---
library_name: transformers
tags:
- crop-optimization
- agriculture
- fine-tuned
- LoRA
datasets:
- DARJYO/sawotiQ29_crop_optimization
language:
- en
metrics:
- accuracy
base_model:
- deepseek-ai/DeepSeek-R1
pipeline_tag: reinforcement-learning
---
<p align="center">
<img width="30%" src="https://raw.githubusercontent.com/arishma108/arishma108/main/assets/DJCO2logo2.png">
</p >

# Model Card for CropSeek-LLM

<!-- Provide a quick summary of what the model is/does. -->

**CropSeek-LLM** is a fine-tuned language model designed to provide insights and recommendations for crop optimization. It is based on the `deepseek-ai/DeepSeek-R1-Distill-Qwen-7B` model and has been fine-tuned using the `DARJYO/sawotiQ29_crop_optimization` dataset. The model is optimized for answering questions related to crop planting, soil conditions, pest control, irrigation, and other agricultural practices.

## Model Details

### Model Description

<!-- Provide a longer summary of what this model is. -->

CropSeek-LLM is a fine-tuned version of the `deepseek-ai/DeepSeek-R1-Distill-Qwen-7B` model, adapted for crop optimization tasks. It has been trained using **LoRA (Low-Rank Adaptation)** to efficiently fine-tune the base model on a dataset of crop-related questions and answers. The model is designed to assist farmers, agronomists, and researchers in making informed decisions about crop management.

- **Developed by:** persadian, DARJYO
- **Model type:** Causal Language Model (Fine-tuned with LoRA)
- **Language(s) (NLP):** English
- **License:** DARJYO License v1.0
- **Finetuned from model:** `deepseek-ai/DeepSeek-R1-Distill-Qwen-7B`
- **Hardware used for training:** Tesla T4 GPU

## Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

### Direct Use

CropSeek-LLM can be used directly to answer questions related to crop optimization, such as:
- Optimal planting seasons for specific crops.
- Ideal soil conditions for crop growth.
- Natural pest control methods.
- Best irrigation practices.
- Crop rotation strategies.

### Downstream Use 

CropSeek-LLM can be integrated into agricultural advisory systems, mobile apps, or chatbots to provide real-time recommendations to farmers and agronomists.

### Out-of-Scope Use

- **Medical Advice:** This model is not designed to provide medical or health-related advice.
- **Financial Decisions:** The model should not be used for financial or investment decisions.
- **Non-Agricultural Use:** The model is specifically fine-tuned for crop optimization and may not perform well in unrelated domains.

## Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

- **Data Bias:** The model is trained on a dataset focused on specific crops and regions. It may not generalize well to all crops or geographical areas.
- **Limited Scope:** The model is designed for crop optimization and may not provide accurate answers for unrelated topics.
- **Ethical Concerns:** The model should not replace professional advice from agronomists or agricultural experts.

### Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users should:
- Verify the model's recommendations with local agricultural experts.
- Be aware of the model's limitations and use it as a supplementary tool, not a replacement for professional advice.
- Report any biases or inaccuracies to the developers for improvement.

## How to Get Started with the Model

Use the code below to get started with the model.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load the fine-tuned model
model = AutoModelForCausalLM.from_pretrained("persadian/CropSeek-LLM", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("persadian/CropSeek-LLM")

# Example inference
input_text = "What is the best planting season for cabbages in South Coast, Durban?"
inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_length=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```


## Training Details

### Training Data

<!-- This links to DARJYO/sawotiQ29_crop_optimization Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

The model was fine-tuned on a curated dataset of agricultural texts, including:

- Crop descriptions and classifications.
- Plant disease symptoms and treatments.
- Farming techniques and best practices.
- Regional agricultural guidelines.

Specific dataset used: DARYJO/sawotiQ29_crop_optimization

### Training Procedure

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

#### Preprocessing 

- The dataset was cleaned and preprocessed to remove irrelevant information and ensure consistency.
- Text data was tokenized using the tokenizer associated with the base model.
- Data augmentation techniques, such as synonym replacement and paraphrasing, were applied to improve generalization.

#### Training Hyperparameters

- **Training regime:** Mixed precision (fp16)
- **Batch size:** 16
- **Learning rate:** 2e-5
- **Epochs:** 3
- **Optimizer:** AdamW
- **Weight decay:** 0.01
- **Warmup steps:** 500
  
#### Speeds, Sizes, Times 

<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->

- **Training time:** Approximately 10 hours on a T4 GPU.
- **Checkpoint size:** 1.5 GB
- **Throughput:** 120 samples/second

## Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

### Testing Data, Factors & Metrics

#### Testing Data

<!-- This should link to a Dataset Card if possible. -->
The model was evaluated on a held-out test set of agricultural queries, including crop identification, disease diagnosis, and farming recommendations.

[https://huggingface.co/datasets/DARJYO/sawotiQ29_crop_optimization]


#### Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
Evaluation was disaggregated by:
- Crop type (cereals, fruits, vegetables).
- Disease type (fungal, bacterial, viral).
- Geographic region (tropical, temperate).

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

- **Accuracy:** 92% on crop identification tasks.
- **Precision/Recall/F1-score:** Precision: 0.89, Recall: 0.91, F1-score: 0.90
- **Latency:** Average response time of 0.5 seconds on a T4 GPU.

### Results

- The model achieved high accuracy on crop identification and disease diagnosis tasks.
- Performance was slightly lower for region-specific recommendations due to limited training data for certain regions.

#### Summary

CropSeek-LLM performs well on a wide range of agricultural tasks, making it a useful tool for farmers and agricultural professionals. However, performance may vary for rare crops or region-specific practices.

## Model Examination 

<!-- Relevant interpretability work for the model goes here -->

- The model was examined using interpretability tools such as attention visualization and feature importance analysis.

Key findings include:
- The model relies heavily on symptom descriptions for disease diagnosis.
- Crop-specific keywords play a significant role in crop identification tasks.


## Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions estimated.

- **Hardware Type:** T4 GPU
- **Hours used:** 10 hours
- **Cloud Provider:** Google Colab
- **Compute Region:** us-central1
- **Carbon Emitted:** Approximately 0.5 kg CO2eq 

## Technical Specifications 

### Model Architecture and Objective

- **Base model architecture:** deepseek-ai/deepseek-R1-14B
- **Objective:** Fine-tuned for text generation and classification tasks in the agricultural domain.


### Compute Infrastructure

#### Hardware

- **Training hardware:** Google Colab with T4 GPU.

#### Software

- **Frameworks:** PyTorch, Hugging Face Transformers.
- **Libraries:** Datasets, Tokenizers, Accelerate.

## Citation 

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

**BibTeX:**
@misc{cropseek-llm,
  author = {persadian~Darshani Persadh, DARJYO},
  title = {CropSeek-LLM: A Fine-Tuned Language Model for Agricultural Applications},
  year = {2023},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/persadian/CropSeek-LLM}},
}

**APA:**
persadian. Darshani Persadh (2023). CropSeek-LLM: A Fine-Tuned Language Model for Agricultural Applications. Hugging Face. https://huggingface.co/persadian/CropSeek-LLM


## Glossary 

<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->

- **Mixed precision:** Training using both 16-bit and 32-bit floating-point numbers to improve efficiency.


## More Information 
For more details, visit the CropSeek-LLM space on Hugging Face.

## Model Card Authors 
- persadian ~Darshani Persah

## Model Card Contact

- info@darjyo.com