---
datasets:
- prithivMLmods/Poseidon-Reasoning-5M
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen3-1.7B
library_name: transformers
tags:
- text-generation-inference
- moe
- code
- science
- biology
- chemistry
- thinking
pipeline_tag: text-generation
---

![1.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/vXEwxMVMiov1zhOFUt6AJ.png)

# **Poseidon-Reasoning-1.7B**

> **Poseidon-Reasoning-1.7B** is a general-purpose, high-efficiency reasoning model fine-tuned on **Qwen3-1.7B** using the **Poseidon-Reasoning-5M** dataset (first 70K entries). Designed for **mathematical, scientific, and code-related reasoning**, this model strikes a balance between structured logic and contextual fluency—ideal for domains demanding symbolic precision and algorithmic thought.

> \[!note]
> GGUF: [https://huggingface.co/prithivMLmods/Poseidon-Reasoning-1.7B-GGUF](https://huggingface.co/prithivMLmods/Poseidon-Reasoning-1.7B-GGUF)

## **Key Features**

1. **Versatile Reasoning Model**
   Finely tuned for multi-domain reasoning tasks, including mathematics, scientific computation, and code logic—capable of navigating structured problem-solving and analytic workflows.

2. **Qwen3-1.7B Foundation**
   Built upon **Qwen3-1.7B**, providing multilingual reasoning capability, efficient token handling, and strong alignment with instruction-following tasks.

3. **Powered by Poseidon-Reasoning-5M (70K Sample Subset)**
   Trained on a carefully selected subset of 70K entries from the **Poseidon-Reasoning-5M** dataset—focusing on tasks that emphasize **symbolic accuracy**, **step-by-step thinking**, and **STEM-relevant clarity**.

4. **Balanced Thinking Mode**
   Supports structured, guided thinking without excessive hallucination or unnecessary verbosity. Ideal for prompt-driven logic tasks with moderate complexity.

5. **Rich Format Output**
   Outputs include **Markdown**, **Python**, **LaTeX**, and tabular structures—helpful for notebooks, scientific documentation, and programmatic outputs.

6. **1.7B Parameter Footprint**
   Lightweight enough to run on **mid-tier GPUs or CPU-only environments**, while offering scalable reasoning power for research, teaching, and light automation.

## **Quickstart with Transformers**

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Poseidon-Reasoning-1.7B"

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

prompt = "Solve: What is the derivative of sin(x) * ln(x)?"

messages = [
    {"role": "system", "content": "You are a structured reasoning assistant for math, science, and code."},
    {"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=256
)
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]
print(response)
```

## **Intended Use**

* General-purpose symbolic reasoning
* Math and science tutoring, theorem solving, and computational guidance
* Structured coding under constraints or STEM-based tasks
* Lightweight environments where interpretability and precision matter
* Prompt-driven reasoning with deterministic steps

## **Limitations**

* Not designed for broad open-domain conversation
* May underperform on creative writing or emotional expression
* Best results occur with **clear problem statements and goal-directed prompts**
* Less suitable for speculative or abstract reasoning without structure