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 2026-10-13
stories 6
RSS Archive Reviews

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

verl adds a TPU GRPO recipe, drops SGLANG_USE_CPU_ENGINE

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

lixali kahlun lxb007981
3 people shipped this
lixali @lixali 2 cited kahlun @kahlun 1 cited lxb007981 @lxb007981

verl shipped a runnable Qwen3-4B GRPO script for TPU clusters and removed an environment variable that rollout code used to set unconditionally.

[hardware] feat: enable TPU GRPO with TorchTitan and vLLM via verl-hardware-plugin verl-project/verl

by lixali

examples/tpu/grpo/run_qwen3_4b_torchtitan.sh runs Qwen3-4B-Base GRPO on GSM8K with TorchTitan training, vLLM rollout and verl-hardware-plugin; if you already have a Ray TPU cluster with the plugin, the model and the datasets in place, it is a one-line launch. It builds on prior work in PR #50.

[rollout] remove leftover todo ENV, remove SGLANG_USE_CPU_ENGINE environment (#8069) verl-project/verl

by kahlun

verl/workers/rollout/replica.py::_load_sglang() no longer sets SGLANG_USE_CPU_ENGINE=1 before importing SGLang, along with the now-unused os import. If your SGLang rollout depended on that variable arriving from verl rather than your own launch environment, you now have to set it yourself.

[hardware] feat: enable TPU GRPO with TorchTitan and vLLM via verl-hardware-plugin (#8155) verl-project/verl

by lixali

Same TPU GRPO recipe as the item above; no separate adoption step.

$ 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 13, 2026?
verl shipped a runnable Qwen3-4B GRPO script for TPU clusters and removed an environment variable that rollout code used to set unconditionally. In total, 3 commits and 3 pull requests landed.
Who contributed to Alignment & Post-training on October 13, 2026?
3 developers shipped this update, including lixali, kahlun, and lxb007981.
What were the notable Alignment & Post-training updates?
[hardware] feat: enable TPU GRPO with TorchTitan and vLLM via verl-hardware-plugin, [rollout] remove leftover todo ENV, remove SGLANG_USE_CPU_ENGINE environment (#8069), and [hardware] feat: enable TPU GRPO with TorchTitan and vLLM via verl-hardware-plugin (#8155).
♥

$ 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.