Update README.md

README.md

@@ -1,3 +1,28 @@
+---
+language:
+- en
+license: cc-by-nc-4.0
+tags:
+- code
+- python
+- django
+- flask
+- fastapi
+- react
+- nextjs
+- typescript
+- sql
+- docker
+- qlora
+- fine-tuned
+- coding-assistant
+- web-development
+base_model: Qwen/Qwen2.5-Coder-7B-Instruct
+pipeline_tag: text-generation
+model_type: qwen2
+library_name: transformers
+---
+
 # ScriptSwiftAI v12 (Dikshan)
 
 > QLoRA fine-tuned full-stack code generation model by **Dikshan** (India).
@@ -22,19 +47,101 @@ https://creativecommons.org/licenses/by-nc/4.0/
 | **Method** | QLoRA (4-bit, r=64) |
 | **Hardware** | NVIDIA RTX 5070 Ti 16GB |
 | **Target Steps** | 40,000 |
-| **Dataset** | ~450-500k samples (17 datasets) |
+| **Dataset** | ~570k samples (17 datasets) |
 | **Max Length** | 1536 tokens |
 | **Developer** | Dikshan (India) |
 | **Training Started** | June 2026 |
 
 ---
 
+## How to Use
+
+Load the adapter on top of the base model:
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+from peft import PeftModel
+import torch
+
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.bfloat16
+)
+
+base = AutoModelForCausalLM.from_pretrained(
+    "Qwen/Qwen2.5-Coder-7B-Instruct",
+    quantization_config=bnb_config,
+    device_map="auto"
+)
+model = PeftModel.from_pretrained(base, "Dikshan1234/ScriptSwiftAIV2/checkpoint-14000")
+tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-Coder-7B-Instruct")
+
+prompt = """You are ScriptSwiftAI, a professional full-stack and Python web development assistant made by Dikshan.
+
+### Instruction:
+Build a Django REST API with JWT authentication.
+
+### Response:
+"""
+
+inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+output = model.generate(**inputs, max_new_tokens=1536, temperature=0.3, top_p=0.85, top_k=40)
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+
+> GGUF version coming soon for easy local use with Ollama.
+
+---
+
+## Dataset Sources (17 datasets, ~570k samples)
+
+| # | Dataset | Focus |
+|---|---------|-------|
+| 1 | sahil2801/CodeAlpaca-20k | General coding |
+| 2 | TokenBender/code_instructions_122k_alpaca_style | Python |
+| 3 | flytech/python-codes-25k | Python |
+| 4 | AhmedSSoliman/DJANGO | Django |
+| 5 | nickrosh/Evol-Instruct-Code-80k-v1 | Web dev |
+| 6 | iamtarun/python_code_instructions_18k_alpaca | Python |
+| 7 | ise-uiuc/Magicoder-OSS-Instruct-75K | General coding |
+| 8 | glaiveai/glaive-code-assistant-v3 | Web dev |
+| 9 | ajibawa-2023/Python-Code-23k-ShareGPT | Python |
+| 10 | m-a-p/CodeFeedback-Filtered-Instruction | Web dev |
+| 11 | cfahlgren1/react-code-instructions | React |
+| 12 | Tesslate/Next.js-Dataset | Next.js |
+| 13 | mhhmm/typescript-instruct-20k | TypeScript |
+| 14 | grenishrai/typescript-dataset | TypeScript |
+| 15 | NumbersStation/NSText2SQL | SQL |
+| 16 | nvidia/OpenCodeInstruct | General coding |
+| 17 | Identity QA | ScriptSwiftAI identity |
+
+---
+
+## Training Config
+
+| Setting | Value |
+|---|---|
+| **Base Model** | Qwen2.5-Coder-7B-Instruct |
+| **Method** | QLoRA (4-bit NF4) |
+| **LoRA Rank** | r=64, alpha=128 |
+| **Target Steps** | 40,000 |
+| **Batch Size** | 1 (effective 16 with grad accum) |
+| **Learning Rate** | 2e-4 cosine, 1000 warmup steps |
+| **Optimizer** | paged_adamw_8bit |
+| **Max Length** | 1536 tokens |
+| **Hardware** | RTX 5070 Ti 16GB |
+
+---
+
+---
+
 ## Changelog / Training Timeline
 
 *Auto-updated by train_v12.py on every checkpoint save.*
 *Append-only — entries never overwritten. HuggingFace commit history provides timestamps.*
 
-
 ---
 
 ### 🔵 Checkpoint — Step 13
@@ -352,4 +459,4 @@ https://creativecommons.org/licenses/by-nc/4.0/
 **Loss:** 0.2634
 **Epoch:** 0.409
 **Zone:** 📈 Good
-**Status:** Training ongoing — auto-uploaded via train_v12.py
+**Status:** Training ongoing — auto-uploaded via train_v12.py