Parameter	Value
Algorithm	ReMax
Advantage Baseline	Greedy-decoded response reward
Extra Inference	1 deterministic rollout per prompt
Group Structure	Not required
Policy Loss	`ppo`
KL Coefficient	`0.0`
Epsilon (clip)	`0.2`
Discount Factor (`gamma`)	`1.0`
Divide Advantage by Std	`False`
Filter Zero Advantage Groups	`False`
Rollouts per Problem	`16`

ReMax uses a greedy-decoded response's reward as the baseline for advantages.

Training Hyperparameters

Parameter	Value
Base Model	`Qwen/Qwen2.5-3B`
Learning Rate	`1e-06`
LR Scheduler	`cosine`
Warmup Steps	`25`
Max Training Steps	`1500`
Micro Batch Size	`2`
Gradient Accumulation	`128`
Effective Batch Size	`256`
Sequence Length	`8192`
Gradient Clipping	`0.3`
Weight Decay	`0.01`
Optimizer	`adamw_torch`
Precision	`bf16`
DeepSpeed	ZeRO Stage 3

Evaluation Results

Pass@k on math reasoning benchmarks (N=32 samples per problem, temperature=1.0):

Dataset	pass@1	pass@2	pass@4	pass@8	pass@16	pass@32
GSM8K (test)	85.99	90.50	93.34	95.29	96.64	97.50
MATH-500	67.36	74.99	81.23	85.92	89.09	91.20
Overall	80.87	86.24	90.01	92.71	94.56	95.77

GSM8K test: 1319 problems · MATH-500: 500 problems · Overall: 1819 problems (overall weighted by problem count).

Training Curves

W&B Run

Full training logs: https://wandb.ai/jaygala24-team/rl-post-training/runs/qwen2.5_3b_remax_3a1f_4xh100_214753_finetune_1c7d72aa

Usage

Transformers

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("jaygala24/Qwen2.5-3B-ReMax-math-reasoning", revision="step-0200")  # optional branch, e.g. "step-0400"
tokenizer = AutoTokenizer.from_pretrained("jaygala24/Qwen2.5-3B-ReMax-math-reasoning", revision="step-0200")

prompt = "Please reason step by step, and put your final answer within \\boxed{}.\n\nWhat is the sum of 123 and 456?"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=4096, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

vLLM

python

from vllm import LLM, SamplingParams

llm = LLM(model="jaygala24/Qwen2.5-3B-ReMax-math-reasoning", revision="step-0200")  # optional branch, e.g. "step-0400"
sampling_params = SamplingParams(temperature=0.7, max_tokens=4096)

prompt = "Please reason step by step, and put your final answer within \boxed{}.

What is the sum of 123 and 456?"
outputs = llm.generate([prompt], sampling_params)
print(outputs[0].outputs[0].text)

jaygala24/Qwen2.5-3B-ReMax-math-reasoning

Qwen2.5-3B-ReMax-math-reasoning

Training Details

Datasets

RL Algorithm