per_gpu: 107 },
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 },
|
||||
ttft_ms: 3387, tpot_ms: 20.94, tokens_per_sec_per_gpu: 742 },
|
||||
],
|
||||
},
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "low-latency", quant: "nvfp4", spec: "dspark" },
|
||||
sglang_version: "v0.5.18 @ 71de97b2",
|
||||
speed: [
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 },
|
||||
ttft_ms: 380, tpot_ms: 3.24, tokens_per_sec_per_gpu: 313 },
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 },
|
||||
ttft_ms: 3765, tpot_ms: 9.77, tokens_per_sec_per_gpu: 1345 },
|
||||
],
|
||||
},
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "balanced", quant: "mxfp4", spec: "none" },
|
||||
sglang_version: "v0.5.18 @ 71de97b2",
|
||||
speed: [
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 64 },
|
||||
ttft_ms: 11635, tpot_ms: 40.19, tokens_per_sec_per_gpu: 1395 },
|
||||
],
|
||||
},
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "balanced", quant: "mxfp4", spec: "dspark" },
|
||||
sglang_version: "v0.5.18 @ 71de97b2",
|
||||
speed: [
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 64 },
|
||||
ttft_ms: 12038, tpot_ms: 24.47, tokens_per_sec_per_gpu: 1987 },
|
||||
],
|
||||
},
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "balanced", quant: "nvfp4", spec: "none" },
|
||||
sglang_version: "v0.5.18 @ 71de97b2",
|
||||
speed: [
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 64 },
|
||||
ttft_ms: 11063, tpot_ms: 41.55, tokens_per_sec_per_gpu: 1373 },
|
||||
],
|
||||
},
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "balanced", quant: "nvfp4", spec: "dspark" },
|
||||
sglang_version: "v0.5.18 @ 71de97b2",
|
||||
speed: [
|
||||
{ workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 64 },
|
||||
ttft_ms: 11664, tpot_ms: 22.17, tokens_per_sec_per_gpu: 1946 },
|
||||
],
|
||||
},
|
||||
];
|
||||
|
||||
@@ -454,7 +454,7 @@ export const config = {
|
||||
--dataset-name {{DATASET}} \\
|
||||
--random-input-len {{ISL}} --random-output-len {{OSL}} --random-range-ratio 1.0 \\
|
||||
--num-prompts {{NUM_PROMPTS}} --max-concurrency {{MAX_CONCURRENCY}} \\
|
||||
--flush-cache`,
|
||||
--warmup-requests 64 --flush-cache`,
|
||||
// num_prompts = 5 × concurrency (measured floor 16).
|
||||
numPromptsByConc: { 1: 16, 16: 80, 64: 320, 256: 1280, 1024: 5120 },
|
||||
},
|
||||
@@ -1013,8 +1013,7 @@ export const config = {
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "low-latency" },
|
||||
nnodes: 1,
|
||||
verified: false,
|
||||
verificationStatus: "in-progress",
|
||||
verified: true,
|
||||
env: [],
|
||||
// No --enable-symm-mem: it makes the fused all-reduce auto-probe skip.
|
||||
flags: [
|
||||
@@ -1031,8 +1030,7 @@ export const config = {
|
||||
{
|
||||
match: { hw: "b300", pdMode: "unified", strategy: "balanced" },
|
||||
nnodes: 1,
|
||||
verified: false,
|
||||
verificationStatus: "in-progress",
|
||||
verified: true,
|
||||
env: [],
|
||||
flags: [
|
||||
"--trust-remote-code",
|
||||
|
||||
Reference in New Issue
Block a user