student-abdullah/Llama3.2_trial_16-bit_gguf

Uploaded model

• Developed by: student-abdullah
• License: apache-2.0
• Finetuned from model: meta-llama/Llama-3.2-1B
• Created on: 3st October, 2024

---

Acknowledgement

---

Model Description

This model is fine-tuned from the meta-llama/Llama-3.2-1B base model to enhance its capabilities in generating relevant and accurate responses related to generic medications under the PMBJP scheme. The fine-tuning process included the following hyperparameters:

• Fine Tuning Template: Llama Q&A
• Max Tokens: 512
• LoRA Alpha: 4
• LoRA Rank (r): 128
• Learning rate: 5e-5
• Gradient Accumulation Steps: 2
• Batch Size: 4
• Quantization: 16 bits

---

Model Quantitative Performace

• Training Quantitative Loss: 0.1383 (at final 5th epoch 5150th Step)

---

Limitations

• Token Limitations: With a max token limit of 512, the model might not handle very long queries or contexts effectively.
• Training Data Limitations: The model’s performance is contingent on the quality and coverage of the fine-tuning dataset, which may affect its generalizability to different contexts or medications not covered in the dataset.
• Potential Biases: As with any model fine-tuned on specific data, there may be biases based on the dataset used for training.

---

Model Performace Evaluation:

• Evaluation on 1000 Questions based on dataset (to evaluate the finetuned knowledge base)
• At temperature 0.3
• Correct Responses: 75%
• Incorrect Responses: 25%