面向 8×RTX 4090（SM89）的 DeepSeek-V4-Flash-0731 MXFP4 SGLang 实验分支
-
Updated
Sep 14, 2026 - Python
#
sm89
Here are 5 public repositories matching this topic...
QSA HiSparse for SGLang: 256K KV offload, CUDA Graph benchmarks, and patches tested on dual RTX 4090 48GB
qsa cpu-offload tensor-parallelism sparse-attention long-context fp8 gpu-inference llm-inference qwen sglang qwen3 rtx-4090 cuda-graphs sm89 256k-context hisparse rtx-4090-48gb 4090-48gb dual-rtx-4090 kv-cache-offloading
-
Updated
Sep 16, 2026 - Python
Feeding the Tensor Cores: a dense FP16 GEMM for NVIDIA Ada (sm_89) at 96.5% of cuBLAS, and what it teaches about how each GPU generation handles async copy and Tensor Core issue.
-
Updated
Aug 3, 2026 - Cuda
Run DeepSeek-V4-Flash with 1M-token context on pre-Hopper NVIDIA GPUs (sm_89) — a vLLM 0.26.0 overlay replacing all SM90+ DeepGEMM/FlashMLA operators with Triton/TileLang/Marlin kernels + CPU-KV-pool (pinned host memory via CUDA UVA).
-
Updated
Sep 16, 2026 - Python
Add this topic to your repo
To associate your repository with the sm89 topic, visit your repo's landing page and select "manage topics."