110 lines
3.8 KiB
Markdown
110 lines
3.8 KiB
Markdown
---
|
|
license: apache-2.0
|
|
base_model:
|
|
- Qwen/Qwen3-1.7B
|
|
datasets:
|
|
- prithivMLmods/Demeter-LongCoT-400K
|
|
language:
|
|
- en
|
|
pipeline_tag: text-generation
|
|
library_name: transformers
|
|
tags:
|
|
- text-generation-inference
|
|
- LongCoT
|
|
- trl
|
|
- math
|
|
- code
|
|
- stem
|
|
---
|
|
|
|
|
|
|
|
# **Demeter-LongCoT-Qwen3-1.7B**
|
|
|
|
> **Demeter-LongCoT-Qwen3-1.7B** is a reasoning-focused model fine-tuned on **Qwen/Qwen3-1.7B** using the **Demeter-LongCoT-400K** dataset.
|
|
> It is designed for **math and code chain-of-thought reasoning**, blending symbolic precision, scientific logic, and structured output fluency—making it an effective tool for developers, educators, and researchers seeking reliable step-by-step reasoning.
|
|
|
|
> \[!note]
|
|
> GGUF: [https://huggingface.co/prithivMLmods/Demeter-LongCoT-Qwen3-1.7B-GGUF](https://huggingface.co/prithivMLmods/Demeter-LongCoT-Qwen3-1.7B-GGUF)
|
|
|
|
---
|
|
|
|
## **Key Features**
|
|
|
|
1. **Unified Reasoning in Math & Code**
|
|
Fine-tuned on **Demeter-LongCoT-400K**, which emphasizes extended chain-of-thought reasoning in mathematics, algorithms, and programming workflows.
|
|
|
|
2. **Advanced Code Understanding & Generation**
|
|
Handles multi-language programming tasks with explanations, optimization hints, and error detection—suited for algorithm synthesis, debugging, and prototyping.
|
|
|
|
3. **Mathematical Problem Solving**
|
|
Excels at step-by-step derivations, symbolic manipulations, and applied problem solving across calculus, algebra, and logic-based reasoning.
|
|
|
|
4. **Chain-of-Thought Focused Reasoning**
|
|
Optimized to produce clear, structured thought processes for both **STEM explanations** and **computational logic** tasks.
|
|
|
|
5. **Structured Output Mastery**
|
|
Generates well-formed outputs in **LaTeX**, **Markdown**, **JSON**, **CSV**, and **YAML**, enabling smooth integration with research pipelines and technical documentation.
|
|
|
|
6. **Balanced Performance for Deployment**
|
|
Designed to deliver strong reasoning under moderate compute budgets, deployable on **mid-range GPUs**, **offline clusters**, and **specialized edge AI systems**.
|
|
|
|
---
|
|
|
|
## **Quickstart with Transformers**
|
|
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "prithivMLmods/Demeter-LongCoT-Qwen3-1.7B"
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
model_name,
|
|
torch_dtype="auto",
|
|
device_map="auto"
|
|
)
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
prompt = "Solve the integral of x^2 * e^x step by step."
|
|
|
|
messages = [
|
|
{"role": "system", "content": "You are a tutor skilled in math, code, and step-by-step reasoning."},
|
|
{"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]
|
|
print(response)
|
|
```
|
|
|
|
---
|
|
|
|
## **Intended Use**
|
|
|
|
* Step-by-step math tutoring and symbolic derivation
|
|
* Advanced coding assistant for algorithms, debugging, and structured reasoning
|
|
* Chain-of-thought generation for research and education tools
|
|
* Producing structured outputs for technical documentation and computational pipelines
|
|
* Deployments requiring reliable reasoning under constrained compute
|
|
|
|
## **Limitations**
|
|
|
|
* Not tuned for general-purpose or conversational tasks
|
|
* May underperform in long-form multi-document contexts
|
|
* Specialized in math and code—general writing or casual dialogue may be weak
|
|
* Prioritizes structured reasoning over natural or emotional tone generation |