SangBin Cho
SangBin Cho
### What happened + What you expected to happen ### Versions / Dependencies master ### Reproduction script n/a ### Issue Severity None
### What happened + What you expected to happen ``` 2024-07-24 23:46:13,852 INFO compiled_dag_node.py:1242 -- Tearing down compiled DAG Exception ignored in: Traceback (most recent call last): File "/home/ray/default/vllm/vllm/executor/ray_gpu_executor.py", line...
This PR supports - full SPMD (i.e., sampler runs in all workers) when SPMD mode is on - spec decoding works with SPMD + ray - creates a draft worker...
## Why are these changes needed? When a actor is killed before dependencies are resolved, it doesn't send the reply to the caller (there's a wrong assumption here), and the...
### Description Document Python type checking best practices. It is important for mature and stable codebases cc @jjyao ### Link _No response_
### What happened + What you expected to happen This is a real bug. The test checks if the actor who borrows a reference dies, if the borrowed reference is...
### What happened + What you expected to happen We call ray.get on ref del, but if the dag is already teardown, it raises an weird exception ### Versions /...
### What happened + What you expected to happen ``` from time import perf_counter from time import sleep from contextlib import contextmanager from typing import Callable STATIC_SHAPE = False NCCL...