Small LLM Comparison in 2024

Timeline of major LLM developments from 2019 to 2024, here are the top 3 Large Language Models (LLMs) in various domains, their key specifications, and performance suitability:

---

Domain	Model Name	Developer
General Q&A (Text Generation)	Mistral 7B	Mistral AI (2023)
	LLaMA 2 (7B)	Meta (2023)
	Phi-3-Mini	Microsoft (2024)
Coding	Phi-3-Small	Microsoft (2024)
	DeciCoder-1B	Deci (2023)
	GPT-2 Small	OpenAI (2020)
Math	Dolly-v2-3B	Databricks (2023)
	Phi-1.5	Microsoft (2024)
	LLaMA 2 (7B)	Meta (2023)
Image Understanding	Dolly-v2-3B	Databricks (2023)
	LLaMA 2 (7B)	Meta (2023)
	Phi-3-Mini	Microsoft (2024)
Translation	Mistral 7B	Mistral AI (2023)
	Phi-3-Small	Microsoft (2024)
	LLaMA 2 (7B)	Meta (2023)

---

1. General Q&A (Text Generation)

Mistral 7B (2023 by Mistral AI)

1. Technical Specs and Hardware Requirements:
   • Parameters: 7 billion
   • RAM: 4GB (smartphones), 8GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 20GB
2. Benchmark: Outperforms Llama 2 13B on all benchmarks; limited context window of 8,000 tokens.

---

Phi-3-Mini (2024 by Microsoft)

1. Technical Specs and Hardware Requirements:
   • Parameters: 3 billion
   • RAM: 4GB (smartphones), 6GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 10GB
2. Benchmark: Fast response times, limited factual recall.

---

LLaMA 2 (7B) (2023 by Meta)

1. Technical Specs and Hardware Requirements:
   • Parameters: 7 billion
   • RAM: 6GB (smartphones)
   • CPU: Snapdragon 855 or higher
   • Storage: 8GB
2. Benchmark: Slower on-device performance compared to Mistral 7B, efficient for basic tasks.

---

2. Coding

DeciCoder-1B (2023 by Deci)

1. Technical Specs and Hardware Requirements:
   • Parameters: 1 billion
   • RAM: 2GB (smartphones), 4GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 2GB
2. Benchmark: Great for basic coding tasks, runs well on low resources.

---

Phi-3-Small (2024 by Microsoft)

1. Technical Specs and Hardware Requirements:
   • Parameters: 7 billion
   • RAM: 4GB (smartphones), 8GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 20GB
2. Benchmark: Fast and efficient in code understanding.

---

GPT-2 Small (2020 by OpenAI)

1. Technical Specs and Hardware Requirements:
   • Parameters: 117 million
   • RAM: 2GB (smartphones), 4GB (PCs)
   • CPU: Snapdragon 660 or higher
   • Storage: 1GB
2. Benchmark: Basic coding capabilities, low-resource requirements.

---

3. Math

Phi-1.5 (2024 by Microsoft)

1. Technical Specs and Hardware Requirements:
   • Parameters: 1.5 billion
   • RAM: 3GB (smartphones), 6GB (PCs)
   • CPU: Snapdragon 845 or higher
   • Storage: 15GB
2. Benchmark: Fast performance, struggles with complex math problems.

---

Dolly-v2-3B (2023 by Databricks)

1. Technical Specs and Hardware Requirements:
   • Parameters: 3 billion
   • RAM: 4GB (smartphones), 8GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 25GB
2. Benchmark: High accuracy but higher RAM usage.

---

LLaMA 2 (7B) (2023 by Meta)

1. Technical Specs and Hardware Requirements:
   • Parameters: 7 billion
   • RAM: 6GB (smartphones)
   • CPU: Snapdragon 855 or higher
   • Storage: 8GB
2. Benchmark: Efficient for basic math, struggles with advanced calculations.

---

4. Image Understanding

LLaMA 2 (7B) (2023 by Meta)

1. Technical Specs and Hardware Requirements:
   • Parameters: 7 billion
   • RAM: 6GB (smartphones), 8GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 10GB
2. Benchmark: Basic image understanding tasks, works on mid-tier smartphones.

---

Dolly-v2-3B (2023 by Databricks)

1. Technical Specs and Hardware Requirements:
   • Parameters: 3 billion
   • RAM: 4GB (smartphones), 8GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 20GB
2. Benchmark: Decent performance in image-to-text tasks but requires more RAM.

---

Phi-3-Mini (2024 by Microsoft)

1. Technical Specs and Hardware Requirements:
   • Parameters: 3 billion
   • RAM: 4GB (smartphones), 6GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 10GB
2. Benchmark: Works well with basic image recognition but limited context handling.

---

5. Translation (Chinese, Japanese)

LLaMA 2 (7B) (2023 by Meta)

1. Technical Specs and Hardware Requirements:
   • Parameters: 7 billion
   • RAM: 6GB (smartphones)
   • CPU: Snapdragon 855 or higher
   • Storage: 10GB
2. Benchmark: Efficient for on-device translations but limited language precision.

---

Phi-3-Small (2024 by Microsoft)

1. Technical Specs and Hardware Requirements:
   • Parameters: 3 billion
   • RAM: 4GB (smartphones), 6GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 8GB
2. Benchmark: Good for basic translations but slower in complex phrases.

---

Mistral 7B (2023 by Mistral AI)

1. Technical Specs and Hardware Requirements:
   • Parameters: 7 billion
   • RAM: 4GB (smartphones), 8GB (PCs)
   • CPU: Snapdragon 855 or higher
   • Storage: 20GB
2. Benchmark: Outperforms Llama 2 13B on all benchmarks; limited context window of 8,000 tokens; good for basic translation tasks, handles multiple languages decently.

---