35 commits

Author	SHA1	Message	Date
Azure	ca7366d2db	Merge remote-tracking branch 'upstream/develop-0.2.2' into support-fp8	2025-02-24 11:58:10 +00:00
Azure	581a524f65	Add data loader to read special weights for fp8; Add special weight process script	2025-02-24 11:34:17 +00:00
Azure	7b7c6a657d	Add fp8 linear kernel;\n Add empty cache to fit in 16G VRAM; By 'wkGCaSS - 知乎 https://zhuanlan.zhihu.com/p/25491611225'	2025-02-22 13:05:08 +00:00
Atream	024009675e	Merge branch 'main' into feat-more-context	2025-02-22 06:17:39 +00:00
Atream	5ec33d046d	optimize gguf dequant, save mem, support Q2_K ... use marlin for lm_head, lm_head only calc last token for prefill extend context window to 19K for DeepSeek-V3/R1 within 24GB VRAM	2025-02-22 06:13:01 +00:00
Atream	a529518346	clean PR code and disable flashinfer	2025-02-19 04:42:47 +00:00
ceerrep	2ffb43f974	fix: adapt prefix cache in forward_linux_flashinfer	2025-02-18 11:40:15 +08:00
ceerrep	bb1cadfff3	Merge branch 'fix_precision_MLA' of https://github.com/kvcache-ai/ktransformers into server-prefix-cache	2025-02-17 18:08:04 +08:00
Atream	038bc30888	fix precision bug imported by position_ids in 0.2.0	2025-02-17 09:23:14 +00:00
ceerrep	5ac266085e	fix: use flash_attn for faster prefill	2025-02-17 00:11:24 +08:00
Azure	ff6b265e53	Mock triton mla due to precision issue	2025-02-16 06:03:12 +00:00
Atream	c5f036e8a4	Merge pull request #333 from kvcache-ai/feat_experts_gpu ... toy support for experts on GPU, no CUDA Graph	2025-02-15 23:30:24 +08:00
Atream	c189d55bd1	toy support for experts on GPU, no CUDA Graph	2025-02-15 15:16:00 +00:00
Atream	92399283b6	Update attention.py	2025-02-15 15:43:35 +08:00
Atream	d90749d35d	Update triton_attention.py	2025-02-15 15:41:01 +08:00
Atream	885a91e7db	Merge pull request #294 from kvcache-ai/feat-fast-MLA ... Feat fast mla	2025-02-14 19:40:36 +08:00
Atream	1084d4e4b4	linux support triton MLA kernel	2025-02-14 11:38:55 +00:00
Atream	bb35dc5b0d	init support for MLA using Attention kernel	2025-02-13 15:01:14 +00:00
liam	8d5ebe49ab	📝 ⚡ fix some debug output and update doc	2025-02-13 17:25:12 +08:00
liam	c74453d8ca	📝 add doc support and fix bug in qwen2	2025-02-13 16:37:43 +08:00
liam	83401dbb3b	⚡ ready to publish	2025-02-10 12:29:23 +08:00
Azure	c4d9bc6670	support KExpertsMarlin backend	2025-02-07 05:57:40 +00:00
Azure	907251c743	done support deepseekv3	2025-02-04 15:53:38 +00:00
Azure	f748cd29f0	fix rope; update moegate	2025-02-01 18:05:45 +00:00
Azure	f873558a89	update rope calculation; update modeling.py; update gate for moe	2025-02-01 07:32:21 +00:00
Azure	476b1d8dc6	support deepseekv3; runable but have precition problem	2025-01-31 08:27:24 +00:00
xhedit	234faf7987	typo fix: KMisrtal -> KMistral	2024-09-12 15:58:01 +00:00
Azure	c55de02f7b	fix qlen > 1000 mask is none error	2024-09-02 02:58:10 +00:00
TangJingqi	6735beb5b6	Fix cannot offload whole layer in cpu	2024-08-29 19:10:14 +08:00
chenxl	4d1d561d28	[feature] release 0.1.3	2024-08-28 16:11:43 +00:00
TangJingqi	67043b4b5c	[fix] format classes and files name	2024-08-15 10:44:59 +08:00
Atream	412055d450	[feature] experts can be injected using CPUInfer ... [fix] fix ktransformers interface when use new CUDAGraphRunner [fix] fix YAML and optimize logic, the top rule has the highest priority	2024-08-14 16:10:54 +08:00
chenxl	f5f79f5c0e	[ADD] support multi-gpu qlen>1 q5_k	2024-08-12 11:41:26 +00:00
chenht2022	c1cc7d2cd2	1) Linear and MLP operators support qlen>1; 2) All operators now share a single memory buffer; 3) Refactor CPUInfer submit/sync logic.	2024-08-08 09:04:36 +00:00
chenxl	18c42e67df	Initial commit	2024-07-27 16:06:58 +08:00