 AlpinDale
|
eb2c5c77df
feat: enforce the max possible seqlen
|
8 months ago |
|
AlpinDale
|
de62ceb18c
refactor: eliminate parallel worker per-step task scheduling overhead
|
8 months ago |
 AlpinDale
|
236be273e5
feat: tensor parallel speculative decoding (#554)
|
8 months ago |
|
AlpinDale
|
7bcff4ac03
implement sharded state dict
|
8 months ago |
|
AlpinDale
|
b984fe4a91
refactor custom allreduce to support multiple tp groups
|
8 months ago |
|
AlpinDale
|
be8154a8a0
feat: proper embeddings API with e5-mistral-7b support
|
8 months ago |
|
AlpinDale
|
8ae2cce237
refactor pynccl
|
8 months ago |
|
AlpinDale
|
0e062e66d3
set block size at init
|
8 months ago |
|
AlpinDale
|
8b56dc4347
dict -> torch.Tensor for blocks_to_swap
|
8 months ago |
|
AlpinDale
|
21ce19b3ea
blocks_to_copy dict -> torch.Tensor
|
8 months ago |
|
AlpinDale
|
ef733aee43
implement ExecuteModelData to reduce executor complexity
|
8 months ago |
|
AlpinDale
|
1879e32510
enable all-reduce for multiple tp groups
|
8 months ago |
|
AlpinDale
|
46159b107a
formatting: pt1
|
8 months ago |
|
AlpinDale
|
4c746d8baa
chore: init nccl using the gloo backend
|
8 months ago |
|
AlpinDale
|
fca911ee0a
vLLM Upstream Sync (#526)
|
8 months ago |
|
AlpinDale
|
f894f7b176
Revert "reduce dedupe by wrapping in general worker class"
|
10 months ago |
|
AlpinDale
|
9fff6fb507
reduce dedupe by wrapping in general worker class
|
10 months ago |
|
AlpinDale
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
10 months ago |
|
AlpinDale
|
e3252edd07
fix: remove event and stream, add typing (#382)
|
11 months ago |
|
AlpinDale
|
f8dfac6372
chore: attention refactor and upstream sync apr01 (#365)
|
11 months ago |
|
AlpinDale
|
e42a78381a
feat: switch from pylint to ruff (#322)
|
1 year ago |
|
AlpinDale
|
9810daa699
feat: INT8 KV Cache (#298)
|
1 year ago |
|
AlpinDale
|
4b80b42362
fix: memory leaks due to nccl cuda graphs (#275)
|
1 year ago |
 Thomas Xin
|
43cf0e98a0
fix: worker initialization on WSL (#260)
|
1 year ago |
|
AlpinDale
|
ea0f57b233
feat: allow further support for non-cuda devices (#247)
|
1 year ago |
|
AlpinDale
|
31c95011a6
feat: FP8 E5M2 KV Cache (#226)
|
1 year ago |
|
AlpinDale
|
641bb0f6e9
feat: add custom allreduce kernels (#224)
|
1 year ago |
|
AlpinDale
|
c0aac15421
feat: S-LoRA support (#222)
|
1 year ago |
|
AlpinDale
|
8fa608aeb7
feat: replace Ray with NCCL for control plane comms (#221)
|
1 year ago |
|
AlpinDale
|
15a0454172
feat: FP8 KV Cache (#185)
|
1 year ago |
