[PR #38] [MERGED] feat(deepseek-v4-flash): DeepSeek V4 Flash model + consumer-GPU (sm_120) support #38

New issue

Closed

opened 2026-10-29 21:06:40 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-29 21:06:40 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/sglang/pull/38
Author: @yyj6666667
Created: 5/2/2026
Status: ✅ Merged
Merged: 5/2/2026
Merged by: @yyj6666667

Base: main ← Head: merge-deepseek-v4

📝 Commits (1)

911d5f7 feat(deepseek-v4-flash): DeepSeek V4 Flash model + consumer-GPU (SM_120) support

📊 Changes

162 files changed (+25388 additions, -587 deletions)

View changed files

📝 python/sglang/jit_kernel/.clang-format (+1 -1)
➕ python/sglang/jit_kernel/all_reduce.py (+198 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/c128.cuh (+524 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/c4.cuh (+549 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/common.cuh (+208 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/fused_norm_rope.cuh (+254 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/hash_topk.cuh (+190 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/hisparse_transfer.cuh (+82 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/mega_moe_pre_dispatch.cuh (+219 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/paged_mqa_metadata.cuh (+119 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/rmsnorm.cuh (+133 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/rope.cuh (+169 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/silu_and_mul_masked_post_quant.cuh (+540 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/silu_and_mul_masked_post_quant_tmp.cuh (+371 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/store.cuh (+205 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/topk.cuh (+336 -0)
➕ python/sglang/jit_kernel/csrc/deepseek_v4/topk_v2.cuh (+498 -0)
➕ python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_base.cuh (+27 -0)
➕ python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_pull.cuh (+205 -0)
➕ python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_push.cuh (+253 -0)

...and 80 more files

📄 Description

No description provided

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/kvcache-ai/sglang/pull/38 **Author:** [@yyj6666667](https://github.com/yyj6666667) **Created:** 5/2/2026 **Status:** ✅ Merged **Merged:** 5/2/2026 **Merged by:** [@yyj6666667](https://github.com/yyj6666667) **Base:** `main` ← **Head:** `merge-deepseek-v4` --- ### 📝 Commits (1) - [`911d5f7`](https://github.com/kvcache-ai/sglang/commit/911d5f746567f3a9fa257d5b822a8e7b0b2fe84c) feat(deepseek-v4-flash): DeepSeek V4 Flash model + consumer-GPU (SM_120) support ### 📊 Changes **162 files changed** (+25388 additions, -587 deletions) <details> <summary>View changed files</summary> 📝 `python/sglang/jit_kernel/.clang-format` (+1 -1) ➕ `python/sglang/jit_kernel/all_reduce.py` (+198 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/c128.cuh` (+524 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/c4.cuh` (+549 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/common.cuh` (+208 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/fused_norm_rope.cuh` (+254 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/hash_topk.cuh` (+190 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/hisparse_transfer.cuh` (+82 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/mega_moe_pre_dispatch.cuh` (+219 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/paged_mqa_metadata.cuh` (+119 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/rmsnorm.cuh` (+133 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/rope.cuh` (+169 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/silu_and_mul_masked_post_quant.cuh` (+540 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/silu_and_mul_masked_post_quant_tmp.cuh` (+371 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/store.cuh` (+205 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/topk.cuh` (+336 -0) ➕ `python/sglang/jit_kernel/csrc/deepseek_v4/topk_v2.cuh` (+498 -0) ➕ `python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_base.cuh` (+27 -0) ➕ `python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_pull.cuh` (+205 -0) ➕ `python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_push.cuh` (+253 -0) _...and 80 more files_ </details> ### 📄 Description _No description provided_ --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-29 21:06:40 +00:00