Update README.md

README.md

@@ -95,20 +95,16 @@ The training data consists of a diverse set of questions and answers related to
 
 The model was fine-tuned on the vaccination dataset using the following hyperparameters:
 
-- **Training regime:** Mixed precision (fp16)
-- **Batch Size:** 32
-- **Learning Rate:** 2e-5
-- **Epochs:** 5
+- **Fine-Tuning Epochs:** 3
+- **Batch Size:** 1 (per device for training and evaluation)
+- **Learning Rate:** 2e-4
+- **Max Tokens per Response:** 512
+
 
 #### Preprocessing
 
 The data was cleaned and tokenized to ensure high-quality input for the model training process.
 
-#### Speeds, Sizes, Times
-
-- **Training Time:** Approximately 72 hours
-- **Checkpoint Size:** 8GB
-
 ## Evaluation
 
 ### Testing Data, Factors & Metrics
@@ -123,10 +119,10 @@ The evaluation considered various factors, including the accuracy and relevance
 
 #### Metrics
 
-- **Accuracy:** 92%
-- **Response Relevance:** 90%
-- **Average Latency:** 200ms
-- **Max Tokens per Response:** 150
+- **Loss:** 0.3554
+- **Runtime:** 195.8647 seconds
+- **Samples per Second:** 0.735
+
 
 ### Results
 
@@ -148,9 +144,6 @@ The Vax-Llama-1 is a transformer-based language model built on the Llama3 archit
 
 ### Compute Infrastructure
 
-#### Hardware
-
-- **GPUs:** NVIDIA A100
 
 #### Software