Vijayendra
commited on
Update README.md
Browse files
README.md
CHANGED
@@ -23,7 +23,57 @@ The Phi-4 Medical QA Model is a fine-tuned version of the "unsloth/phi-4" langua
|
|
23 |
|
24 |
### Model Description
|
25 |
|
26 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
|
28 |
|
29 |
## How to Use
|
|
|
23 |
|
24 |
### Model Description
|
25 |
|
26 |
+
Model Description
|
27 |
+
|
28 |
+
The Phi-4 Medical QA Model builds upon the robust foundation provided by the "unsloth/phi-4" pre-trained language model. Fine-tuned on the PubMedQA dataset, it is specifically designed to answer complex medical questions by integrating domain-specific knowledge and language understanding capabilities.
|
29 |
+
|
30 |
+
The model employs several advanced techniques:
|
31 |
+
|
32 |
+
LoRA Fine-Tuning: Low-Rank Adaptation (LoRA) enhances parameter efficiency, allowing domain adaptation with minimal compute.
|
33 |
+
|
34 |
+
4-Bit Quantization: Memory usage is significantly reduced, making the model deployable on resource-constrained systems.
|
35 |
+
|
36 |
+
Cyclic Attention and Gradient Checkpointing: Further optimizations for handling long sequences and reducing GPU memory usage.
|
37 |
+
|
38 |
+
The model is trained using the SFTTrainer library from the trl package, with parameters optimized for accuracy and resource efficiency.
|
39 |
+
|
40 |
+
Model Architecture
|
41 |
+
|
42 |
+
Base Model: "unsloth/phi-4"
|
43 |
+
|
44 |
+
Tokenization: Custom tokenizer from the unsloth framework
|
45 |
+
|
46 |
+
Fine-Tuning Techniques:
|
47 |
+
|
48 |
+
Targeted modules: q_proj, k_proj, v_proj, and others
|
49 |
+
|
50 |
+
LoRA Rank: 16
|
51 |
+
|
52 |
+
LoRA Alpha: 16
|
53 |
+
|
54 |
+
Dropout: 0 (optimized for this use case)
|
55 |
+
|
56 |
+
Training Dataset: PubMedQA (labeled fold0 source)
|
57 |
+
|
58 |
+
Hardware Used: NVIDIA A100 GPUs
|
59 |
+
|
60 |
+
Intended Use
|
61 |
+
|
62 |
+
This model is intended for:
|
63 |
+
|
64 |
+
Answering medical and healthcare-related questions.
|
65 |
+
|
66 |
+
Supporting healthcare professionals and students with evidence-based insights.
|
67 |
+
|
68 |
+
Enhancing patient care via interactive QA systems.
|
69 |
+
|
70 |
+
Limitations
|
71 |
+
|
72 |
+
Domain Restriction: The model performs best on medical questions and may not generalize well to other domains.
|
73 |
+
|
74 |
+
Bias and Fairness: The model inherits biases from the PubMedQA dataset.
|
75 |
+
|
76 |
+
Hallucination Risks: As with all large language models, responses should be validated by professionals before application in critical scenarios.
|
77 |
|
78 |
|
79 |
## How to Use
|