Developers should ensure that the system that utilizes the model provides relevant transparency to the end users. Important considerations and information include: 

• The users are informed that they are interacting with an AI model which generates text-based responses.
• The users of this AI model are responsible for determining the accuracy of any content it generates.
• The users should have human oversight in basing their decisions leveraging the outputs of the model.
• The users should only use the model to support queries regarding crop protection. The model is not intended to replace the guidance of subject matter experts, and the users should validate the responses with subject matter experts when in doubt.
• The model is based on Microsoft’s Phi-3-mini-128k-instruct model. The capabilities and limitations of the Phi-3 model can be found in the Phi-3-mini-128k-instruct model card on Azure AI Studio .

Intended Use

Primary Use Cases

This model is intended to be used by Agronomists and is focused on providing timely and valuable weed, disease and pest control information, based on a good understanding of Bayer's US crop protection product labels and safety data sheets.

Out-of-Scope Use Cases

This model is intended for crop protection product-specific applications in the Agriculture domain only and cannot be used for scenarios that are not explicitly mentioned in the Primary use cases section above Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.

Model Architecture

This model retains the architecture of the Phi-3-Mini-128K-instruct model which has 3.8 billion parameters and is a dense decoder-only Transformer model.

Supported VM SKUs

• NC A100 v4-series - Azure Virtual Machines | Microsoft Learn
• ND A100 v4-series - Azure Virtual Machines | Microsoft Learn

Trademarks

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

Responsible AI Considerations

The model 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 model is 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: The model 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: the model 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 model can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated.
• Limited Scope for Code: Majority of model 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: Model 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 model 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: Model 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 Data

Our training data includes the following proprietary and public data sources and is a combination of:

• Proprietary Bayer crop protection product data including labels, Safety Data Sheets and supplemental information
• Publicly available general instruction datasets for reasoning and math skills

Domain-Specific Fine-Tuning Benchmark

E.L.Y.Crop-Protection model delivers factual responses to agronomy questions, focusing on US crop protection for weeds, diseases and insects. When benchmarked against Phi-3 baseline, GPT-3.5-Turbo, GPT-4o-mini and GPT-4o on crop protection related evaluation datasets, the E.L.Y.Crop-Protection model demonstrated state-of-the-art performance.

Model	% of fully correct answers	% of partially correct answers
Phi-3 baseline	5%	15.2%
GPT-3.5-Turbo	18.1%	16.9%
GPT-4o-mini	14.8%	22.5%
GPT-4o	16.5%	29.8%
E.L.Y.Crop-Protection model	50.3%	15.6%