microsoft / phi-3.5-moe-instruct

Model Summary

Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.

Third-Party Community Consideration

This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party’s requirements for this application and use case.

Resources

🏡 Phi-3 Portal

📰 Phi-3 Microsoft Blog

📖 Phi-3 Technical Report

🛠️ Phi-3 on Azure AI Studio

👩‍🍳 Phi-3 Cookbook

Intended Use

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.

Usage

Input Formats

Given the nature of the training data, the Phi-3.5-MoE-instruct model is best suited for prompts using the chat format as follows:

<|system|>
You are a helpful assistant.<|end|>
<|user|>
How to explain Internet for a medieval knight?<|end|>
<|assistant|>

Loading the model locally

After obtaining the Phi-3.5-MoE-instruct model checkpoints, users can use this sample code for inference.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline 

torch.random.manual_seed(0) 

model = AutoModelForCausalLM.from_pretrained( 
    "microsoft/Phi-3.5-MoE-instruct",  
    device_map="cuda",  
    torch_dtype="auto",  
    trust_remote_code=True,  
) 

tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3.5-moe-instruct") 

messages = [ 
    {"role": "system", "content": "You are a helpful AI assistant."}, 
    {"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"}, 
    {"role": "assistant", "content": "Sure! Here are some ways to eat bananas and dragonfruits together: 1. Banana and dragonfruit smoothie: Blend bananas and dragonfruits together with some milk and honey. 2. Banana and dragonfruit salad: Mix sliced bananas and dragonfruits together with some lemon juice and honey."}, 
    {"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"}, 
] 

pipe = pipeline( 
    "text-generation", 
    model=model, 
    tokenizer=tokenizer, 
) 

generation_args = { 
    "max_new_tokens": 500, 
    "return_full_text": False, 
    "temperature": 0.0, 
    "do_sample": False, 
} 

output = pipe(messages, **generation_args) 
print(output[0]['generated_text'])

Model Version(s):

[Use unique identifier for identifying the model in the future- this may be part of our internal naming, specifying variation like "pruned," "unpruned," "trained," or "deployable" or "quantized" where necessary and including a versioning number like vX.X along with short description differentiating if multiple versions are available]

Datasets

Our training data includes a wide variety of sources, totaling 4.9 trillion tokens (including 10% multilingual), and is a combination of

1. publicly available documents filtered rigorously for quality, selected high-quality educational data, and code;
2. newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (science, daily activities, theory of mind, etc.);
3. high quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness.

We are focusing on the quality of data that could potentially improve the reasoning ability for the model, and we filter the publicly available documents to contain the correct level of knowledge. As an example, the result of a game in premier league in a particular day might be good training data for frontier models, but we need to remove such information to leave more model capacity for reasoning for the small size models. More details about data can be found in the Phi-3 Technical Report.

Benchmarks

To understand the capabilities, we compare Phi-3.5-MoE with a set of models over a variety of benchmarks using our internal benchmark platform. At the high-level overview of the model quality on representative benchmarks:

We take a closer look at different categories across 80 public benchmark datasets at the table below:
| Category | Phi-3.5-MoE-instruct | Mistral-Nemo-12B-instruct-2407 | Llama-3.1-8B-instruct | Gemma-2-9b-It | Gemini-1.5-Flash | GPT-4o-mini-2024-07-18 (Chat) |
|--|--|--|--|--|--|--|
| Popular aggregated benchmark | 62.6 | 51.9 | 50.3 | 56.7 | 64.5 | 73.9 |
| Reasoning | 78.7 | 72.2 | 70.5 | 75.4 | 77.7 | 80.0 |
| Language understanding | 71.8 | 67.0 | 62.9 | 72.8 | 66.6 | 76.8 |
| Robustness | 75.6 | 65.2 | 59.8 | 64.7 | 68.9 | 77.5 |
| Long context | 25.5 | 24.5 | 25.5 | 0.0 | 27.0 | 25.4 |
| Math | 74.1 | 57.7 | 65.0 | 67.9 | 60.2 | 80.8 |
| Code generation | 68.3 | 56.9 | 65.8 | 58.3 | 66.8 | 69.9 |
| Multilingual | 65.8 | 55.3 | 47.5 | 59.6 | 64.3 | 76.6 |

Overall, Phi-3.5-MoE with only 6.6B active parameters achieves a similar level of language understanding and math as much larger models. Moreover, the model outperforms bigger models in reasoning capability and only behind GPT-4o-mini. However, it is still fundamentally limited by its size for certain tasks. The model simply does not have the capacity to store too much factual knowledge, therefore, users may experience factual incorrectness. However, we believe such weakness can be resolved by augmenting Phi-3.5 with a search engine, particularly when using the model under RAG settings.

Multilingual

The table below highlights multilingual capability of Phi-3.5-MoE on multilingual MMLU, MEGA, and multilingual MMLU-pro datasets. Overall, we observed that even with just 6.6B active parameters, the model is very competitive on multilingual tasks in comparison to other models with a much bigger active parameters.

Long Context

Phi-3.5-MoE supports 128K context length, therefore the model is capable of several long context tasks including long document/meeting summarization, long document QA, multilingual context retrieval. We see that Phi-3.5 is clearly better than Gemma-2 family which only supports 8K context length. Phi-3.5-MoE-instruct is very competitive with other much larger open-weight models such as Llama-3.1-8B-instruct, and Mistral-Nemo-12B-instruct-2407.

RULER: a retrieval-based benchmark for long context understanding
| Model | 4K | 8K | 16K | 32K | 64K | 128K | Average |
|--|--|--|--|--|--|--|--|
| Phi-3.5-MoE-instruct | 94.8 | 93 | 93.2 | 91.6 | 85.7 | 64.2 | 87.1 |
| Llama-3.1-8B-instruct | 95.5 | 93.8 | 91.6 | 87.4 | 84.7 | 77.0 | 88.3 |
| Mistral-Nemo-12B-instruct-2407 | 87.8 | 87.2 | 87.7 | 69.0 | 46.8 | 19.0 | 66.2 |

RepoQA: a benchmark for long context code understanding
| Model | Python | C++ | Rust | Java | TypeScript | Average |
|--|--|--|--|--|--|--|
| Phi-3.5-MoE-instruct | 89 | 74 | 81 | 88 | 95 | 85 |
| Llama-3.1-8B-instruct | 80 | 65 | 73 | 76 | 63 | 71 |
| Mistral-7B-instruct-v0.3 | 61 | 57 | 51 | 61 | 80 | 62 |

Inference:

Engine: Tensor(RT)

Test Hardware [Name the specific test hardware model]:

• 512 A100-80G
  

Responsible AI Considerations

Like other models, the Phi family of 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.
• Identification of individuals: models with vision capabilities may have the potential to uniquely identify individuals in images. Safety post-training steers the model to refuse such requests, but developers should consider and implement, as appropriate, additional mitigations or user consent flows as required in their respective jurisdiction, (e.g., building measures to blur faces in image inputs before processing).

Security and AI Concerns:

Please report security vulnerabilities or NVIDIA AI Concerns here.

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.