MathMate
Collection
Capstone Project LLM models
•
5 items
•
Updated
Mathmate-7B-dare-ties is a Mixture of Experts (MoE) made with the following models using LazyMergekit:
base_model: AI-MO/NuminaMath-7B-TIR
gate_mode: hidden
dtype: bfloat16
experts:
- source_model: AI-MO/NuminaMath-7B-TIR
positive_prompts:
- "This model is good at solving math questions at high school level and generating python code for the same"
# - source_model: Qwen/Qwen2-Math-7B-Instruct
# positive_prompts:
# - "This model is really good at solving college level math to olympiad level questions"
- source_model: deepseek-ai/DeepSeek-Prover-V1.5-RL
positive_prompts:
- "This model is good at formal theorem providing math problems"
!pip install -qU transformers bitsandbytes accelerate
from transformers import AutoTokenizer
import transformers
import torch
model = "Haleshot/Mathmate-7B-dare-ties"
tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
"text-generation",
model=model,
model_kwargs={"torch_dtype": torch.float16, "load_in_4bit": True},
)
messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
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"])