AlpinDale
|
cda0e93a10
abstract away the platform for device capability
|
7 kuukautta sitten |
AlpinDale
|
ae04f57ec1
feat: Pipeline Parallel support (#581)
|
7 kuukautta sitten |
AlpinDale
|
7d79c0e726
chore: use nvml query to avoid accidental cuda initialization
|
7 kuukautta sitten |
AlpinDale
|
cdff8e89f9
feat: introduce `DraftModelRunner`
|
7 kuukautta sitten |
AlpinDale
|
405bb74612
Control plane comms refactor (#573)
|
7 kuukautta sitten |
AlpinDale
|
25feb1d592
chore: add support for pinning lora adapters in the lru cache
|
7 kuukautta sitten |
AlpinDale
|
af43576da0
feat: add MLPSpeculator speculative decoding support (#572)
|
7 kuukautta sitten |
AlpinDale
|
6a57861fca
feat: initial XPU support via intel_extension_for_pytorch (#571)
|
7 kuukautta sitten |
AlpinDale
|
d0cca80b8b
feat: support sharded tensorizer models
|
7 kuukautta sitten |
AlpinDale
|
eb2c5c77df
feat: enforce the max possible seqlen
|
8 kuukautta sitten |
AlpinDale
|
de62ceb18c
refactor: eliminate parallel worker per-step task scheduling overhead
|
8 kuukautta sitten |
AlpinDale
|
236be273e5
feat: tensor parallel speculative decoding (#554)
|
8 kuukautta sitten |
AlpinDale
|
7bcff4ac03
implement sharded state dict
|
8 kuukautta sitten |
AlpinDale
|
b984fe4a91
refactor custom allreduce to support multiple tp groups
|
8 kuukautta sitten |
AlpinDale
|
be8154a8a0
feat: proper embeddings API with e5-mistral-7b support
|
8 kuukautta sitten |
AlpinDale
|
8ae2cce237
refactor pynccl
|
8 kuukautta sitten |
AlpinDale
|
0e062e66d3
set block size at init
|
8 kuukautta sitten |
AlpinDale
|
8b56dc4347
dict -> torch.Tensor for blocks_to_swap
|
8 kuukautta sitten |
AlpinDale
|
21ce19b3ea
blocks_to_copy dict -> torch.Tensor
|
8 kuukautta sitten |
AlpinDale
|
ef733aee43
implement ExecuteModelData to reduce executor complexity
|
8 kuukautta sitten |
AlpinDale
|
1879e32510
enable all-reduce for multiple tp groups
|
8 kuukautta sitten |
AlpinDale
|
46159b107a
formatting: pt1
|
8 kuukautta sitten |
AlpinDale
|
4c746d8baa
chore: init nccl using the gloo backend
|
8 kuukautta sitten |
AlpinDale
|
fca911ee0a
vLLM Upstream Sync (#526)
|
8 kuukautta sitten |
AlpinDale
|
f894f7b176
Revert "reduce dedupe by wrapping in general worker class"
|
10 kuukautta sitten |
AlpinDale
|
9fff6fb507
reduce dedupe by wrapping in general worker class
|
10 kuukautta sitten |
AlpinDale
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
10 kuukautta sitten |
AlpinDale
|
e3252edd07
fix: remove event and stream, add typing (#382)
|
11 kuukautta sitten |
AlpinDale
|
f8dfac6372
chore: attention refactor and upstream sync apr01 (#365)
|
11 kuukautta sitten |
AlpinDale
|
e42a78381a
feat: switch from pylint to ruff (#322)
|
1 vuosi sitten |