File size: 4,784 Bytes
13daa8c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 | # ScriptSwiftAI (Dikshan)
> A QLoRA fine-tuned code generation model built and owned by **Dikshan** (India).
> Specializes in generating clean, production-ready code across multiple languages and frameworks.
---
## License
This model is licensed under **Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)**.
- ✅ You may use, modify, and distribute this model for **personal and research purposes**
- ✅ You must **credit Dikshan / ScriptSwiftAI** with a link to this repository
- ✅ You must **state any changes** you made
- ❌ You may **NOT** use this model or any derivative for **commercial purposes**
- ❌ You may **NOT** sell, sublicense, or monetize any product built on this model
Full license: https://creativecommons.org/licenses/by-nc/4.0/
**Copyright © 2026 Dikshan. All rights reserved.**
---
## Model Details
| Property | Value |
|---|---|
| **Model Name** | ScriptSwiftAI (Dikshan) |
| **Base Model** | Qwen2.5-Coder-7B-Instruct |
| **Fine-tuning Method** | QLoRA (4-bit quantization) |
| **Training Hardware** | NVIDIA RTX 5070 Ti |
| **Framework** | Hugging Face Transformers + PEFT |
| **Target Steps** | ~37,500 (~2.21 epochs) |
| **Dataset Size** | ~271,689 samples (11 datasets + 1,900 identity QA pairs) |
| **Developer** | Dikshan (India) |
| **Training Started** | June 13, 2026 |
---
## What is ScriptSwiftAI?
ScriptSwiftAI is a fine-tuned AI model trained from scratch by Dikshan to generate fast, accurate, and production-ready code. It is trained on a custom curated dataset covering:
- Python, JavaScript, TypeScript, React, Node.js
- API development, full-stack web apps
- Code explanation, debugging, and refactoring
- Custom identity: responds as "ScriptSwiftAI" / "Dikshan" (not as base model)
---
## Training Configuration
- **LoRA Rank:** 64, Alpha: 128
- **Quantization:** 4-bit (NF4) via bitsandbytes
- **Optimizer:** paged_adamw_8bit
- **Batch Size:** 2 (gradient accumulation: 8, effective batch: 16)
- **Learning Rate:** 2e-4 with cosine scheduler
- **Max Sequence Length:** 2048 tokens
- **Warmup Steps:** 100
- **Checkpointing:** Every 500 steps with UPS auto-resume support
---
## Changelog / Training Timeline
This section tracks every major checkpoint upload as proof of development history.
All uploads are additionally timestamped via HuggingFace commit history.
---
### 🟢 v0.1 — Training Kickoff
**Date:** June 13, 2026
**Status:** Training started
- Base model loaded: Qwen2.5-Coder-7B-Instruct (Apache 2.0)
- QLoRA configuration finalized
- Dataset pipeline built: ~271,689 samples across 11 datasets
- Identity QA pairs injected: 1,900 samples (ScriptSwiftAI / Dikshan persona)
- Training infrastructure set up: train_v11_monitor.py with live loss monitoring UI
---
### 🟡 v0.2 — Best Loss Checkpoint
**Date:** June 17, 2026
**Checkpoint:** checkpoint-12000
**Step:** ~12,000 / 37,500
**Loss:** 0.1920 (best recorded)
**Status:** Training ongoing
- Best performing checkpoint saved
- Model showing strong code generation capability
- Identity responses stable (responds as ScriptSwiftAI/Dikshan correctly)
- Repetition penalty and top_p inference parameters tuned
---
### 🔵 v0.3 — ~77% Training Complete *(Latest)*
**Date:** June 20, 2026
**Checkpoint:** checkpoint-29000 *(update with actual checkpoint number)*
**Step:** ~29000+ / 37,500
**Loss:** ~0.27–0.32 (normal QLoRA batch variance)
**Status:** Training ongoing
- First public upload to HuggingFace
- README and LICENSE added
- Commercial product stack in development (Desktop + Web app)
---
*This changelog will be updated with every checkpoint upload going forward.*
---
## Intended Use
| Use Case | Allowed? |
|---|---|
| Personal coding assistant | ✅ Yes |
| Research and academic use | ✅ Yes |
| Open source projects | ✅ Yes (with credit) |
| Commercial SaaS / products | ❌ No |
| Reselling or sublicensing | ❌ No |
---
## Credit / Attribution
If you use ScriptSwiftAI in any project, you must include:
> *"Powered by ScriptSwiftAI, developed by Dikshan — https://huggingface.co/Dikshan1234/ScriptSwiftAI"*
---
## Contact
For licensing inquiries or commercial use permissions, contact the developer directly via HuggingFace.
---
*ScriptSwiftAI is an independent AI project developed entirely by Dikshan (India).
---
### 🔵 Checkpoint — Step 29,513
**Date:** June 20, 2026 at 23:50 IST
**Checkpoint:** checkpoint-29513
**Step:** 29,513 / 37,500 (78.7% complete)
**Loss:** 0.2475
**Epoch:** 1.738
**Zone:** 🎯 Sweet Spot
**Status:** Training ongoing — auto-uploaded via train_v11_monitor.py
|