GanitLLM-4B_CGRPO

Highlights

GanitLLM-4B_CGRPO is a Bengali mathematical reasoning model trained with Curriculum-GRPO directly on the base model (without SFT). This variant achieves the highest raw accuracy but reasons primarily in English. Key results:

+13.2 accuracy on Bn-MGSM benchmark (69.2 → 82.4)
+8.0 accuracy on Bn-MSVAMP benchmark (70.5 → 78.5)
14.94% Bengali reasoning (similar to base model)
10.5% fewer tokens in generated solutions (943 → 844 words)

Note: This model achieves high accuracy but does not reason in Bengali. For Bengali reasoning, use GanitLLM-4B_SFT_CGRPO instead.

Model Overview

Property	Value
Model Type	Causal Language Model
Base Model	Qwen/Qwen3-4B
Parameters	4B
Training	Curriculum-GRPO (no SFT)
Context Length	4,096 tokens
Language	Bengali, English

Training Details

This model was trained with a single-stage pipeline:

Curriculum-GRPO: Reinforcement learning with difficulty-aware sampling directly on the base model using GANIT-RLVR (~7.3k examples)

Reward Functions

Format Reward: Validates <think> and <answer> tag structure
Correctness Reward: +2.0 for Bengali answer match, +1.0 for English match
Bengali Reasoning Reward: Ensures >80% Bengali text in reasoning

Quickstart

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "dipta007/GanitLLM-4B_CGRPO"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

problem = "একটি দোকানে ১২টি আপেল আছে। যদি ৫টি আপেল বিক্রি হয়, তাহলে কতটি আপেল বাকি থাকবে?"

prompt = f"""A conversation takes place between the user and the assistant. The user asks a question, and the assistant solves the problem. Please reason step by step in Bengali, and put your final answer in the <answer> </answer> tags.

Question: {problem}"""

messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(**model_inputs, max_new_tokens=2048, temperature=0.7)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
response = tokenizer.decode(output_ids, skip_special_tokens=True)
print(response)

Using vLLM

vllm serve dipta007/GanitLLM-4B_CGRPO --max-model-len 4096

Performance

Model	Bn-MGSM	Bn-MSVAMP	Avg. Words	Bengali %
Qwen3-4B (base)	69.20	70.50	943	14.79%
GanitLLM-4B_CGRPO	82.40	78.50	844	14.94%

Related Models

Model	Parameters	Training	Link
GanitLLM-4B_SFT_CGRPO	4B	SFT + CGRPO	Link
GanitLLM-4B_SFT_GRPO	4B	SFT + GRPO	Link
GanitLLM-4B_CGRPO	4B	CGRPO	Link
GanitLLM-1.7B_CGRPO	1.7B	CGRPO	Link
GanitLLM-0.6B_CGRPO	0.6B	CGRPO	Link