[Question] I found that ois is numerically equivalent to tis during training

Maintainers usually reply within 1 day

Assessment

Research direction

Description

Your Question

Hi,
I am training Qwen3.5-4B with Slime v0.2.4 with the following config:

• async
• update weight step = 2
• use tis
  When I look at the logged metrics, I found that OIS is numerically equal to TIS during training, which should not happen. Is this a bug or something expected? Thanks in advance for any help.

Here are the figures:

What I've Tried

I read through the code and found they should not be equal to each other

Environment (if relevant)

• slime version: v0.2.4
• Python version:
• PyTorch version:
• CUDA/ROCm version:
• GPU type and count:
• OS:

Additional Context

No response

Pre-submission Checklist

• I have read the CONTRIBUTING.md and understand the collaboration scope.
• I have read the documentation and FAQ and my question is not answered there.
• I have searched for existing issues and my question has not been asked before.

Getting set up

First steps

1. Read the whole issue, then the project's contributing guide.
2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
3. Fork the repository and make your change on a branch.
4. Open a pull request that references the issue number.

More from THUDM/slime

• [Bug] Math reward accepts distinct large integers after float normalization

  Possibly taken @lll-hhh claimed this 2 days ago. Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel

  Difficulty 2/5 1-3 hours Newbie friendliness 78/100

  THUDM/slime#2509 · Opened Oct 30, 2026

  Maintainers usually reply within 1 day

• gpqa: when no option letter can be extracted, a single-letter gold label is substring-matched against the whole reply, so a reply that never answers can score 1.0

  Possibly taken @shaurya416 claimed this 38 days ago. Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel

  Difficulty 2/5 1-3 hours Newbie friendliness 88/100

  THUDM/slime#2409 · Opened Sep 24, 2026

  Maintainers usually reply within 1 day

• [Bug] shipped dynamic-sampling filters crash on fan-out groups (list[list[Sample]]) from multi-turn agent rollouts

  Possibly taken @QIU-Guanzong claimed this 50 days ago. Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel

  Difficulty 2/5 1-3 hours Newbie friendliness 84/100

  THUDM/slime#2374 · Opened Sep 9, 2026

  Maintainers usually reply within 1 day

• [Bug] W&B initialization fails with wandb 0.28.2 because wandb.util.generate_id was removed

  Possibly taken @cyyecao-lappland claimed this 68 days ago. Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel
  bug

  Difficulty 1/5 1-3 hours Newbie friendliness 90/100

  THUDM/slime#2332 · 1 comment · Opened Aug 26, 2026

  Maintainers usually reply within 1 day

• [Bug] solution_str = sample.prompt + sample.response

  Possibly taken @mvanhorn claimed this 132 days ago. Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel
  bug

  Difficulty 2/5 1-3 hours Newbie friendliness 68/100

  THUDM/slime#1829 · 1 comment · Opened Apr 13, 2026

  Maintainers usually reply within 1 day

All issues in THUDM/slime

Similar issues

• URLValidator ignores an explicit empty scheme allowlist

  Possibly taken @IlyaKhalafi claimed this today. Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel

  Difficulty 2/5 1-3 hours Newbie friendliness 85/100

  tortoise/tortoise-orm#2393 · Opened Nov 2, 2026

  Maintainers usually reply within 2 days

• Bench image workflow can publish frappe/bench:null when the release lookup fails

  Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel
  bug

  Difficulty 2/5 1-3 hours Newbie friendliness 62/100

  frappe/frappe_docker#2046 · Opened Nov 2, 2026

  Maintainers usually reply within 1 day

• Weekly link check failed

  Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel

  Difficulty 2/5 1-3 hours Newbie friendliness 62/100

  awslabs/operational-readiness-review-custom-war-lens#12 · Opened Nov 2, 2026

• complete_box_iou_loss / complete_box_iou: NaN gradients for zero-height boxes, NaN for point boxes

  Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel

  Difficulty 2/5 1-3 hours Newbie friendliness 62/100

  pytorch/vision#9730 · Opened Nov 2, 2026

  Maintainers usually reply within 1 day

• [SPEC/SMU-AXI] State the AXI ID widths of the smu_axi_in and smu_axi_out boundary ports

  Open

  ⋯Report a problem with this issue

  • Is something wrong?

  • Already taken
  • Not beginner-friendly
  • Stale or invalid
  • Something else…
  • Send Cancel
  AXI SMU SPEC

  Difficulty 2/5 1-3 hours Newbie friendliness 62/100

  tenstorrent/tt-oca-harness#3556 · Opened Nov 2, 2026

  Maintainers usually reply within 1 day

More Python issues