Commit Graph
39 Commits
Author SHA1 Message Date
陈一涵 48b8dcd42e [jit kernel] support dtype as a cpp template parameter (#16452) 2026-01-08 13:54:33 +08:00
fy 25fa2ac290 Convert cu_seqlens to CPU for npu_flash_attention_unpad operator (#15434) 2026-01-04 08:16:42 +08:00
Nan JiangandXinyuan Tong 7254986342 [VLM] feat: true on policy for vlm + fsdp (#14636)
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
2026-01-01 16:54:39 -08:00
Yuan Luoandluoyuan.luo 3a42c5e341 [VLM] Adopt jit qk_norm kernel in VLM (#16171)
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2026-01-01 10:10:36 +08:00
Vitaly Tuzov 1048803c1f Reworked fast_pos_embed_interpolate() using torch (#10959) 2025-12-30 14:45:34 +08:00
Yuan Luoandluoyuan.luo b9af8d2eb9 [VLM] Support apply qk norm in multi cuda streams (#15720)
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2025-12-25 14:35:07 +08:00
Yuan Luoandluoyuan.luo 828dec1c7c [VLM] Tiny: Unify VLM environment variables (#15572)
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2025-12-22 17:32:37 +08:00
Yuan Luoandluoyuan.luo 019517a356 [VLM] Support ViT Piecewise CUDA Graph for Qwen3-VL (#15320)
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2025-12-20 21:00:07 +08:00
Yuan Luoandluoyuan.luo 8fa3dc36c5 [VLM] Support cos sin cache for Qwen3-VL & GLM-4.1V (#15205)
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2025-12-18 08:39:50 +08:00
3b8a824b8b [VLM] Support VLM ViT Piecewise CUDA Graph (#14422)
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
Co-authored-by: kousakawang <wanghanpei@bytedance.com>
2025-12-13 20:49:40 +08:00
Netanel Haber 8308cd3632 Support internvl on Blackwell (which doesn't support fa3): add SingletonCache support to Vision{Sdpa|Triton|Ascend}Attention (#13151) 2025-11-26 10:31:45 -08:00
ac81db66c2 [VLM][feat] Support encoder DP for Qwen2.5-VL (#13126)
Co-authored-by: Shangming Cai <csmthu@gmail.com>
Co-authored-by: liusy58 <xiehang.lsy@alibaba-inc.com>
Co-authored-by: Yuan Luo <yuan.luo@hotmail.com>
2025-11-18 16:13:18 +08:00
Morpheus Guoandyuechguo e7b57b0d04 [BugFix] weight load bug when checkpoint expert.gate and exepert.up_proj are not fused (#13113)
Co-authored-by: yuechguo <yuechguo@amd.com>
2025-11-14 00:35:44 -08:00
Morpheus Guoandyuechguo c8547ecddd Enable Aiter Attention for VL model (#12699)
Co-authored-by: yuechguo <yuechguo@amd.com>
2025-11-05 13:01:23 -08:00
Liangsheng Yin 516738b096 Depreate global_server_args_dict (#11528) 2025-10-13 19:34:43 +08:00
Cheng Wan 1bdd010291 Revert "Deprecate global_server_args_dict" (#11520) 2025-10-12 17:40:40 -07:00
Liangsheng Yin 1083e7e3df Deprecate global_server_args_dict (#11331) 2025-10-13 01:20:47 +08:00
ronnie_zhengandwangqihui01 095093ee5a [Ascend] optimize Qwen-vl on Ascend (#10556)
Co-authored-by: wangqihui01 <wangqh10@163.com>
2025-09-22 17:18:16 -07:00
Mick a1f011d09a minor: determine mm attn backend based on platforms (#9303) 2025-08-22 01:08:41 -07:00
Kevin Xiang LiandXiang Li 3b3b3baf9f Double vision prefill throughput by defaulting to optimal vision attention backend (#8484)
Co-authored-by: Xiang (Kevin) Li <lik@nvidia.com>
2025-08-13 02:08:30 -07:00
yhyang201 00da906584 feat: Support DP Attention for step3_vl (#8699) 2025-08-03 19:35:26 +08:00
Wenchen Loandroot ea93079b30 model: adapt mllama4 to VisionAttention (#8512)
Co-authored-by: root <mickjagger19@icloud.com>
2025-08-02 00:39:40 -07:00
b7094a5ef1 model: support intern-s1 (#8350)
Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: zxy <zhou0493@e.ntu.edu.sg>
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: Mick <mickjagger19@icloud.com>
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
2025-07-26 13:48:51 -07:00
Mick 83d87685c5 vlm: adapt internvl to VisionAttention (#6870) 2025-06-11 01:16:04 -07:00
Mick 01dd39bac1 refactor: minor refactors regarding multimodal processing (#6187) 2025-05-17 22:53:20 -07:00
Lianmin Zheng e8e18dcdcc Revert "fix some typos" (#6244) 2025-05-12 12:53:26 -07:00
applesaucethebunandBrayden Zhong d738ab52f8 fix some typos (#6209)
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
2025-05-13 01:42:38 +08:00
fa7d7fd9e5 [Feature] Add FlashAttention3 as a backend for VisionAttention (#5764)
Co-authored-by: othame <chenzhu_912@zju.edu.cn>
Co-authored-by: Mick <mickjagger19@icloud.com>
Co-authored-by: Yi Zhang <1109276519@qq.com>
2025-05-08 10:01:19 -07:00
liwenju0 4d1e52abea Add an assertion to enhance the robustness of the operator (#5736) 2025-04-26 18:09:12 -07:00
Mick e53a0b3d5b [fix] fix mrope positions not picked up (#5265) 2025-04-11 01:29:45 -07:00
Mick 11577cedb7 refactor: bug fixes and refactor for vlm (#4661) 2025-03-22 22:48:49 -07:00
Mick 9d02bb3e2a Urgent model support: support gemma-3-it (#4424) 2025-03-16 17:37:32 -07:00
Mick 01090e8ac3 model: Support Janus-pro (#3203) 2025-03-12 11:02:11 -07:00
Mick ff2ce0b86f refactor: move image processors to separate files (#4229) 2025-03-11 12:35:35 -07:00
Qubitium-ModelCloudandZX-ModelCloud 56a724eba3 [QUANT] Add GPTQModel Dynamic Quantization + lm_head Quantization (#3790)
Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>
Co-authored-by: ZX-ModelCloud <zx@modelcloud.ai>
2025-03-05 01:11:00 -08:00
Mick 99c1b9d2ee fix: apply cache size limit of attention mask for VisionAttention (#3657) 2025-02-19 20:16:48 +08:00
Mick 9f635ea50d [Fix] Address remaining issues of supporting MiniCPMV (#2977) 2025-01-28 00:22:13 -08:00
Lianmin Zheng 89cd923581 Roll back to use vllm custom allreduce (#3006) 2025-01-20 04:03:15 -08:00
3d93f84a00 [Feature] Support minicpmv v2.6 (#2785)
Co-authored-by: Chayenne <zhaochen20@outlook.com>
Co-authored-by: yizhang2077 <1109276519@qq.com>
2025-01-18 14:14:19 -08:00