EnversonAI/DeepSeek-R1-FineTuned-AdaptiveQGen-COT

DeepSeek-R1-FineTuned-AdaptiveQGen

Model Description

DeepSeek-R1-FineTuned-AdaptiveQGen is a fine-tuned version of the DeepSeek R1 (Reasoning Model) specifically designed for adaptive question generation in English language learning. The model has been fine-tuned on a custom Chain-of-Thought (CoT) dataset to generate context-aware, adaptive follow-up questions tailored to a student's proficiency level, learning goals, interests, and common errors.

Intended Use

This model is intended for use in educational settings, particularly for English language learning. It can be integrated into AI-driven tutoring systems to provide personalized and adaptive learning experiences. The model is designed to:

1. Generate follow-up questions based on student responses.

2. Provide gentle corrective feedback for grammatical errors.

3. Align questions with the student's interests and learning goals.

Training Data

The model was fine-tuned on a custom dataset consisting of interactive exchanges between an AI tutor and a student. Each data row includes:

• An initial question from the AI tutor.
• A student's response.
• A detailed Chain-of-Thought (CoT) explanation for generating an adaptive follow-up question.
• The AI's response, including feedback and the follow-up question.

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import unsloth

model_name = "EnversonAI/DeepSeek-R1-FineTuned-AdaptiveQGen-COT"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

device = "cuda"

model.to(device)

prompt_style = """Below is an instruction that describes a task, paired with an input that provides further context. 
Write a response that appropriately completes the request. 
Before answering, think carefully about the student's message and create a step-by-step chain of thoughts to ensure pedagogically effective feedback.

### Instruction:
You are an English language instructor with expertise in ESL education and communicative language teaching. 
Please respond to the following student message with appropriate feedback and generate context-aware questions.

### Question:
{}

### Response:
<think>
{}"""

input_text = "Do you like swimming? Student: but i dont know no i am mostly like hiking"
inputs = tokenizer(input_text, return_tensors="pt")

# Move the input tensors to the same device as the model
inputs = tokenizer([prompt_style.format(input_text, "")], return_tensors="pt").to("cuda")


outputs = model.generate(
    input_ids=inputs.input_ids,
    attention_mask=inputs.attention_mask, max_length=2056, num_return_sequences=1, no_repeat_ngram_size=2, top_p=0.95, top_k=50
)
response = tokenizer.batch_decode(outputs)
print(response[0].split("### Response:")[1])

<think>
The student uses an incorrect verb tense ('like') and incorrect word order ('but i don't know'). To maintain engagement, i will ask a follow-up question about hiking. I will provide feedback on verb usage and word structure to help the learner correct their mistake.
</think>
That's interesting! Hiking sounds like a fun activity. What do you think about exploring more trails?<｜end▁of▁sentence｜>

More examples:

input_text = "Do you like tea? Student: Yes i am mostly like cofee"

<think>
The student has provided a yes/no response, signaling a beginner-level proficiency. I identify that the topic is food preferences, which can be used to engage the learner and provide further practice. The student mentioned 'cof ee', indicating a possible error in vocabulary. To encourage correct usage of vocabulary, i will provide feedback in a supportive way and ask a follow-up question that reinforces correct grammar.
</think>
Great choice! You meant to say 'i mostly prefer coffee.' do you have any favorite brands of coffee?<｜end▁of▁sentence｜>


input_text = "How do you think technology will change the way we work in the next 10 years? Student: I think I was be hard to find job in future"

<think>
The student has provided a sentence with incorrect verb tense ('be' instead of 'will be').  I identify that the topic of future jobs aligns with the learning goal of discussing technology's impact on society. To reinforce correct verb usage and encourage further discussion on the subject, i will provide feedback in a supportive way and ask a follow-up question that requires the use of the correct tense. By asking about potential job markets in 2030, the question engages the learner in critical thinking and encourages them to consider future possibilities.
</think>
That's an interesting point! You meant to say 'i think it will be difficult tofind jobs inthe future.' what do you think are some industries that might grow rapidly in 10 years?<｜end▁of▁sentence｜>

About Enverson

Our mission is to make personalized language learning accessible, engaging, and effective for learners worldwide. We believe that every individual has unique learning needs, preferences, and goals, and our technology is designed to adapt to these differences, creating a truly personalized educational experience.

To achieve our mission, Enverson is actively developing and fine-tuning multiple in-house Large Language Models (LLMs). These models are specifically designed to address the complexities of language learning and provide adaptive, context-aware interactions.