 AlpinDale
|
ae04f57ec1
feat: Pipeline Parallel support (#581)
|
hai 7 meses |
|
AlpinDale
|
405bb74612
Control plane comms refactor (#573)
|
hai 7 meses |
 AlpinDale
|
323fe23b21
chore: use 127.0.0.1 for single-node setups
|
hai 7 meses |
|
AlpinDale
|
dfa59bc5f9
fix: 16 GPUs in a cluster
|
hai 7 meses |
|
AlpinDale
|
17eb1b7eb9
chore: remove ray health check
|
hai 7 meses |
|
AlpinDale
|
de62ceb18c
refactor: eliminate parallel worker per-step task scheduling overhead
|
hai 7 meses |
|
AlpinDale
|
9f3d6205ce
fix ray gpu executor
|
hai 7 meses |
|
AlpinDale
|
236be273e5
feat: tensor parallel speculative decoding (#554)
|
hai 7 meses |
|
AlpinDale
|
c6a501f682
add multiprocessing executor; make ray optional
|
hai 7 meses |
|
AlpinDale
|
ef733aee43
implement ExecuteModelData to reduce executor complexity
|
hai 7 meses |
|
AlpinDale
|
7bcf4c3fc9
centralize gpu worker construction
|
hai 7 meses |
|
AlpinDale
|
fb982981ce
num_lookahead_slots in neuron and ray executors
|
hai 7 meses |
|
AlpinDale
|
957ed7d244
type hints
|
hai 7 meses |
|
AlpinDale
|
c21af7acad
feat: `DistributedGPUExecutor` abstract class (#541)
|
hai 8 meses |
|
AlpinDale
|
199e776722
chore: move ray utils to executor dir
|
hai 8 meses |
|
AlpinDale
|
46159b107a
formatting: pt1
|
hai 8 meses |
|
AlpinDale
|
fca911ee0a
vLLM Upstream Sync (#526)
|
hai 8 meses |
|
AlpinDale
|
f894f7b176
Revert "reduce dedupe by wrapping in general worker class"
|
hai 9 meses |
|
AlpinDale
|
082b0b03bc
Revert "actually run the workers"
|
hai 9 meses |
|
AlpinDale
|
36cf32649d
actually run the workers
|
hai 9 meses |
|
AlpinDale
|
9fff6fb507
reduce dedupe by wrapping in general worker class
|
hai 9 meses |
|
AlpinDale
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
hai 10 meses |
|
AlpinDale
|
0f6d56b07f
feat: model executor refactor (#367)
|
hai 11 meses |
|
AlpinDale
|
f8dfac6372
chore: attention refactor and upstream sync apr01 (#365)
|
hai 11 meses |
