This repository contains executable weights (which we call
llamafiles) that run on
Linux, MacOS, Windows, FreeBSD, OpenBSD, and NetBSD for AMD64 and ARM64.
Alternatively, you may download an official llamafile executable from
Mozilla Ocho on GitHub, in which case you can use the Mixtral llamafiles
as a simple weights data file.
llamafile is a new format introduced by Mozilla Ocho on Nov 20th 2023.
It uses Cosmopolitan Libc to turn LLM weights into runnable llama.cpp
binaries that run on the stock installs of six OSes for both ARM64 and
AMD64.
In addition to being executables, llamafiles are also zip archives. Each
llamafile contains a GGUF file, which you can extract using the unzip
command. If you want to change or add files to your llamafiles, then the
zipalign command (distributed on the llamafile github) should be used
instead of the traditional zip command.
Licensing (Phi-3 Specific)
The Phi-3 llamafiles are licensed Apache 2.0 because some of the
software that went into creating these llamafiles uses that as its
license. The Phi-3 weights themselves were published by Microsoft under
the even more permissive MIT license. You can use the unzip command to
extract the MIT-licensed GGUF file from each llamafile, which contains
only the Microsoft Phi-3 weights.
For further details on the complete picture, read our LICENSE file,
since it documents the copyright notice of every transitive dependency.
About Quantization Formats (General Advice)
Your choice of quantization format depends on three things:
Will it fit in RAM or VRAM?
Is your use case reading (e.g. summarization) or writing (e.g. chatbot)?
llamafiles bigger than 4.30 GB are hard to run on Windows (see gotchas)
Good quants for writing (prediction speed) are Q5_K_M, and Q4_0. Text
generation is bounded by memory speed, so smaller quants help, but they
cause the LLM to hallucinate more. However that doesn't mean they can't
think correctly. A highly degraded quant like Q2_K may not make a
great encyclopedia, but it's still capable of logical reasoning and
the emergent capabilities LLMs exhibit.
Good quants for reading (evaluation speed) are BF16, F16, Q8_0, and
Q4_0 (ordered from fastest to slowest). Prompt evaluation is bounded by
flop count, which means perf can be improved through software
engineering alone, e.g. BLAS algorithms, in which case quantization
starts hurting more than it helps, since it competes for CPU resources
and makes it harder for the compiler to parallelize instructions. You
want to ideally use the simplest smallest floating point format that's
natively implemented by your hardware. In most cases, that's BF16 or
FP16. However, llamafile is able to still offer respectable tinyBLAS
speedups for llama.cpp's simplest quants: Q8_0 and Q4_0.
--
Model Summary
The Phi-3-Mini-4K-Instruct is a 3.8B parameters, lightweight, state-of-the-art open model trained with the Phi-3 datasets that includes both synthetic data and the filtered publicly available websites data with a focus on high-quality and reasoning dense properties.
The model belongs to the Phi-3 family with the Mini version in two variants 4K and 128K which is the context length (in tokens) that it can support.
The model has underwent a post-training process that incorporates both supervised fine-tuning and direct preference optimization for the instruction following and safety measures.
When assessed against benchmarks testing common sense, language understanding, math, code, long context and logical reasoning, Phi-3 Mini-4K-Instruct showcased a robust and state-of-the-art performance among models with less than 13 billion parameters.
The model is intended for commercial and research use in English. The model provides uses for applications which require:
Memory/compute constrained environments
Latency bound scenarios
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.
Use case considerations
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 fariness 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.
How to Use
Phi-3 Mini-4K-Instruct has been integrated in the development version (4.40.0) of transformers. Until the official version is released through pip, ensure that you are doing one of the following:
When loading the model, ensure that trust_remote_code=True is passed as an argument of the from_pretrained() function.
Update your local transformers to the development version: pip uninstall -y transformers && pip install git+https://github.com/huggingface/transformers. The previous command is an alternative to cloning and installing from the source.
The current transformers version can be verified with: pip list | grep transformers.
Phi-3 Mini-4K-Instruct is also available in HuggingChat.
Tokenizer
Phi-3 Mini-4K-Instruct supports a vocabulary size of up to 32064 tokens. The tokenizer files already provide placeholder tokens that can be used for downstream fine-tuning, but they can also be extended up to the model's vocabulary size.
Chat Format
Given the nature of the training data, the Phi-3 Mini-4K-Instruct model is best suited for prompts using the chat format as follows.
You can provide the prompt as a question with a generic template as follow:
<|user|>\nQuestion <|end|>\n<|assistant|>
For example:
markdown
1<|system|>2You are a helpful AI assistant.<|end|>3<|user|>4How to explain Internet for a medieval knight?<|end|>5<|assistant|>
where the model generates the text after <|assistant|> . In case of few-shots prompt, the prompt can be formatted as the following:
markdown
1<|system|>2You are a helpful AI assistant.<|end|>3<|user|>4I am going to Paris, what should I see?<|end|>5<|assistant|>6Paris, the capital of France, is known for its stunning architecture, art museums, historical landmarks, and romantic atmosphere. Here are some of the top attractions to see in Paris:\n\n1. The Eiffel Tower: The iconic Eiffel Tower is one of the most recognizable landmarks in the world and offers breathtaking views of the city.\n2. The Louvre Museum: The Louvre is one of the world's largest and most famous museums, housing an impressive collection of art and artifacts, including the Mona Lisa.\n3. Notre-Dame Cathedral: This beautiful cathedral is one of the most famous landmarks in Paris and is known for its Gothic architecture and stunning stained glass windows.\n\nThese are just a few of the many attractions that Paris has to offer. With so much to see and do, it's no wonder that Paris is one of the most popular tourist destinations in the world."<|end|>7<|user|>8What is so great about #1?<|end|>9<|assistant|>
Sample inference code
This code snippets show how to get quickly started with running the model on a GPU:
python
1import torch
2from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
34torch.random.manual_seed(0)56model = AutoModelForCausalLM.from_pretrained(7"microsoft/Phi-3-mini-4k-instruct",8 device_map="cuda",9 torch_dtype="auto",10 trust_remote_code=True,11)12tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct")1314messages =[15{"role":"system","content":"You are a helpful digital assistant. Please provide safe, ethical and accurate information to the user."},16{"role":"user","content":"Can you provide ways to eat combinations of bananas and dragonfruits?"},17{"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."},18{"role":"user","content":"What about solving an 2x + 3 = 7 equation?"},19]2021pipe = pipeline(22"text-generation",23 model=model,24 tokenizer=tokenizer,25)2627generation_args ={28"max_new_tokens":500,29"return_full_text":False,30"temperature":0.0,31"do_sample":False,32}3334output = pipe(messages,**generation_args)35print(output[0]['generated_text'])
Some applications/frameworks might not include a BOS token (<s>) at the start of the conversation. Please ensure that it is included since it provides more reliable results.
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.
Training
Model
Architecture: Phi-3 Mini-4K-Instruct has 3.8B parameters and is a dense decoder-only Transformer model. The model is fine-tuned with Supervised fine-tuning (SFT) and Direct Preference Optimization (DPO) to ensure alignment with human preferences and safety guidlines.
Inputs: Text. It is best suited for prompts using chat format.
Context length: 4K tokens
GPUs: 512 H100-80G
Training time: 7 days
Training data: 3.3T tokens
Outputs: Generated text in response to the input
Dates: Our models were trained between February and April 2024
Status: This is a static model trained on an offline dataset with cutoff date October 2023. Future versions of the tuned models may be released as we improve models.
Datasets
Our training data includes a wide variety of sources, totaling 3.3 trillion tokens, and is a combination of
Publicly available documents filtered rigorously for quality, selected high-quality educational data, and code;
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.);
High quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness.
Fine-tuning
A basic example of multi-GPUs supervised fine-tuning (SFT) with TRL and Accelerate modules is provided here.
Benchmarks
We report the results for Phi-3-Mini-4K-Instruct on standard open-source benchmarks measuring the model's reasoning ability (both common sense reasoning and logical reasoning). We compare to Phi-2, Mistral-7b-v0.1, Mixtral-8x7b, Gemma 7B, Llama-3-8B-Instruct, and GPT-3.5.
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.
Note that by default, the Phi-3-mini model uses flash attention, which requires certain types of GPU hardware to run. We have tested on the following GPU types:
NVIDIA A100
NVIDIA A6000
NVIDIA H100
If you want to run the model on:
NVIDIA V100 or earlier generation GPUs: call AutoModelForCausalLM.from_pretrained() with attn_implementation="eager"
Optimized inference on GPU, CPU, and Mobile: use the ONNX models 4K
Cross Platform Support
ONNX runtime ecosystem now supports Phi-3 Mini models across platforms and hardware. You can find the optimized Phi-3 Mini-4K-Instruct ONNX model here.
Optimized Phi-3 models are also published here in ONNX format, to run with ONNX Runtime on CPU and GPU across devices, including server platforms, Windows, Linux and Mac desktops, and mobile CPUs, with the precision best suited to each of these targets. DirectML support lets developers bring hardware acceleration to Windows devices at scale across AMD, Intel, and NVIDIA GPUs.
Along with DirectML, ONNX Runtime provides cross platform support for Phi-3 across a range of devices CPU, GPU, and mobile.
Here are some of the optimized configurations we have added:
ONNX models for int4 DML: Quantized to int4 via AWQ
ONNX model for fp16 CUDA
ONNX model for int4 CUDA: Quantized to int4 via RTN
ONNX model for int4 CPU and Mobile: Quantized to int4 via RTN
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.
