 AlpinDale
|
cabca7383d
fix: use loopback address for single node again
|
hace 5 meses |
|
AlpinDale
|
523ac99aca
chore: pipeline parallel with Ray accelerated dag
|
hace 5 meses |
|
AlpinDale
|
6124140a45
fix: remove error_on_invalid_device_count_status
|
hace 5 meses |
|
AlpinDale
|
8fc6bc6f8e
fix: define self.forward_dag before init_workers_ray
|
hace 5 meses |
|
AlpinDale
|
45a004874c
chore: allow specifying custom Executor
|
hace 5 meses |
|
AlpinDale
|
b7a2d52e47
fix: allow using mp executor for pipeline parallel
|
hace 5 meses |
|
AlpinDale
|
052a6e1eb6
feat: add SPMD worker execution using Ray accelerated DAG
|
hace 5 meses |
|
AlpinDale
|
6f8beb8583
fix: 4-node crash with PP
|
hace 5 meses |
|
AlpinDale
|
23408b9b2b
chore: skip the driver worker
|
hace 5 meses |
|
AlpinDale
|
1562e073c6
fix: ray worker rank assigment
|
hace 5 meses |
|
AlpinDale
|
4c3bb0b436
fix: pipeline parallel on python 3.8 and 3.9
|
hace 5 meses |
|
AlpinDale
|
5257ebce8c
fix: device >= 0 && device < num_gpus INTERNAL_ASSERT FAILED
|
hace 5 meses |
 AlpinDale
|
ae04f57ec1
feat: Pipeline Parallel support (#581)
|
hace 6 meses |
|
AlpinDale
|
405bb74612
Control plane comms refactor (#573)
|
hace 6 meses |
|
AlpinDale
|
323fe23b21
chore: use 127.0.0.1 for single-node setups
|
hace 6 meses |
|
AlpinDale
|
dfa59bc5f9
fix: 16 GPUs in a cluster
|
hace 6 meses |
|
AlpinDale
|
17eb1b7eb9
chore: remove ray health check
|
hace 6 meses |
|
AlpinDale
|
de62ceb18c
refactor: eliminate parallel worker per-step task scheduling overhead
|
hace 6 meses |
|
AlpinDale
|
9f3d6205ce
fix ray gpu executor
|
hace 6 meses |
|
AlpinDale
|
236be273e5
feat: tensor parallel speculative decoding (#554)
|
hace 6 meses |
|
AlpinDale
|
c6a501f682
add multiprocessing executor; make ray optional
|
hace 6 meses |
|
AlpinDale
|
ef733aee43
implement ExecuteModelData to reduce executor complexity
|
hace 6 meses |
|
AlpinDale
|
7bcf4c3fc9
centralize gpu worker construction
|
hace 6 meses |
|
AlpinDale
|
fb982981ce
num_lookahead_slots in neuron and ray executors
|
hace 6 meses |
|
AlpinDale
|
957ed7d244
type hints
|
hace 6 meses |
|
AlpinDale
|
c21af7acad
feat: `DistributedGPUExecutor` abstract class (#541)
|
hace 6 meses |
|
AlpinDale
|
199e776722
chore: move ray utils to executor dir
|
hace 7 meses |
|
AlpinDale
|
46159b107a
formatting: pt1
|
hace 7 meses |
|
AlpinDale
|
fca911ee0a
vLLM Upstream Sync (#526)
|
hace 7 meses |
|
AlpinDale
|
f894f7b176
Revert "reduce dedupe by wrapping in general worker class"
|
hace 8 meses |
