cricketmind-nemotron-mini

CricketMind — Cricket Domain Expert (Nemotron Mini 4B)

A fine-tuned version of nvidia/Nemotron-Mini-4B-Instruct specialized in MCC Laws of Cricket and match situation analysis.

Training

• Method: LoRA (r=16, alpha=32) on bfloat16
• Target modules: q_proj, v_proj
• Data: ~170 examples — Laws QA + response distillation from Claude
• Hardware: NVIDIA A100 80GB SXM
• Epochs: 3
• Final training loss: 1.65

Evaluation — CricketBench v0.1

LLM-as-judge evaluation (Claude) across 20 questions in 4 categories:

Usage

With Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "brettleehari/cricketmind-nemotron-mini"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)

prompt = """### Instruction:
You are CricketMind, an expert in the Laws of Cricket. Cite Law numbers and reason step by step.

### Input:
A batter is struck on the pad outside the line of off stump. They played a shot. Is it out LBW?

### Response:
"""

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.1)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Google Colab (free GPU)

1. Go to colab.google.com → New notebook
2. Runtime → Change runtime type → T4 GPU
3. Paste the code above and run

Dataset

Training data and evaluation suite: brettleehari/cricketbench-v1

Author

Hariprasad Sudharshan — AI Product Manager

• LinkedIn
• GitHub