Alignment & Post-training — full archive
11 daily briefings · 2 reviews.
$ ls alignment/reviews/ # the step-back reads
- week verl adds TPU SFT via TorchTitan and ETP checkpoint conversion
- week verl adds score centering to bypass-mode REINFORCE
- verl rejects conflicting vLLM engine_kwargs, open-instruct adds MILES OPD behind a Megatron learner
- FSDP2 CPU offload was silently corrupting parameters, and MILES gets FlashAttention
- Fused cross-entropy lands on Ascend NPUs and Megatron, checkpoint offload stops assuming CUDA
- DeepSeek MTP support fixed for vLLM and Megatron backends
- verl adds a TPU GRPO recipe, drops SGLANG_USE_CPU_ENGINE
- Chunked-entropy backward fix for Megatron, plus a Quay description for Ascend images
- veRL merges veomni checkpoints, automates Ascend doc translation
- TPU SFT via TorchTitan, Megatron TP/ETP checkpoint conversion
- verl opens the profiler and attention paths to non-CUDA backends
- verl docs rename the Ascend 950 product line
- Score centering lands for bypass-mode REINFORCE in verl
Never miss a Alignment & Post-training briefing.
Get Alignment & Post-training's week in review every Monday - and security advisories the day they land. Free, unsubscribe in one click.
One click. Every Monday: what mattered across the projects you've been reading (or our AI picks) — security news same-day. You refine anytime. Choose specific journals · Read a sample issue →