 AlpinDale
|
f91991f584
fix: f-string fixes
|
5 meses atrás |
 AlpinDale
|
5289c14b24
feat: Asymmetric Tensor Parallel (#594)
|
5 meses atrás |
|
AlpinDale
|
dba22e4f83
fix: add zeromq fallback for broadcasting large objects (e.g. vlm images)
|
5 meses atrás |
|
AlpinDale
|
bdf1cc1aec
fix: allow using custom all reduce when pp_size > 1
|
5 meses atrás |
|
AlpinDale
|
ae04f57ec1
feat: Pipeline Parallel support (#581)
|
5 meses atrás |
|
AlpinDale
|
4cdc810b1c
fix: minor TP issues with vision models
|
5 meses atrás |
|
AlpinDale
|
9868bb2290
chore: make it clear that '%' should NOT be in tensor dict keys
|
5 meses atrás |
|
AlpinDale
|
bb4da84623
fix: make sure multi modal kwargs can broadcast properly with ring buffer
|
5 meses atrás |
|
AlpinDale
|
bc5ac9584a
fix: make tensor_dict flattening/unflattening more generic
|
5 meses atrás |
|
AlpinDale
|
abbb730607
feat: support draft model on different tensor parallel size
|
5 meses atrás |
|
AlpinDale
|
e238abf0cc
chore: send and recv helper functions
|
5 meses atrás |
|
AlpinDale
|
1b340083b1
feat: add shm broadcast
|
5 meses atrás |
|
AlpinDale
|
6a57861fca
feat: initial XPU support via intel_extension_for_pytorch (#571)
|
5 meses atrás |
|
AlpinDale
|
cc3486477e
fix: benign multiprocessing error
|
5 meses atrás |
|
AlpinDale
|
1d00b61622
feat: w4a16 support for compressed-tensors
|
5 meses atrás |
|
AlpinDale
|
34b41e0a87
chore: add coordinator to reduce code duplication in tp and pp
|
5 meses atrás |
|
AlpinDale
|
270bd333af
chore: check if process is on the same node
|
5 meses atrás |
|
AlpinDale
|
5b0c11d190
support pipeline parallel pynccl groups
|
6 meses atrás |
|
AlpinDale
|
b984fe4a91
refactor custom allreduce to support multiple tp groups
|
6 meses atrás |
|
AlpinDale
|
8ae2cce237
refactor pynccl
|
6 meses atrás |
|
AlpinDale
|
1879e32510
enable all-reduce for multiple tp groups
|
6 meses atrás |
|
AlpinDale
|
4c746d8baa
chore: init nccl using the gloo backend
|
7 meses atrás |
|
AlpinDale
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
8 meses atrás |
