Phi-1 Models by Microsoft Research
About
The Phi-1 family of large language models (LLMs), developed by Microsoft, comprises several models designed for specific tasks, primarily focusing on code generation and reasoning. Phi-1, the initial model in the family, is a transformer-based model with 1.3 billion parameters, specializing in basic Python coding 18. Its training utilized a blend of "textbook quality" data sourced from the web and synthetic data generated using GPT-3.5 18. Despite its relatively small size compared to other LLMs, Phi-1 demonstrates impressive accuracy, exceeding 50% on the HumanEval benchmark for simple Python coding tasks 18. Subsequent models in the Phi family, such as Phi-1.5 and Phi-2, build upon this foundation, expanding capabilities to encompass broader natural language tasks while maintaining a focus on efficiency and high-quality data 911. These models showcase Microsoft's research into creating smaller, more efficient LLMs that rival the performance of much larger models 29.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
Use when the workload needs 2k context and 1.3B parameters.
Use when the workload needs 2k context and 1.3B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Phi-1.5 | Use when the workload needs 2k context and 1.3B parameters. | 2023-09 | 2k context1.3B parameters | Current |
| Phi-1 | Use when the workload needs 2k context and 1.3B parameters. | 2023-06 | 2k context1.3B parameters | Current |
Release Timeline
2 release groupsSpecifications(2 models)
Available From(1 provider)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Phi-1.5 | Microsoft Foundry | $0.07 | $0.07 | Provisioned |
Frequently Asked Questions
- What is Phi-1 used for?
- Phi-1 is used for coding, math-heavy prompts, and structured outputs. The family description and listed model capabilities point to those workloads as the best fit.
- How does Phi-1 compare to Harrier?
- Phi-1 by Microsoft Research is strongest where you need coding, while Harrier by Microsoft Research is the closest related family to check for embedding. Phi-1 has 2 listed variants and reaches up to 2k context, while Harrier reaches up to 33k context, so compare the specs and pricing tables before choosing a production model.
