Instalar DeepSeek R1 1.5B en Ubuntu 24.04 sin GPU

Instalar DeepSeek en tu sistema sin GPU, pasos:

Especificaciones del Entorno de Pruebas

Componente	Detalle
SO	Ubuntu Cinnamon 24.04 LTS x86_64
Kernel	6.8.0-51-generic
CPU	Intel i7-6820HQ (8 núcleos) @ 3.600GHz
GPUs	AMD ATI Radeon HD 8830M / R7 250 / R7 M465X Intel HD Graphics 530
RAM	15.882 GB (3.716 GB en uso)
Resolución	1440x810
Escritorio	Cinnamon 6.0.4

1. Instalar Git LFS

sudo apt-get install git-lfs
git lfs install

2. Clonar el repositorio

cd /opt
sudo mkdir deepseek && sudo chown $USER:$USER deepseek
cd deepseek
git clone https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B

3. Crear y activar un entorno virtual

python -m venv deepseek-env
source deepseek-env/bin/activate

4. Instalar dependencias optimizadas para CPU

pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install transformers==4.41.0 accelerate sentencepiece

5. Crear el archivo deepseek_local.py

6. Para ejecutar

python deepseek_local.py

Observación de Rendimiento

Tiempo de Procesamiento: En mi computadora (Intel i7-6820HQ), cada respuesta toma un promedio de 150 segundos (2.5 minutos).

Parámetros de Optimización

Parámetro	Valor Original	Rango Recomendado	Impacto
max_new_tokens	256	64 - 512	Reduce tiempo de generación. Menos tokens = respuestas más cortas
temperature	0.7	0.1 - 1.0	Controla aleatoriedad. Menor valor = respuestas más determinísticas
top_p	0.9	0.5 - 1.0	Probabilidad de tokens. Menor valor = más predecible
top_k	50	10 - 100	Filtra los k tokens más probables. Reduce complejidad computacional
repetition_penalty	1.1	1.0 - 2.0	Evita repeticiones. Mayor valor = menos repetición

Consejos:

• Reducir max_new_tokens disminuye significativamente el tiempo de procesamiento
• Valores más bajos de temperature generan respuestas más consistentes
• Experimentar con estos parámetros para encontrar el equilibrio ideal

Ejemplo de Modificación

generation_config = {
    "max_new_tokens": 128,  # Reducido de 256
    "temperature": 0.5,     # Más determinístico
    "top_p": 0.8,           # Menos tokens considerados
    "top_k": 30,            # Menos opciones
    "repetition_penalty": 1.2
}

Ejemplos de algunas preguntas

Explicaciones Técnicas

python deepseek_local.py "Explain quantum computing and its potential to revolutionize cryptography"
python deepseek_local.py "Describe the fundamental principles of blockchain technology"
python deepseek_local.py "How do microprocessors process binary information?"
python deepseek_local.py "Explain the concept of edge computing in distributed systems"
python deepseek_local.py "What are the core principles of quantum entanglement?"
    

Ayuda para Código y Depuración

python deepseek_local.py "Implement a memory-efficient algorithm for finding the median in a large dataset"
python deepseek_local.py "Design a robust error handling mechanism for distributed microservices"
python deepseek_local.py "Create a recursive solution for traversing a complex tree structure"
python deepseek_local.py "Optimize a Python script with high time complexity"
python deepseek_local.py "Develop a secure authentication system using modern cryptographic techniques"
    

Escritura Creativa

python deepseek_local.py "Write a dystopian short story about AI-controlled social credit systems"
python deepseek_local.py "Compose a narrative exploring consciousness in artificial intelligence"
python deepseek_local.py "Create a poetic metaphor comparing technological evolution to natural adaptation"
python deepseek_local.py "Write a speculative fiction piece about interstellar communication"
python deepseek_local.py "Develop a character study of a sentient AI struggling with ethical dilemmas"
    

Guías Prácticas

python deepseek_local.py "Comprehensive guide to implementing microservices architecture"
python deepseek_local.py "Step-by-step process for conducting a thorough code security audit"
python deepseek_local.py "Detailed workflow for machine learning model deployment"
python deepseek_local.py "Best practices for designing scalable cloud infrastructure"
python deepseek_local.py "Complete tutorial for setting up a continuous integration pipeline"
    

Análisis Conceptual

python deepseek_local.py "Deep analysis of privacy implications in big data technologies"
python deepseek_local.py "Comprehensive comparison of different AI alignment approaches"
python deepseek_local.py "Philosophical implications of artificial general intelligence"
python deepseek_local.py "Systemic risks of autonomous weapon technologies"
python deepseek_local.py "Interdisciplinary perspectives on technological singularity"
    

Lenguaje y Traducción

python deepseek_local.py "Analyze the linguistic evolution of programming language syntax"
python deepseek_local.py "Explore the etymology of technical terminology in computer science"
python deepseek_local.py "Compare communication patterns in different programming paradigms"
python deepseek_local.py "Linguistic analysis of code comments across multiple programming cultures"
python deepseek_local.py "Investigate the semantic nuances in technical translation"
    

Escenarios Hipotéticos

python deepseek_local.py "Simulate global economic impact of complete automation"
python deepseek_local.py "Predict societal transformation with breakthrough neurotechnology"
python deepseek_local.py "Model potential outcomes of universal basic income implementation"
python deepseek_local.py "Explore civilization scenarios with unlimited renewable energy"
python deepseek_local.py "Analyze potential global changes with radical life extension"