Run DeepScaleR 1.5B on macOS : Step by Step Installation Guide

DeepScaleR 1.5B represents a fine-tuned iteration of the Deepseek-R1-Distilled-Qwen-1.5B model, engineered to advance accessibility in Reinforcement Learning (RL) for Large Language Models (LLMs).

This model exhibits cross-platform compatibility, supporting macOS, Linux, and Windows, thereby facilitating a broad adoption among researchers and developers.

Key Features of DeepScaleR 1.5B

• Hybrid Architecture: MoE (Mixture of Experts) SSM-Transformer model optimized for speed.
• 256k Context Window: Industry-leading capacity for long-context RAG and complex workflows.
• Multilingual Mastery: Supports 9 languages including English, Arabic, and European languages.
• Tool Integration: Native JSON support and API compatibility for AI app development.

System Requirements

DeepScaleR 1.5B: Architectural and Functional Overview

• Fine-Tuned Foundation: Built upon Deepseek-R1-Distilled-Qwen-1.5B.
• Objective: Enhances the accessibility and efficiency of reinforcement learning frameworks for LLMs.
• Cross-Platform Compatibility: Operational on macOS, Linux, and Windows.
• Development Entity: AI21 Labs.
• Structural Composition: Employs a Mixture of Experts (MoE) hybrid architecture, integrating State Space Models (SSMs) with Transformer paradigms.
• Contextual Capacity: Accommodates input sequences of up to 256k tokens.
• Optimization Strategies: Supports long-context Retrieval-Augmented Generation (RAG), advanced tool utilization, and structured JSON modeling.
• Multilingual Proficiency: Extends support to English, Spanish, French, Portuguese, Italian, Dutch, German, Arabic, and Hebrew.

Installation and Execution Procedures

For an efficient setup and execution of DeepScaleR 1.5B on macOS, adhere to the following procedural framework:

1. Referencing the Installation Guide:
   • A meticulously structured tutorial elucidates the stepwise process for local deployment of DeepScaleR-1.5B-Preview.
   • The instructional content is accessible via YouTube, delivering an in-depth walkthrough.
2. Executing DeepScaleR via Ollama:
   • This directive instantiates and operationalizes the model within the local computational infrastructure.

Given a pre-installed Ollama environment, execution is initiated with the following command:

ollama run deepscaler

How to install DeepScaleR 1.5B on MacOS

Method 1: Ollama One-Click Installation

1. Install Ollama
2. Wait for automatic model download (~5GB)

Open Terminal and run:

ollama run deepscaler  

Method 2: Manual Local Installation

Install dependencies:

pip install transformers vllm torch  

Clone the model repository:

git clone https://github.com/deepscaler/DeepScaleR-1.5B-Preview  

Install Python 3.10+ via Homebrew:

brew install [email protected]  

Optimization Tips for Mac Users

Context Window Tuning: Adjust chunk size for RAG applications:

tokenizer.model_max_length = 262144  # 256k tokens  

Memory Management: Use 4-bit quantization for M1/M2 Macs:

from transformers import BitsAndBytesConfig  
bnb_config = BitsAndBytesConfig(load_in_4bit=True)  

Metal Performance: Enable GPU acceleration with:

model.to('mps')  # PyTorch Metal backend  

Multilingual Support Guide

Practical Implementation Scenarios

Example 1: Summarization of Extensive Texts

from transformers import pipeline

summarizer = pipeline("summarization", model="deepscaler")
text = "DeepScaleR significantly enhances NLP capabilities, particularly for long-context comprehension and reinforcement learning applications."
summary = summarizer(text, max_length=50, min_length=25, do_sample=False)
print(summary)

Example 2: Sentiment Classification of Textual Data

from transformers import pipeline

sentiment_analyzer = pipeline("sentiment-analysis", model="deepscaler")
text = "DeepScaleR demonstrates remarkable efficacy in large-scale language modeling."
result = sentiment_analyzer(text)
print(result)

Example 3: Context-Aware Question Answering

from transformers import pipeline

qa_pipeline = pipeline("question-answering", model="deepscaler")
context = "DeepScaleR has been engineered for advanced long-context processing and reinforcement learning integrations."
question = "What are the primary optimizations of DeepScaleR?"
answer = qa_pipeline(question=question, context=context)
print(answer)

Advanced Capabilities and Ecosystem Integrations

• Transformer and VLLM Compatibility: Natively supported within the Transformers and VLLM frameworks.
• Optimized Long-Context Processing: Architected for high-efficiency handling of extended text sequences.
• Interoperability with Toolchains and APIs: Seamlessly integrates with systems requiring structured JSON-based outputs

Conclusion

By adhering to these procedural guidelines, users can efficiently deploy and leverage DeepScaleR 1.5B on macOS, harnessing its cutting-edge reinforcement learning and large-scale language modeling capabilities for advanced computational tasks.

References

1. Run DeepSeek Janus-Pro 7B on Mac: A Comprehensive Guide Using ComfyUI
2. Run DeepSeek Janus-Pro 7B on Mac: Step-by-Step Guide
3. Run DeepSeek Janus-Pro 7B on Windows: A Complete Installation Guide