RepoJournal
Alignment & Post-training Alignment & Post-training
11 wires and counting

$ follow Alignment & Post-training

Keep up with Alignment & Post-training in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-10-12
stories 4
RSS Archive Reviews

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

Chunked-entropy backward fix for Megatron, plus a Quay description for Ascend images

By RepoJournal · Filed · About Alignment & Post-training · Composed from the cited sources · methodology

lxb007981 aoshen02
2 people shipped this
lxb007981 @lxb007981 1 cited aoshen02 @aoshen02 1 cited

verl landed a backward-pass fix for chunked vocab-parallel entropy under the Megatron engine and put the quay.io/ascend/verl repository description under version control, a quiet day for operators of the trainer.

[megatron] fix: backward through chunked vocab-parallel entropy verl-project/verl

by aoshen02

Operators running the Megatron engine with `entropy_from_logits_with_chunking=True` no longer hit a RuntimeError in backward when entropy is part of the graph; the failure came from applying `_VocabParallelEntropy` to chunks that are views of one tensor, so the gradient reference was modified in place before it was needed.

[ci] feat: Add Ascend Quay repository description verl-project/verl

by lxb007981

The description for quay.io/ascend/verl now lives in the repo rather than being set by hand on the registry, so anyone pulling the Ascend image sees a description that stays in step with the source tree.

$ ls alignment/reviews/ # the step-back reads

the week in review verl adds TPU SFT via TorchTitan and ETP checkpoint conversion
Full archive → Weekly reviews →

Quick answers

What shipped in Alignment & Post-training on October 12, 2026?
verl landed a backward-pass fix for chunked vocab-parallel entropy under the Megatron engine and put the quay.io/ascend/verl repository description under version control, a quiet day for operators of the trainer. In total, 2 commits and 2 pull requests landed.
Who contributed to Alignment & Post-training on October 12, 2026?
2 developers shipped this update, including lxb007981 and aoshen02.
What were the notable Alignment & Post-training updates?
[megatron] fix: backward through chunked vocab-parallel entropy and [ci] feat: Add Ascend Quay repository description.
♥

$ log Alignment & Post-training --recent

verl rejects conflicting vLLM engine_kwargs, open-instruct adds MILES OPD behind a Megatron learner 8 shipped FSDP2 CPU offload was silently corrupting parameters, and MILES gets FlashAttention 8 shipped Fused cross-entropy lands on Ascend NPUs and Megatron, checkpoint offload stops assuming CUDA 26 shipped DeepSeek MTP support fixed for vLLM and Megatron backends 10 shipped

$ log --all # elsewhere on the wire

Interpretability nnsight 0.8.0 cuts import time from 7.6s to 1.6s Agent Security Approval digests get RFC 8785 float serialization; sandbox-runtime 0.0.80 kills a 300 ms path

$ ls showcases/

Safety Evals AI Red Teaming Agent Security Interpretability

$ init your-desk

RepoJournal can write one of these for any repo you care about. Sign in with GitHub, pick your repos and sources, and your desk gets a wire like this one every morning.