iCIIT
/

redqueenprotocol-sin-llama3.2-3B-model

@@ -1,199 +1,123 @@
 ---
 library_name: transformers
-tags: []
 ---
-# Model Card for Model ID
-<!-- Provide a quick summary of what the model is/does. -->
-## Model Details
-### Model Description
-<!-- Provide a longer summary of what this model is. -->
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-### Model Sources [optional]
-<!-- Provide the basic links for the model. -->
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
-## Uses
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-### Direct Use
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-[More Information Needed]
-### Downstream Use [optional]
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-[More Information Needed]
-### Out-of-Scope Use
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-[More Information Needed]
-## Bias, Risks, and Limitations
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-[More Information Needed]
-### Recommendations
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-## How to Get Started with the Model
-Use the code below to get started with the model.
-[More Information Needed]
-## Training Details
-### Training Data
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-[More Information Needed]
-### Training Procedure
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-#### Preprocessing [optional]
-[More Information Needed]
-#### Training Hyperparameters
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-#### Speeds, Sizes, Times [optional]
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-[More Information Needed]
-## Evaluation
-<!-- This section describes the evaluation protocols and provides the results. -->
-### Testing Data, Factors & Metrics
-#### Testing Data
-<!-- This should link to a Dataset Card if possible. -->
-[More Information Needed]
-#### Factors
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-[More Information Needed]
-#### Metrics
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-[More Information Needed]
-### Results
-[More Information Needed]
-#### Summary
-## Model Examination [optional]
-<!-- Relevant interpretability work for the model goes here -->
-[More Information Needed]
-## Environmental Impact
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-## Technical Specifications [optional]
-### Model Architecture and Objective
-[More Information Needed]
-### Compute Infrastructure
-[More Information Needed]
-#### Hardware
-[More Information Needed]
-#### Software
-[More Information Needed]
-## Citation [optional]
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-**BibTeX:**
-[More Information Needed]
-**APA:**
-[More Information Needed]
-## Glossary [optional]
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-[More Information Needed]
-## More Information [optional]
-[More Information Needed]
-## Model Card Authors [optional]
-[More Information Needed]
-## Model Card Contact
-[More Information Needed]

 ---
+license: mit
+language:
+- si
 library_name: transformers
+tags:
+- llama-3
+- sinhala
+- generative-qa
+- iciit-2025
+- lora
+datasets:
+- RedQueenProtocol/all-articles-from-sinhala-wikipedia-2025-parquet
+- RedQueenProtocol/sinhala-qna-530-rows
+- ihalage/sinhala-finetune-qa-eli5
+- janani-rane/SiQuAD
+base_model:
+- meta-llama/Llama-3.2-3B-Instruct
 ---
+# RedQueen Llama 3.2 3B - Sinhala Generative QA
+**Technical Report:** [Click here for pdf](https://drive.google.com/file/d/1XFPwiwTx5j8yxcBCxmyDZgK5ldpulFw-/view?usp=sharing)
+<br>
+**GitHub Repo for Scripts and Notebooks:** [Click here](https://github.com/scythe410/Below-8B-Sinhala-LLM-Training---RedQueen-Protocol)
+- **Developed by:** [Red Queen Protocol](https://huggingface.co/RedQueenProtocol)
+- **Team:** [Ramiru De Silva](https://www.linkedin.com/in/ramirudesilva/), [Senadhi Thimanya](https://www.linkedin.com/in/senadhi-chandrasekara/)
+- **Language(s) (NLP):** Sinhala
+- **Finetuned from model:** [Llama 3.2 3B IT](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct)
+This model and LoRA was developed by Ramiru De Silva and Senadhi Thimanya (Team: [RedQueen Protocol](https://huggingface.co/RedQueenProtocol)) for the iCIIT Conclave 2025 Shared Task on Building Compact Sinhala & Tamil LLMs.
+This is a 3-billion parameter, instruction-tuned model that has undergone a novel two-stage fine-tuning process to achieve proficiency in both the Sinhala language and the specific task of generative QA. The entire fine-tuning process was performed efficiently using Low-Rank Adaptation (LoRA) technique.
+<br>
+The model's creation follows a hierarchical training strategy designed to first build a strong linguistic foundation and then specialize it for a specific task.
+### Stage 1: Domain Adaptation (Language Foundation)
+The initial model, `RedQueenProtocol/llama-3.2-3b-it-sinhala-rq` (Meta's Llama-3.2-3B-IT copies into a private repo for ease of use), was fine-tuned on the entirety of the Sinhala Wikipedia to create a foundational model with a comprehensive grasp of the language.
+- **Dataset:** `RedQueenProtocol/all-articles-from-sinhala-wikipedia-2025-parquet`.
+- **Method:** Long articles were tokenized and split into overlapping chunks of 512 tokens to ensure full context was seen during training.
+- **Output Model:** The resulting adapter was merged to create the Sinhala domain-expert base model for the next stage: `RedQueenProtocol/sinhala-wiki-2025-LoRA-merged`.
+### Stage 2: Task Adaptation (Sequential QA Fine-tuning)
+Using the Wikipedia-tuned model as the new base, a single LoRA adapter was sequentially fine-tuned on three distinct QA datasets to progressively accumulate question-answering skills.
+<br>
+The training sequence was as follows:
+1. **Custom Dataset:** Fine-tuned on a manually curated dataset of 528 Sinhala QA pairs (`RedQueenProtocol/sinhala-qna-530-rows`).
+2. **Ihalage ELI5 Dataset:** Continued training the same adapter on 10,000 samples from the `ihalage/sinhala-finetune-qa-eli5` dataset.
+3. **SiQuAD Dataset:** Performed a final round of training on 13,500 samples from the `janani-rane/SiQuAD` dataset, formatting the inputs as "Context: ... Question: ... Answer: ...".
+The **final LoRA adapter**, containing the combined knowledge of all three datasets **and the Wikipedia-tuned base model** was then uploaded here in seperate repositories.
+## How to Use
+```python
+# For Kaggle:
+#from kaggle_secrets import UserSecretsClient
+#from huggingface_hub import login
+#user_secrets = UserSecretsClient()
+#hf_token = user_secrets.get_secret("HF_TOKEN")
+#login(token=hf_token)
+# For Colab:
+#from huggingface_hub import notebook_login
+#notebook_login()
+# --- 1. Install Libraries ---
+!pip install -q -U transformers accelerate bitsandbytes peft
+# --- 2. Import Libraries ---
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+from peft import PeftModel
+import warnings
+# --- 3. Configuration ---
+# Now both the base model and adapter are loaded from the iCIIT organization.
+base_model_id = "iCIIT/sinhala-llama-rq-model"
+adapter_id = "iCIIT/sinhala-llama-rq-LoRA"
+device = "cuda" if torch.cuda.is_available() else "cpu"
+# --- 4. Load Model and Adapter ---
+print(f"Loading base model from: {base_model_id}")
+base_model = AutoModelForCausalLM.from_pretrained(
+    base_model_id,
+    torch_dtype=torch.bfloat16,
+    device_map=device,
+)
+tokenizer = AutoTokenizer.from_pretrained(base_model_id)
+tokenizer.pad_token = tokenizer.eos_token
+print(f"Applying LoRA adapter from: {adapter_id}")
+model = PeftModel.from_pretrained(base_model, adapter_id)
+print("\n Model and adapter loaded successfully from the iCIIT repositories.")
+# --- 5. Run a Sample Prompt ---
+generator = pipeline("text-generation", model=model, tokenizer=tokenizer)
+question = "ශ්‍රී ලංකා ජාතික ධජය නිර්මාණය කළේ කවුද?"
+prompt = f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\\n\\n{question}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\\n\\n"
+print("\n" + "="*50)
+print(f"USER: {question}")
+print("\nASSISTANT: Generating...")
+outputs = generator(
+    prompt,
+    max_new_tokens=256,
+    eos_token_id=tokenizer.eos_token_id,
+    do_sample=True,
+    temperature=0.6,
+    top_p=0.9,
+)
+full_response = outputs[0]['generated_text']
+answer = full_response.split("<|start_header_id|>assistant<|end_header_id|>\\n\\n")[1].replace("<|eot_id|>", "")
+print(answer.strip())
+print("="*50)