xiaoxiaohehe001 f265a26f8b support mtp rope_3d (#3791) ...

* support mtp rope_3d

* Update speculate_write_cache_with_rope_kernel.cu

2025-09-04 17:18:05 +08:00

..

template_instantiation

support qk norm (#3145)

2025-08-05 16:46:14 +08:00

append_attention_c4_impl.cuh

Modify mask_offset‘s format (#3525)

2025-09-02 03:05:35 -07:00

append_attention_c8_impl.cuh

Modify mask_offset‘s format (#3525)

2025-09-02 03:05:35 -07:00

append_attention_c16_impl.cuh

Modify mask_offset‘s format (#3525)

2025-09-02 03:05:35 -07:00

append_attention_func.cuh

Modify mask_offset‘s format (#3525)

2025-09-02 03:05:35 -07:00

append_attention_kernel.h

…

decode_attention_func.cuh

…

decode_attention_kernel.cu

…

decoder_write_cache_with_rope_impl.cuh

qk norm for speculate decode C16 (#3637)

2025-09-03 14:53:56 +08:00

decoder_write_cache_with_rope_kernel.cu

fix mask_offset in append_attn (#3745)

2025-08-31 15:03:16 +08:00

decoder_write_cache_with_rope_kernel.h

support qk norm (#3145)

2025-08-05 16:46:14 +08:00

encoder_write_cache_with_rope_impl.cuh

fix mask_offset in append_attn (#3745)

2025-08-31 15:03:16 +08:00

encoder_write_cache_with_rope_kernel.h

fix mask_offset in append_attn (#3745)

2025-08-31 15:03:16 +08:00

get_block_shape_and_split_kv_block.cu

[Excutor] Increase buffer size to prevent address corruption; add forward metadata debug tool (#3404)

2025-08-18 16:14:09 +08:00

gqa_rope_write_cache.cu

support fa3 rope3d (#3622)

2025-08-27 11:31:29 +08:00

mem_util.cuh

…

mla_cache_kernel.cu

…

mla_cache_kernel.cuh

…

mma_tensor_op.cuh

…

multi_head_latent_attention_kernel.h

…

pre_cache_len_concat.cu

…

speculate_write_cache_with_rope_impl.cuh

support mtp rope_3d (#3791)

2025-09-04 17:18:05 +08:00

speculate_write_cache_with_rope_kernel.cu

support mtp rope_3d (#3791)

2025-09-04 17:18:05 +08:00

speculate_write_cache_with_rope_kernel.h

support mtp rope_3d (#3791)

2025-09-04 17:18:05 +08:00

utils.cuh

make append_attn supports mask_offset (#3138)

2025-08-14 03:40:55 -07:00