Fix an intermittent hang in TRL's colocated vLLM generation when a PEFT model uses tensor parallelism across…
Fix an intermittent hang in TRL's colocated vLLM generation when a PEFT model uses tensor parallelism across…: a task in HF ML Tasksmith (Harbor dataset). Make this transition use the GPU assigned to the local worker explicitly and complete safely before inference begins. Keep the existing…
The task
Make this transition use the GPU assigned to the local worker explicitly and complete safely before inference begins. Keep the existing synchronization limited to the affected configuration: ordinary models and single-device tensor parallelism should retain their current behavior without an extra collective.…
Part of FineEnvs/HF_ML_Tasksmith.