Any support for torch 2.10 and higher vllm？

[NaNillll]

Context/Motivation

Hello, is there any plan for torch 2.10 and higher vllm？Thanks!

Pseudo-code + acceptance criteria [Optional]

No response

[lonexreb]

Hi @NaNillll — wanted to share the current state since the dependency picture is interlocked.

Current pins (pyproject.toml):

• torch==2.9.0
• vllm>=0.13.0,<0.14.0
• torchmonarch==0.2.0
• torchtitan==0.2.0
• torchstore (git-pinned to no-monarch-2026.01.05)

Upstream today:

• PyTorch latest: 2.12.0 (2.10 was 2026-01-21)
• vLLM latest: v0.22.x — and v0.22 pins torch==2.11.0 in requirements/cuda.txt, so it isn't a torch-2.10 path either
• Monarch latest: v0.5.0 vs. forge's v0.2.0
• torchtitan latest tag: v0.2.2 (forge pins v0.2.0)

The blocker isn't torch alone — bumping torch to 2.10+ requires a coordinated bump of vLLM, Monarch, torchtitan, and torchstore, none of which can move unilaterally. Existing tracking issues for parts of this:

• Update monarch to the latest version #766 — Update Monarch to the latest version
• Titan Nightly Implementation in Forge #730 — Titan Nightly Implementation in Forge
• [CI] Ping specific torchstore & torchtitan nightly version  #656 — Ping specific torchstore & torchtitan nightly version
• [ROCm] Upgrade vLLM Version Used in ROCm to v0.13.0 #741 — [ROCm] Upgrade vLLM to v0.13.0

The bigger context — per the README banner, active development on torchforge is paused; LLM training at PyTorch is being consolidated into torchtitan. If you want to track newer torch + vLLM, that's where the work is moving.

Short answer for now: no roadmap to torch 2.10+ in torchforge that I can find.