Bump torch from 2.4.0+cu121 to 2.8.0 in /experiments/agentcompany/openhands

[dependabot]

Bumps torch from 2.4.0+cu121 to 2.8.0.

Release notes

Sourced from torch's releases.

PyTorch 2.8.0 Release Notes

• Highlights
• Backwards Incompatible Changes
• Deprecations
• New Features
• Improvements
• Bug fixes
• Performance
• Documentation
• Developers

Highlights

... (truncated)

Commits

• See full diff in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)
  You can disable automated security fix PRs for this repo from the Security Alerts page.

Greptile Summary

This is a Dependabot-generated bump of torch from 2.4.0+cu121 to 2.8.0 in the OpenHands experiment requirements, but the companion PyTorch packages were not updated alongside it.

• torchaudio remains at 2.4.0+cu121 and torchvision at 0.19.0, both aligned to the torch 2.4.x release train — using them with torch 2.8.0 will cause import errors or silent runtime failures due to ABI incompatibilities.
• torch==2.8.0 drops the +cu121 CUDA build suffix that was present on the original pin, while torchaudio still carries +cu121; this inconsistency risks resolving a CPU-only torch wheel in a GPU environment.

Confidence Score: 2/5

Not safe to merge as-is; the environment will likely fail to import torchaudio/torchvision or lose GPU acceleration.

The torch version bump is incomplete: torchaudio and torchvision are left at their 2.4.x counterparts, which are ABI-incompatible with torch 2.8.0. Additionally, dropping the +cu121 CUDA suffix from torch while retaining it on torchaudio creates an inconsistent build environment that could silently install a CPU-only torch wheel.

experiments/agentcompany/openhands/requirements.txt — torchaudio and torchvision pins need to be updated to their torch 2.8.0-compatible versions, and the CUDA build variant should be applied consistently.

Important Files Changed

Filename	Overview
experiments/agentcompany/openhands/requirements.txt	Bumps torch from 2.4.0+cu121 to 2.8.0, but leaves torchaudio at 2.4.0 and torchvision at 0.19.0 (both tied to the 2.4.x release); also drops the +cu121 CUDA build tag from torch while torchaudio retains it.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[pip install -r requirements.txt] --> B[torch==2.8.0\nno CUDA suffix]
    A --> C[torchaudio==2.4.0+cu121\nCUDA build, old version]
    A --> D[torchvision==0.19.0\nold version]

    B -- "ABI mismatch at runtime" --> E[❌ ImportError or wrong dispatch]
    C -- "version mismatch with torch 2.8.0" --> E
    D -- "version mismatch with torch 2.8.0" --> E

    B -- "no +cu121 tag → PyPI resolution" --> F{CUDA wheel\navailable on PyPI?}
    F -- "CPU-only wheel" --> G[❌ GPU acceleration lost]
    F -- "CUDA wheel" --> H[✅ GPU works]

    style E fill:#f66,color:#fff
    style G fill:#f66,color:#fff

Loading

Prompt To Fix All With AI

Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
experiments/agentcompany/openhands/requirements.txt:177-179
**torch/torchaudio/torchvision version mismatch**

`torch` was bumped to 2.8.0 but `torchaudio` remains at 2.4.0 and `torchvision` at 0.19.0 — both of which correspond to the torch 2.4.x release train. These three packages must be kept in lockstep; mismatched versions cause `ImportError` or silent wrong-dispatch bugs at runtime when torchvision/torchaudio call into torch C++ internals. The expected companions for torch 2.8.0 are `torchvision==0.23.0` and `torchaudio==2.8.0`.

### Issue 2 of 2
experiments/agentcompany/openhands/requirements.txt:177-179
**CUDA build variant dropped for torch**

`torch` was changed from `torch==2.4.0+cu121` to `torch==2.8.0` (no CUDA suffix), while `torchaudio==2.4.0+cu121` still carries the `+cu121` CUDA tag. Without the `+cu121` variant specifier (and the corresponding `--extra-index-url https://download.pytorch.org/whl/cu121`), pip will resolve `torch==2.8.0` to whichever wheel is on PyPI — which may be the CPU-only build. If CUDA acceleration is required in this environment, the CUDA-tagged build should be specified consistently across all three PyTorch packages.

_{Reviews (1): Last reviewed commit: "Bump torch in /experiments/agentcompany/..." | Re-trigger Greptile}

Greptile also left 2 inline comments on this PR.

[greptile-apps]

torch/torchaudio/torchvision version mismatch

torch was bumped to 2.8.0 but torchaudio remains at 2.4.0 and torchvision at 0.19.0 — both of which correspond to the torch 2.4.x release train. These three packages must be kept in lockstep; mismatched versions cause ImportError or silent wrong-dispatch bugs at runtime when torchvision/torchaudio call into torch C++ internals. The expected companions for torch 2.8.0 are torchvision==0.23.0 and torchaudio==2.8.0.

Prompt To Fix With AI

This is a comment left during a code review.
Path: experiments/agentcompany/openhands/requirements.txt
Line: 177-179

Comment:
**torch/torchaudio/torchvision version mismatch**

`torch` was bumped to 2.8.0 but `torchaudio` remains at 2.4.0 and `torchvision` at 0.19.0 — both of which correspond to the torch 2.4.x release train. These three packages must be kept in lockstep; mismatched versions cause `ImportError` or silent wrong-dispatch bugs at runtime when torchvision/torchaudio call into torch C++ internals. The expected companions for torch 2.8.0 are `torchvision==0.23.0` and `torchaudio==2.8.0`.

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

CUDA build variant dropped for torch

torch was changed from torch==2.4.0+cu121 to torch==2.8.0 (no CUDA suffix), while torchaudio==2.4.0+cu121 still carries the +cu121 CUDA tag. Without the +cu121 variant specifier (and the corresponding --extra-index-url https://download.pytorch.org/whl/cu121), pip will resolve torch==2.8.0 to whichever wheel is on PyPI — which may be the CPU-only build. If CUDA acceleration is required in this environment, the CUDA-tagged build should be specified consistently across all three PyTorch packages.

Prompt To Fix With AI

This is a comment left during a code review.
Path: experiments/agentcompany/openhands/requirements.txt
Line: 177-179

Comment:
**CUDA build variant dropped for torch**

`torch` was changed from `torch==2.4.0+cu121` to `torch==2.8.0` (no CUDA suffix), while `torchaudio==2.4.0+cu121` still carries the `+cu121` CUDA tag. Without the `+cu121` variant specifier (and the corresponding `--extra-index-url https://download.pytorch.org/whl/cu121`), pip will resolve `torch==2.8.0` to whichever wheel is on PyPI — which may be the CPU-only build. If CUDA acceleration is required in this environment, the CUDA-tagged build should be specified consistently across all three PyTorch packages.

How can I resolve this? If you propose a fix, please make it concise.