• GPU: NVIDIA RTX 4080 SUPER (16 GB VRAM)
• CPU: Intel Core Ultra 9 285K @ 3.70 GHz
• RAM: 64 GB DDR5
• OS: Windows 11 with WSL2 (Ubuntu 22.04)
• CUDA: 12.4
• Torch: 2.5.1+cu124
• Port 32111 (TCP): ✅ Fully open (inbound and outbound)
• Running locally at home, not cloud/VPS.

My Config - Model B 1.5

Model arguments

model_revision: main
torch_dtype: bfloat16
attn_implementation: flash_attention_2
bf16: true
tf32: true

Dataset arguments

dataset_id_or_path: 'openai/gsm8k'

Training arguments

max_steps: 100 # Original 450
num_train_epochs: 1
gradient_accumulation_steps: 4
gradient_checkpointing: true
gradient_checkpointing_kwargs:
use_reentrant: false
learning_rate: 8.5e-7 # 1.0e-6 as in the deepseek math paper 5-e7 from https://hijkzzz.notion.site/unraveling-rlhf-and-its-variants-engineering-insights#147d9a33ecc9806090f3d5c749d31f05
lr_scheduler_type: cosine
warmup_ratio: 0.03

GRPO arguments

use_vllm: true
num_generations: 4
per_device_train_batch_size: 4
beta: 0.001 # 0.04 as in the deepseek math paper 0.001 from https://hijkzzz.notion.site/unraveling-rlhf-and-its-variants-engineering-insights#147d9a33ecc9806090f3d5c749d31f05
max_prompt_length: 256
max_completion_length: 768

Logging arguments

logging_strategy: steps
logging_steps: 2
report_to:

• tensorboard
  save_strategy: "steps"
  save_steps: 25
  seed: 42

Script arguments

public_maddr: "/ip4/38.101.215.12/tcp/30002"
host_maddr: "/ip4/0.0.0.0/tcp/38331"
max_rounds: 10000

Model-specific arguments

model_name_or_path: Gensyn/Qwen2.5-1.5B-Instruct
output_dir: runs/gsm8k/multinode/Qwen2.5-1.5B-Instruct-Gensyn-Swarm

Everything runs on time — I never fall behind other participants, and sometimes I even wait for others to finish. I complete all stages successfully without any errors. My rewards are high, but Participation is not increasing. Why?

When I use the default configuration, Participation is always counted — but either the rewards are very low, or I don’t get any at all!

Please provide a clear answer to this — what exactly is preventing my Participation from being counted with a custom config?

My logs o

n screenshots