[sgl-router] Add SGLang-compatible DeepSeek V4.1 Flash rendering (#40532)

This commit is contained in:
Kan Wu
2026-09-21 18:35:09 +08:00
committed by GitHub
parent 0abb251a20
commit 0f6761b54f
9 changed files with 247 additions and 44 deletions
+6 -1
View File
@@ -204,7 +204,7 @@ settings are not inferred from the router's environment. Disabling forwarding pr
engine behavior but does not establish parity for local routing hashes.
Also set `--disable-input-ids-forwarding` for array-only templates: Dynamo may wrap
string content into arrays differently from the worker. Dynamo 5.1.2 does not expose
string content into arrays differently from the worker. The pinned Dynamo renderer does not expose
its conversion flag, so the router cannot automatically block these templates.
Detailed content-format parity coverage follows in #39133.
@@ -221,6 +221,11 @@ official/preview effort profile is detected from the checkpoint's
`config.json`, as in SGLang. Reference prompts live in `tests/fixtures/deepseek/`
and are regenerated by `tests/scripts/generate_deepseek_parity.py`.
V4.1 Flash uses Dynamo's separate V4.1 encoder with SGLang's numeric reasoning
budgets, tool payloads, and `<|System|>` markers. Developer messages and media
are left to the worker (the pinned encoder renders them differently), and a
non-default `SGLANG_DSV41_REASONING_EFFORT` needs the forwarding precautions above.
## Kimi-K3
Kimi-K3 renders through dynamo-render's native XTML formatter with SGLang's