minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Clean deprecated DeepSeek V4 Environs (#34926)

Browse Source
This commit is contained in:
Baizhou Zhang
2026-08-17 16:07:00 -07:00
committed by GitHub
parent b42abbb1ba
commit bc312d185d
28 changed files with 187 additions and 639 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/cookbook/autoregressive/DeepSeek/DeepSeek-V4.mdx
+1 -3
View File
@@ -356,9 +356,7 @@ stay in **FP8** — keep `SGLANG_DSV4_FP4_EXPERTS=0`. It requires a `sgl-deep-ge
build with SM90 FP8 MegaMoE support. **Please use the latest image for this
feature.**
Enable the MegaMoE path with `--moe-a2a-backend megamoe` — or equivalently set
`SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE=1`, which auto-configures the same backend:
Enable the MegaMoE path with `--moe-a2a-backend megamoe`
```bash Command
SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK=4096 \
SGLANG_DSV4_FP4_EXPERTS=0 \
docs/cookbook/autoregressive/Moonshotai/Kimi-K3.mdx
+1 -1
View File
@@ -160,7 +160,7 @@ Speculation: DSPARK holds block size + 1 (= 8) intermediate states per request
- No `--enable-symm-mem` under DCP (force-disabled for decode-graph correctness).
- Explicit `tokenspeed_mla` force-rewrites `--kv-cache-dtype` to fp8; the default `cutedsl_mla` serves either dtype.
- 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 `SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE=1` or `--moe-a2a-backend` is set.
- 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.
docs/docs/references/environment_variables.mdx
-5
View File
@@ -381,11 +381,6 @@ SGLang supports various environment variables that can be used to configure its
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>FlashInfer workspace size in bytes (default ≈ 384 MiB).</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}><code>402653184</code></td>
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>SGLANG_PREP_IN_CUDA_GRAPH</code></td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Capture input preparation inside the CUDA graph.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}><code>true</code></td>
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>SGLANG_EAGER_INPUT_NO_COPY</code></td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>In eager forward, wrap the ForwardBatch's own tensors instead of copying them into the CUDA graph buffer registry (skips a per-iter device-to-device copy).</td>