APHRC-v2-q: African Health Research Model (Quantized)
This is a quantized version of the APHRC model fine-tuned on African health and population research data using QLoRA.
Model Details
- Base Model: mistralai/Mistral-7B-Instruct-v0.3
- Fine-tuned On: African health and population research datasets
- Method: QLoRA (4-bit quantization)
- Architecture: Mistral 7B with LoRA adapters
- Purpose: Specialized for African health and demographic research questions
How to Use
Load the model using the PEFT library:
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
# Load base model
base_model = "mistralai/Mistral-7B-Instruct-v0.3"
tokenizer = AutoTokenizer.from_pretrained(base_model)
# Load the LoRA adapter from this repository
model = PeftModel.from_pretrained(base_model, "adamsw/aphrc-v2-q")
model = model.merge_and_unload() # Optional: merge for full performance
# Use the model
inputs = tokenizer("Explain African health statistics.", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Training Configuration
- QLoRA Configuration: 4-bit NF4 quantization
- LoRA Ranks: 32
- LoRA Alpha: 64
- Training Epochs: 3
- Batch Size: 2 per device with 4 gradient accumulation steps
Intended Use
This model is designed to answer questions related to:
- African health research
- Population studies in Africa
- Demographic surveillance systems (HDSS)
- Health statistics and trends in Africa
- Public health interventions in African contexts
Limitations
- The model is specialized for African health topics and may not perform well on other domains
- As a quantized model, it may have slightly reduced accuracy compared to the full-precision version
- Responses should be verified by domain experts for critical applications
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support