Thinking-Camel-7b

Model Description

Thinking-Camel-7b is a 7 billion parameter large language model fine-tuned from ALLaM-7B-Instruct-preview. This model is designed to enhance reasoning capabilities while maintaining the core strengths of the ALLAM architecture.

Model Details

Developed by: Mohaddz
Model type: Causal Language Model
Language(s): English (primary), with potential capabilities in other languages supported by the base model
License: [Same as base model - apache-2.0]
Base model: ALLAM-7b-Instruct
Training paradigm: GRPO

Intended Uses

Thinking-Camel-7b is intended for a variety of applications requiring strong reasoning capabilities, including but not limited to:

Complex problem-solving
Step-by-step reasoning for mathematical and logical problems
Enhanced chain-of-thought processing
Research assistance and information synthesis
Educational applications requiring explanatory capabilities

How to Use

You can use Thinking-Camel-7b with the Hugging Face Transformers library:

from transformers import AutoTokenizer, AutoModelForCausalLM

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("Mohaddz/Thinking-Camel-7b")
model = AutoModelForCausalLM.from_pretrained("Mohaddz/Thinking-Camel-7b")

# Generate text
inputs = tokenizer("Question: What would happen if we doubled the Earth's gravity? Think through this step by step.", return_tensors="pt")
outputs = model.generate(**inputs, max_length=512)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Using with vLLM for faster inference:

from vllm import LLM, SamplingParams

# Initialize the model
llm = LLM(model="Mohaddz/Thinking-Camel-7b")

# Set sampling parameters
sampling_params = SamplingParams(temperature=0.7, top_p=0.9, max_tokens=512)

# Generate completions
prompts = ["Question: How would you solve the Tower of Hanoi problem with 3 disks? Think step by step."]
outputs = llm.generate(prompts, sampling_params)

# Print the generated text
for output in outputs:
    print(output.outputs[0].text)

Prompt Format

Thinking-Camel-7b works best with prompts that explicitly ask the model to think step by step:

Question: [Your complex problem or question here]
Think through this step by step.

For general use, you can also use a standard instruction format:

[Instruction or question]

Limitations

As with all LLMs, Thinking-Camel-7b may occasionally generate factually incorrect information
The model inherits limitations from its base model, ALLAM-7b-Instruct
Benchmarks are not yet available to quantify performance improvements
Like most models in this size range, it may struggle with highly specialized domain knowledge
Performance on complex reasoning tasks may vary

Training

Thinking-Camel-7b was fine-tuned from ALLaM-7B-Instruct-preview with a focus on enhancing reasoning capabilities. The training approach prioritized:

Chain-of-thought examples
Step-by-step problem solving
Improved logical reasoning structures
Explicit thinking processes

Ethical Considerations

Users should be aware of common LLM limitations including potential biases inherited from the training data, hallucinations, and the need for human oversight particularly in sensitive applications. This model should not be used as the sole decision-maker for critical applications.

Future Work

Comprehensive benchmarking across standard LLM evaluation suites
Further fine-tuning on specialized reasoning tasks
Potential instruction-tuning with human feedback

Citation

If you use this model in your research, please cite:

@misc{mohaddz2025thinkingcamel,
  author = {Mohaddz},
  title = {Thinking-Camel-7b},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Mohaddz/Thinking-Camel-7b}}
}

Contact

For questions, feedback, or issues related to Thinking-Camel-7b, please contact Mohaddz through Hugging Face or open an issue in the model repository.

Mohaddz
/

Thinking-Camel-7b