License

The model is licensed under the MIT license.

Intended Uses

Primary use cases The model is intended for broad commercial and research use in English. The model provides uses for general purpose AI systems and applications which require:

1. Memory/compute constrained environments
2. Latency bound scenarios
3. Strong reasoning (especially code, math and logic)

Our model is designed to accelerate research on language and multimodal models, for use as a building block for generative AI powered features. Out-of-scope use cases Our models are not specifically designed or evaluated for all downstream purposes. Developers should consider common limitations of language models as they select use cases, and evaluate and mitigate for accuracy, safety, and fairness before using within a specific downstream use case, particularly for high-risk scenarios. Developers should be aware of and adhere to applicable laws or regulations (including privacy, trade compliance laws, etc.) that are relevant to their use case. Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft’s Trademark & Brand Guidelines . Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.

Responsible AI Considerations

Like other language models, the Phi series models can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include:

• Quality of Service: the Phi models are trained primarily on English text. Languages other than English will experience worse performance. English language varieties with less representation in the training data might experience worse performance than standard American English.
• Representation of Harms & Perpetuation of Stereotypes: These models can over- or under-represent groups of people, erase representation of some groups, or reinforce demeaning or negative stereotypes. Despite safety post-training, these limitations may still be present due to differing levels of representation of different groups or prevalence of examples of negative stereotypes in training data that reflect real-world patterns and societal biases.
• Inappropriate or Offensive Content: these models may produce other types of inappropriate or offensive content, which may make it inappropriate to deploy for sensitive contexts without additional mitigations that are specific to the use case.
• Information Reliability: Language models can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated.
• Limited Scope for Code: Majority of Phi-3 training data is based in Python and use common packages such as "typing, math, random, collections, datetime, itertools". If the model generates Python scripts that utilize other packages or scripts in other languages, we strongly recommend users manually verify all API uses.

Developers should apply responsible AI best practices and are responsible for ensuring that a specific use case complies with relevant laws and regulations (e.g. privacy, trade, etc.). Important areas for consideration include:

• Allocation: Models may not be suitable for scenarios that could have consequential impact on legal status or the allocation of resources or life opportunities (ex: housing, employment, credit, etc.) without further assessments and additional debiasing techniques.
• High-Risk Scenarios: Developers should assess suitability of using models in high-risk scenarios where unfair, unreliable or offensive outputs might be extremely costly or lead to harm. This includes providing advice in sensitive or expert domains where accuracy and reliability are critical (ex: legal or health advice). Additional safeguards should be implemented at the application level according to the deployment context.
• Misinformation: Models may produce inaccurate information. Developers should follow transparency best practices and inform end-users they are interacting with an AI system. At the application level, developers can build feedback mechanisms and pipelines to ground responses in use-case specific, contextual information, a technique known as Retrieval Augmented Generation (RAG).
• Generation of Harmful Content: Developers should assess outputs for their context and use available safety classifiers or custom solutions appropriate for their use case.
• Misuse: Other forms of misuse such as fraud, spam, or malware production may be possible, and developers should ensure that their applications do not violate applicable laws and regulations.

Content Filtering

Prompts and completions are passed through a default configuration of Azure AI Content Safety classification models to detect and prevent the output of harmful content. Learn more about Azure AI Content Safety . Configuration options for content filtering vary when you deploy a model for production in Azure AI; learn more .

We report the results for Phi-3-Small-8K-Instruct on standard open-source benchmarks measuring the model's reasoning ability (both common sense reasoning and logical reasoning). We compare to Mixtral-8x7b, Gemini-Pro, Gemma 7B, Llama-3-8B-Instruct, GPT-3.5-Turbo-1106, and GPT-4-Turbo-1106. All the reported numbers are produced with the exact same pipeline to ensure that the numbers are comparable. These numbers might differ from other published numbers due to slightly different choices in the evaluation. As is now standard, we use few-shot prompts to evaluate the models, at temperature 0.
The prompts and number of shots are part of a Microsoft internal tool to evaluate language models, and in particular we did no optimization to the pipeline for Phi-3.
More specifically, we do not change prompts, pick different few-shot examples, change prompt format, or do any other form of optimization for the model. The number of k–shot examples is listed per-benchmark.

Benchmark	Phi-3-Small-8K-Instruct 7b	Gemma 7B	Mixtral 8x7B	Llama-3-Instruct 8b	GPT-3.5-Turbo version 1106	Gemini Pro	GPT-4-Turbo version 1106 (Chat)
AGI Eval 5-shot	45.1	42.1	45.2	42.0	48.4	49.0	59.6
MMLU 5-shot	75.7	63.6	70.5	66.5	71.4	66.7	84.0
BigBench Hard 3-shot	79.1	59.6	69.7	51.5	68.3	75.6	87.7
ANLI 7-shot	58.1	48.7	55.2	57.3	58.1	64.2	71.7
HellaSwag 5-shot	77.0	49.8	70.4	71.1	78.8	76.2	88.3
ARC Challenge 10-shot	90.7	78.3	87.3	82.8	87.4	88.3	95.6
ARC Easy 10-shot	97.0	91.4	95.6	93.4	96.3	96.1	98.8
BoolQ 2-shot	84.8	66.0	76.6	80.9	79.1	86.4	91.3
CommonsenseQA 10-shot	80.0	76.2	78.1	79.0	79.6	81.8	86.7
MedQA 2-shot	65.4	49.6	62.2	60.5	63.4	58.2	83.7
OpenBookQA 10-shot	88.0	78.6	85.8	82.6	86.0	86.4	93.4
PIQA 5-shot	86.9	78.1	86.0	75.7	86.6	86.2	90.1
Social IQA 5-shot	79.2	65.5	75.9	73.9	68.3	75.4	81.7
TruthfulQA (MC2) 10-shot	70.2	52.1	60.1	63.2	67.7	72.6	85.2
WinoGrande 5-shot	81.5	55.6	62.0	65.0	68.8	72.2	86.7
TriviaQA 5-shot	58.1	72.3	82.2	67.7	85.8	80.2	73.3
GSM8K Chain of Thought 8-shot	89.6	59.8	64.7	77.4	78.1	80.4	94.2
HumanEval 0-shot	61.0	34.1	37.8	60.4	62.2	64.4	79.9
MBPP 3-shot	71.7	51.5	60.2	67.7	77.8	73.2	86.7
Average	75.7	61.8	69.8	69.4	74.3	75.4	85.2

We take a closer look at different categories across 80 public benchmark datasets at the table below:

Benchmark	Phi-3-Small-8K-Instruct 7b	Gemma 7B	Mixtral 8x7B	Llama-3-Instruct 8b	GPT-3.5-Turbo version 1106	Gemini Pro	GPT-4-Turbo version 1106 (Chat)
Popular aggregated benchmark	71.1	59.4	66.2	59.9	67.0	67.5	80.5
Reasoning	82.4	69.1	77.0	75.7	78.3	80.4	89.3
Language understanding	70.6	58.4	64.9	65.4	70.4	75.3	81.6
Code generation	60.7	45.6	52.7	56.4	70.4	66.7	76.1
Math	51.6	35.8	40.3	41.1	52.8	50.9	67.1
Factual knowledge	38.6	46.7	58.6	43.1	63.4	54.6	45.9
Multilingual	62.5	63.2	63.4	65.0	69.1	76.5	82.0
Robustness	72.9	38.4	51.0	64.5	69.3	69.7	84.6