Phi-3.5-mini-instruct

Phi-3.5-mini is the latest model from Microsoft’s Phi series of small language models, combining compactness with high performance. Built on an architecture with 3.8 billion parameters, it can run locally even on modern smartphones, making it one of the most accessible and efficient language models on the market. Thanks to its use of carefully curated and synthetic training data, Phi-3.5-mini delivers results comparable to much larger models such as GPT-3.5 and Mixtral 8x7B, while requiring significantly fewer computational resources.

The uniqueness of Phi-3.5-mini lies in its training approach: instead of simply increasing the model size, developers focused on the quality and relevance of the data. By using carefully filtered web sources and synthetic examples, the model achieves a “data optimal regime”—maximizing the effectiveness of each parameter. This enables Phi-3.5-mini to deliver outstanding performance in reasoning, mathematics, programming, and dialogue tasks, all while remaining compact and fast.

Phi-3.5-mini is particularly well-suited for edge devices, mobile applications, chatbots, educational platforms, and any scenarios where privacy and offline operation are important. The model is ideal for building multilingual assistants, text generation and analysis, solving mathematical and logical problems, and integration into products with limited computational resources.

Announce Date: 23.04.2024
Parameters: 4B
Context: 132K
Layers: 32
Attention Type: Full Attention
Developer: Microsoft
Transformers Version: 4.43.3
License: MIT

Use our pre-built public endpoints for free to test inference and explore Phi-3.5-mini-instruct capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.