---
library_name: transformers
license: apache-2.0
base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
pipeline_tag: text-generation
language:
- en
tags:
- text-generation-inference
- RL
- Math
- Code
- Reasoning
---

![r.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/Tx6eUgqfe4Qd3Cu4kQd0w.png)

# **Fomalhaut-QwenR-1.5B**

> **Fomalhaut-QwenR-1.5B** is a language model fine-tuned from **DeepSeek-R1-Distilled-Qwen-1.5B** using **distributed reinforcement learning (RL)**. This version enhances capabilities in **mathematical reasoning**, **coding ability**, and **error correction**, delivering efficient general-purpose reasoning and intelligent assistance in a lightweight 1.5B parameter architecture.

## **Key Improvements**
1. **Mathematical Reasoning Enhancements**:  
   Equipped with advanced capabilities in handling mathematical logic, symbolic computation, step-by-step problem-solving, and numerical accuracy across topics from basic arithmetic to higher-order mathematics.

2. **Coding and Debugging Proficiency**:  
   Improved performance in code generation, understanding documentation, and identifying and correcting bugs in multiple programming languages, especially Python, JavaScript, and C++. It supports functional, object-oriented, and scripting paradigms.

3. **Intelligent Error Correction**:  
   Capable of identifying inconsistencies or errors in logical reasoning, structured formats (JSON, XML), and code outputs, with suggestions and auto-corrections.

4. **Enhanced Instruction Following**:  
   Fine-tuned for following complex, nested instructions with increased precision and coherence over extended prompts and interactions.

5. **Long-Context Support**:  
   Supports up to **128K tokens** for input context and can generate up to **8K tokens** in one output, making it well-suited for extended problem solving, document generation, and analysis.

## **Quickstart with Transformers**

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Fomalhaut-QwenR-1.5B"

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

prompt = "Explain the difference between breadth-first search and depth-first search with Python code examples."
messages = [
    {"role": "system", "content": "You are a knowledgeable assistant skilled in reasoning, coding, and explanation."},
    {"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]
```

## **Intended Use**

1. **Mathematics and Computation**:  
   Effective for solving math problems, verifying formulas, symbolic logic, algebraic reasoning, and analytical computations.

2. **Programming Assistance**:  
   Ideal for generating, explaining, and debugging code. Suitable for both learning and software development use cases.

3. **Educational and Informational Support**:  
   Provides accurate, well-explained answers to conceptual and applied questions in STEM and humanities.

4. **Conversational AI and Reasoning Agents**:  
   Designed for intelligent chatbots capable of nuanced reasoning, error correction, and structured dialogue.

5. **Multilingual & Global Applications**:  
   Useful for translation, multilingual support bots, and cross-lingual content generation.

6. **Long-Form & Structured Content Generation**:  
   Can create long documents, reports, and structured outputs like JSON, Markdown, and tabular formats.

## **Limitations**

1. **Hardware Requirements**:  
   While lighter than 14B models, it still benefits from modern GPUs/TPUs for inference due to long-context handling.

2. **Real-Time Limitations**:  
   No real-time awareness; knowledge is limited to training data.

3. **Bias and Hallucination**:  
   While reduced, some bias and hallucinations from training data may persist.

4. **Creative Consistency**:  
   Variability in outputs for creative or ambiguous queries (e.g., fiction, storytelling).

5. **Prompt Sensitivity**:  
   Results may vary significantly depending on the structure and clarity of the input prompt.