README.md

---
license: apache-2.0
language:
- en
base_model:
- NovaSky-AI/Sky-T1-32B-Preview
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
- Omni
---
![omni.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/Zz_Uc6M06tyh3Euhm93fn.png)

# **Omni-Reasoner-o1: Overview**

*Omni-Reasoner-o1* is a specialized AI model built upon the Sky T1 32B architecture, combined with **Qwen 2.5 32B**, and fine-tuned using synthetic data from OpenAI pipeline-generated records. It is optimized for mathematical reasoning and complex problem-solving.

# **Quickstart with Transformers**

Here provides a code snippet with `apply_chat_template` to show you how to load the tokenizer and model and how to generate contents.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Omni-Reasoner-o1"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "How many r in strawberry."
messages = [
    {"role": "system", "content": "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
```

# **Key Features**
1. **Hybrid Architecture:**
   - Combines **Sky T1 32B** and **Qwen 2.5 32B** to leverage strengths in both natural language understanding and mathematical reasoning.
   - Enables robust problem-solving across diverse domains.

2. **Mathematical Expertise:**
   - Trained specifically as a **mathematical reasoner and problem solver**.
   - Excels in numerical computations, symbolic mathematics, proofs, and equation-solving.

3. **Synthetic Data Fine-Tuning:**
   - Leveraged high-quality synthetic data generated by OpenAI pipelines.
   - Ensures enhanced generalization across a wide range of problem-solving scenarios.

4. **Natural Language Processing (NLP):**
   - Capable of understanding and interpreting complex language inputs related to mathematical queries.
   - Provides step-by-step explanations for solutions, fostering user understanding.

5. **Multi-Task Capability:**
   - Handles a variety of mathematical tasks including algebra, calculus, combinatorics, and statistics.
   - Suitable for word problems and domain-specific queries requiring logic and reasoning.

6. **Scalability:**
   - Designed for seamless integration into **educational platforms**, **scientific research tools**, and **automated reasoning systems**.

# **Intended Use**
1. **Educational Applications:**
   - Acts as a tutor for students in mathematics and related fields.
   - Provides explanations, step-by-step solutions, and practice problem generation.

2. **Scientific Research:**
   - Aids researchers in automating repetitive mathematical calculations or exploring new problem-solving methodologies.

3. **Professional Use Cases:**
   - Supports professionals in domains like engineering, data science, and finance by solving domain-specific mathematical problems.

4. **AI-Assisted Development:**
   - Assists in coding environments for algorithm development and debugging by identifying mathematical bottlenecks or issues.

5. **Automated Systems:**
   - Integrates into automated reasoning and decision-making systems for operations requiring quantitative analysis.

# **Limitations**
1. **Reliance on Synthetic Data:**
   - Despite its extensive training, reliance on synthetic data might lead to **biases** or **overfitting** in specific scenarios.
   - May struggle with real-world edge cases not reflected in its training data.

2. **Domain-Specific Gaps:**
   - While excelling in mathematics, it may not perform as well in non-mathematical or interdisciplinary problem-solving tasks.

3. **Resource Intensive:**
   - Due to its hybrid 32B architecture, deploying the model requires **significant computational resources**.

4. **Interpretation Errors:**
   - Misinterprets poorly structured or ambiguous natural language queries.
   - May provide overly verbose explanations that aren't always user-friendly.

5. **Limitations in Creativity:**
   - Not designed for creative or abstract tasks outside mathematical reasoning, such as writing, art, or subjective decision-making.

6. **Dependency on Prompt Quality:**
   - Performance can degrade with unclear, poorly framed, or overly complex prompts