[Deps] Bump FlashInfer to 0.7.1 (#43494)

Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
This commit is contained in:
Mohammad Miadh Angkad
2026-10-11 23:23:11 -07:00
committed by GitHub
co-authored by Mohammad Angkad
parent f9bd83e7a9
commit 130e11d50e
6 changed files with 8 additions and 18 deletions
+1 -1
View File
@@ -39,7 +39,7 @@ dependencies = [
"einops",
"fastapi",
"flash-attn-4>=4.0.0b18",
"flashinfer_python[cu13]==0.7.0.post1", # keep it aligned with jit-cache version in Dockerfile
"flashinfer_python[cu13]==0.7.1", # keep it aligned with jit-cache version in Dockerfile
"gguf",
"humming-kernels[cu13]==0.1.12",
"interegular",
@@ -108,13 +108,8 @@ pip install git+https://github.com/thu-ml/SpargeAttn.git --no-build-isolation
SM120 requires FlashInfer's CuTe-DSL SM120 Sage backend and
`quantize_sage_qkv_sm120`, added in
[FlashInfer #4691](https://github.com/flashinfer-ai/flashinfer/pull/4691).
The default FlashInfer `0.7.0.post1` pin does not include these APIs. After installing
SGLang, install the FlashInfer source revision used for SM120 validation:
```bash
pip install "git+https://github.com/flashinfer-ai/flashinfer.git@6a84331eb6013e5e61018dc2be532ae90520d30f"
```
[FlashInfer #4691](https://github.com/flashinfer-ai/flashinfer/pull/4691),
which the default FlashInfer install includes.
Enable it on SM90 or SM120 with:
+1 -1
View File
@@ -1911,7 +1911,7 @@ def _set_envs_and_config(server_args: ServerArgs):
):
assert_pkg_version(
"flashinfer_python",
"0.7.0.post1",
"0.7.1",
"Please uninstall the old version and "
"reinstall the latest version by following the instructions "
"at https://docs.flashinfer.ai/installation.html.",