DeepSeek-R1-Distill-Qwen-14B-unsloth-bnb-4bit-AIMO_CoT

Disclaimer!!

Hello! This model is not perfect yet, I am just experimenting! This is me attempting the AIMO Prize 2 Kaggle contest

I have decided to release the models before the competition ends because I don't care about winning the contest as much!

My research fields are Medical Computing and Reinforcement Learning. Feel free to add me on LinkedIn if you want to chat!

• Developed by: sindhusatish97
• License: apache-2.0
• Finetuned from model : unsloth/DeepSeek-R1-Distill-Qwen-14B

This qwen2 model was trained 2x faster with Unsloth and Huggingface's TRL library. - Huge thanks to the awesome team for releasing these distilled models!

<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>

Test it out!

!pip install unsloth
# Also get the latest nightly Unsloth!
!pip install --force-reinstall --no-cache-dir --no-deps git+https://github.com/unslothai/unsloth.git
from unsloth import FastLanguageModel
import torch
max_seq_length = 5120 # I chose this value based on Qwen's max sequence length.  
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "sindhusatish97/DeepSeek-R1-Distill-Qwen-14B-unsloth-bnb-4bit-AIMO_CoT",
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    # token = "hf_...", # use one if using gated models like meta-llama/Llama-2-7b-hf
)
FastLanguageModel.for_inference(model)
inputs = tokenizer(
[
        """4 pints of a 5% antifreeze solution and 8 pints of a 20% antifreeze solution must be mixed to obtain 12 pints of a 
        solution with what percentage of antifreeze?"""
], return_tensors = "pt").to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)
_ = model.generate(**inputs, streamer = text_streamer, max_length = 2048)