Update README.md
Browse files
README.md
CHANGED
@@ -30,10 +30,8 @@ This repository contains a fine-tuned version of **Meta’s Llama 3.1 3B Instruc
|
|
30 |
The fine-tuning process involves using **QLoRA** to adapt the pre-trained model while maintaining memory efficiency and computational feasibility. This technique allows for fine-tuning large-scale models on consumer-grade GPUs by leveraging **NF4** 4-bit quantization.
|
31 |
|
32 |
- **Developed by [FineTuned]:** Karthik Manjunath Hadagali
|
33 |
-
- **Funded by [optional]:** [More Information Needed]
|
34 |
-
- **Shared by [optional]:** [More Information Needed]
|
35 |
- **Model type:** Text-Generation
|
36 |
-
- **Language(s) (NLP):**
|
37 |
- **License:** [More Information Needed]
|
38 |
- **Fine-Tuned from model [optional]:** Meta Llama 3.1 3B Instruct
|
39 |
- **Fine-Tuning Method:** QLoRA
|
@@ -87,7 +85,21 @@ Users (both direct and downstream) should be made aware of the risks, biases and
|
|
87 |
|
88 |
Use the code below to get started with the model.
|
89 |
|
90 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
|
92 |
## Training Details
|
93 |
|
@@ -97,9 +109,9 @@ Use the code below to get started with the model.
|
|
97 |
|
98 |
The model has been fine-tuned on the **dmedhi/wiki_medical_terms** dataset. This dataset is designed to improve medical terminology comprehension and consists of:
|
99 |
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
|
104 |
This dataset ensures that the fine-tuned model performs well in understanding and responding to medical queries with enhanced accuracy.
|
105 |
|
@@ -140,43 +152,6 @@ This dataset ensures that the fine-tuned model performs well in understanding an
|
|
140 |
- **Training Time:** ~3-4 hours per epoch on A100 40GB GPU
|
141 |
- **Final Checkpoint Size:** ~2.8GB (with LoRA adapters stored separately)
|
142 |
|
143 |
-
## Evaluation
|
144 |
-
|
145 |
-
<!-- This section describes the evaluation protocols and provides the results. -->
|
146 |
-
|
147 |
-
### Testing Data, Factors & Metrics
|
148 |
-
|
149 |
-
#### Testing Data
|
150 |
-
|
151 |
-
<!-- This should link to a Dataset Card if possible. -->
|
152 |
-
|
153 |
-
[More Information Needed]
|
154 |
-
|
155 |
-
#### Factors
|
156 |
-
|
157 |
-
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
158 |
-
|
159 |
-
[More Information Needed]
|
160 |
-
|
161 |
-
#### Metrics
|
162 |
-
|
163 |
-
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
164 |
-
|
165 |
-
[More Information Needed]
|
166 |
-
|
167 |
-
### Results
|
168 |
-
|
169 |
-
[More Information Needed]
|
170 |
-
|
171 |
-
#### Summary
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
## Model Examination [optional]
|
176 |
-
|
177 |
-
<!-- Relevant interpretability work for the model goes here -->
|
178 |
-
|
179 |
-
[More Information Needed]
|
180 |
|
181 |
## Environmental Impact
|
182 |
|
@@ -190,50 +165,26 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
|
|
190 |
- **Compute Region:** US-East
|
191 |
- **Carbon Emitted:** [More Information Needed]
|
192 |
|
193 |
-
##
|
194 |
-
|
195 |
-
### Model Architecture and Objective
|
196 |
-
|
197 |
-
[More Information Needed]
|
198 |
-
|
199 |
-
### Compute Infrastructure
|
200 |
-
|
201 |
-
[More Information Needed]
|
202 |
-
|
203 |
-
#### Hardware
|
204 |
-
|
205 |
-
[More Information Needed]
|
206 |
|
207 |
-
|
|
|
|
|
208 |
|
209 |
-
|
210 |
-
|
211 |
-
## Citation [optional]
|
212 |
|
213 |
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
|
|
214 |
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
[More Information Needed]
|
222 |
-
|
223 |
-
## Glossary [optional]
|
224 |
-
|
225 |
-
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
226 |
-
|
227 |
-
[More Information Needed]
|
228 |
-
|
229 |
-
## More Information [optional]
|
230 |
-
|
231 |
-
[More Information Needed]
|
232 |
-
|
233 |
-
## Model Card Authors [optional]
|
234 |
-
|
235 |
-
[More Information Needed]
|
236 |
|
237 |
-
##
|
238 |
|
239 |
-
|
|
|
|
|
|
30 |
The fine-tuning process involves using **QLoRA** to adapt the pre-trained model while maintaining memory efficiency and computational feasibility. This technique allows for fine-tuning large-scale models on consumer-grade GPUs by leveraging **NF4** 4-bit quantization.
|
31 |
|
32 |
- **Developed by [FineTuned]:** Karthik Manjunath Hadagali
|
|
|
|
|
33 |
- **Model type:** Text-Generation
|
34 |
+
- **Language(s) (NLP):** Python
|
35 |
- **License:** [More Information Needed]
|
36 |
- **Fine-Tuned from model [optional]:** Meta Llama 3.1 3B Instruct
|
37 |
- **Fine-Tuning Method:** QLoRA
|
|
|
85 |
|
86 |
Use the code below to get started with the model.
|
87 |
|
88 |
+
```python
|
89 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
90 |
+
import torch
|
91 |
+
|
92 |
+
# Load the fine-tuned model
|
93 |
+
model_id = "your-hf-username/llama-3.1-3b-medical-qlora"
|
94 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
95 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto")
|
96 |
+
|
97 |
+
# Example query
|
98 |
+
input_text = "What is the medical definition of pneumonia?"
|
99 |
+
inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
|
100 |
+
outputs = model.generate(**inputs, max_new_tokens=100)
|
101 |
+
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
102 |
+
|
103 |
|
104 |
## Training Details
|
105 |
|
|
|
109 |
|
110 |
The model has been fine-tuned on the **dmedhi/wiki_medical_terms** dataset. This dataset is designed to improve medical terminology comprehension and consists of:
|
111 |
|
112 |
+
✅ Medical definitions and terminologies
|
113 |
+
✅ Disease symptoms and conditions
|
114 |
+
✅ Healthcare and clinical knowledge from Wikipedia's medical section
|
115 |
|
116 |
This dataset ensures that the fine-tuned model performs well in understanding and responding to medical queries with enhanced accuracy.
|
117 |
|
|
|
152 |
- **Training Time:** ~3-4 hours per epoch on A100 40GB GPU
|
153 |
- **Final Checkpoint Size:** ~2.8GB (with LoRA adapters stored separately)
|
154 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
155 |
|
156 |
## Environmental Impact
|
157 |
|
|
|
165 |
- **Compute Region:** US-East
|
166 |
- **Carbon Emitted:** [More Information Needed]
|
167 |
|
168 |
+
## Limitations & Considerations
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
169 |
|
170 |
+
❗ Not a substitute for professional medical advice
|
171 |
+
❗ May contain biases from training data
|
172 |
+
❗ Limited knowledge scope (not updated in real-time)
|
173 |
|
174 |
+
## Citation
|
|
|
|
|
175 |
|
176 |
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
177 |
+
If you use this model, please consider citing:
|
178 |
|
179 |
+
@article{llama3.1_medical_qlora,
|
180 |
+
title={Fine-tuned Llama 3.1 3B Instruct for Medical Knowledge with QLoRA},
|
181 |
+
author={Karthik Manjunath Hadagali},
|
182 |
+
year={2024},
|
183 |
+
journal={Hugging Face Model Repository}
|
184 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
185 |
|
186 |
+
## Acknowledgments
|
187 |
|
188 |
+
- Meta AI for the Llama 3.1 3B Instruct Model.
|
189 |
+
- Hugging Face PEFT for QLoRA implementation.
|
190 |
+
- dmedhi/wiki_medical_terms dataset contributors.
|