Mathmate-7B-DELLA

Mathmate-7B-DELLA is a merge of the following models using LazyMergekit:

• AI-MO/NuminaMath-7B-TIR
• deepseek-ai/DeepSeek-Prover-V1.5-RL

🧩 Configuration

models:
  - model: AI-MO/NuminaMath-7B-TIR
    parameters:
      density: 0.5
      weight: 0.3
  - model: deepseek-ai/DeepSeek-Prover-V1.5-RL
    parameters:
      density: 0.5
      weight: 0.2
merge_method: della
base_model: deepseek-ai/deepseek-math-7b-base
parameters:
  normalize: true
dtype: bfloat16

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Haleshot/Mathmate-7B-DELLA"
messages = [{"role": "user", "content": "What is a large language model?"}]

tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])

📊 Evaluation Results

Evaluation results using LLMAutoeval:

AGIEval

Average: 21.95%

GPT4All

Average: 36.5%

TruthfulQA

Average: 48.08%

Bigbench

Average: 28.89%

Average score: 33.86%

Elapsed time: 03:52:09