Search results

Evaluation Time Issue

Status: Open.

#23 In YoungZ365/SOD;

· Lorsense opened on Jul 20, 2026

Qwen3-1.7B SOD does not converge by ~150 steps; AIME avg@32 far below paper (50.83% / 41.72%)

Status: Open.

#22 In YoungZ365/SOD;

· Derek0001 opened on Jul 10, 2026

Source of Tools and Tool-Calling Judgment in Multi-Step Reasoning

Status: Open.

#21 In YoungZ365/SOD;

· Lorsense opened on Jul 8, 2026

Releasing SFT checkpoint and on OPD/SOD training details

Status: Open.

#20 In YoungZ365/SOD;

· ziqi-zhou02 opened on Jul 3, 2026

Training time ~4× longer than reported (8×A800 vs 8×H20) — FP8 / bandwidth difference? Also, step count in run_sod.sh vs Table 4?

Status: Open.

#19 In YoungZ365/SOD;

· Derek0001 opened on Jun 24, 2026

Question regarding SFT checkpoint initialization for RL/Distillation and performance degradation in 1.7B reproduction

Status: Open.

#18 In YoungZ365/SOD;

· Frankkkk233 opened on Jun 24, 2026

Getting tons of Error: Failed to decode tool call, did I configure something wrong?

Status: Open.

#10 In YoungZ365/SOD;

· UrbanFireClash opened on Jun 10, 2026

Clarification on SFT Learning Rate and SOD Model Configuration

Status: Open.

#9 In YoungZ365/SOD;

· SupreCyk opened on Jun 5, 2026

Recommended Sandbox Configuration for SOD Training?

Status: Open.

#2 In YoungZ365/SOD;

· Neebever opened on May 13, 2026