Update README.md
Browse files
README.md
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
---
|
| 2 |
-
license: other
|
| 3 |
-
license_name: proprietary
|
| 4 |
-
license_link: LICENSE
|
| 5 |
---
|
| 6 |
|
| 7 |
<!-- ================= Social Links ================= -->
|
|
@@ -25,124 +25,125 @@ license_link: LICENSE
|
|
| 25 |
|
| 26 |
---
|
| 27 |
|
| 28 |
-
# 🚀 JumpLander Coder 32B
|
| 29 |
-
**Advanced Code-Generation LLM — optimized for Persian
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
|
| 35 |
---
|
| 36 |
|
| 37 |
-
## 🌟 Overview
|
| 38 |
-
JumpLander Coder 32B is a high-performance, bilingual (English–Persian) code-generation LLM
|
| 39 |
-
JumpLander Coder 32B یک مدل تولید کُد با عملکرد بالا و دو زبانه (انگلیسی و فارسی) است که طراحی شده تا:
|
| 40 |
-
- High-quality, executable code generation ⚡ — تولید کدی با کیفیت بالا و قابل اجرا
|
| 41 |
-
- Multi-file reasoning & architecture-level analysis 🧩 — تحلیل سطح معماری و استدلال در چند فایل
|
| 42 |
-
- Security-aware and refactoring-optimized outputs 🛡️ — خروجیهایی با آگاهی امنیتی و بهینهشده برای بازسازی کد
|
| 43 |
-
- Enhanced usability for Persian-speaking developers 🇮🇷 — تجربهٔ بهتر برای توسعهدهندگان فارسیزبان
|
| 44 |
|
| 45 |
-
|
| 46 |
-
|
|
|
|
| 47 |
|
| 48 |
---
|
| 49 |
|
| 50 |
-
## 📊 Current Status
|
| 51 |
-
|
| 52 |
-
-
|
| 53 |
-
|
| 54 |
-
-
|
| 55 |
-
|
|
|
|
|
|
|
|
|
|
| 56 |
|
| 57 |
---
|
| 58 |
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
**
|
| 64 |
-
-
|
| 65 |
-
|
| 66 |
-
- When local usage is available, the flow will be:
|
| 67 |
-
1. Download model weights + checksum + signature
|
| 68 |
-
2. Verify integrity locally (SHA-256, PGP)
|
| 69 |
-
3. Load model via JumpLander App or CLI
|
| 70 |
-
4. Start coding locally — offline or online
|
| 71 |
-
|
| 72 |
-
**جریان کار زمانی که قابلیت لوکال آماده شود:**
|
| 73 |
-
1. دانلود وزنهای مدل + چکسام + امضای دیجیتال
|
| 74 |
-
2. بررسی صحت و یکپارچگی محلی (SHA-256 و PGP)
|
| 75 |
-
3. اجرا از ��ریق JumpLander App یا CLI
|
| 76 |
-
4. شروع به کدنویسی به صورت محلی — آفلاین یا آنلاین
|
| 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 |
-
## 🛠 Planned Capabilities
|
|
|
|
| 123 |
- Repository-wide code generation
|
| 124 |
- Multi-language support: Python, JS/TS, Go, Rust, Java, C/C++, Bash, SQL
|
| 125 |
-
-
|
| 126 |
-
- Test generation
|
| 127 |
-
-
|
| 128 |
-
-
|
| 129 |
-
:contentReference[oaicite:7]{index=7}
|
| 130 |
|
| 131 |
---
|
| 132 |
|
| 133 |
-
## 💻 Example Usage (Future API)
|
|
|
|
| 134 |
```python
|
| 135 |
from jumplander_sdk import JumplanderClient
|
| 136 |
|
| 137 |
client = JumplanderClient(api_key="YOUR_KEY")
|
| 138 |
|
| 139 |
-
# Scaffold a FastAPI app
|
| 140 |
project = client.scaffold(
|
| 141 |
-
"Create a FastAPI service with JWT
|
| 142 |
language="python"
|
| 143 |
)
|
| 144 |
project.save("./generated_app")
|
| 145 |
|
| 146 |
# Refactor an existing repository
|
| 147 |
-
patches = client.refactor("./myrepo", intent="improve
|
| 148 |
client.apply_patches(patches)
|
|
|
|
| 1 |
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: proprietary
|
| 4 |
+
license_link: LICENSE
|
| 5 |
---
|
| 6 |
|
| 7 |
<!-- ================= Social Links ================= -->
|
|
|
|
| 25 |
|
| 26 |
---
|
| 27 |
|
| 28 |
+
# 🚀 JumpLander Coder 32B
|
| 29 |
+
**Advanced Code-Generation LLM — optimized for English & Persian workflows**
|
| 30 |
+
🇮🇷 *چند خط توضیح فارسی:*
|
| 31 |
+
این مدل برای توسعهدهندگان ایرانی طراحی شده و نسخه آنلاین آن در سایت فعال است.
|
| 32 |
+
نسخهی لوکال فقط از طریق نرمافزار رسمی JumpLander ارائه خواهد شد.
|
| 33 |
+
وزنهای مدل عمومی نیستند و تنها در قالب نرمافزار قابل استفاده میباشند.
|
| 34 |
|
| 35 |
---
|
| 36 |
|
| 37 |
+
## 🌟 Overview
|
| 38 |
+
JumpLander Coder 32B is a high-performance, bilingual (English–Persian) code-generation LLM built for advanced programming tasks, repository-wide reasoning, and architecture-level understanding.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
|
| 40 |
+
This release provides documentation, benchmarks, design goals, and usage guidelines.
|
| 41 |
+
**Model weights are not publicly distributed.**
|
| 42 |
+
Local access will be provided exclusively through the official JumpLander desktop/server application.
|
| 43 |
|
| 44 |
---
|
| 45 |
|
| 46 |
+
## 📊 Current Status
|
| 47 |
+
|
| 48 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/69204763af796f2f22ad9f49/A8r0WUkLpEhDAh7Z8Xajx.jpeg" width="600"/>
|
| 49 |
+
|
| 50 |
+
- ✔ Documentation available
|
| 51 |
+
- ✔ Prototype benchmarks included
|
| 52 |
+
- ✔ Online demo available on the website
|
| 53 |
+
- ❌ Weights are *not* public
|
| 54 |
+
- 🔒 Local model execution will be provided only via official software
|
| 55 |
|
| 56 |
---
|
| 57 |
|
| 58 |
+
# 🧪 Benchmarks (Prototype)
|
| 59 |
+
|
| 60 |
+
| Task | Score | Notes |
|
| 61 |
+
|------|-------|--------|
|
| 62 |
+
| **HumanEval** | **72%** | Strong execution accuracy |
|
| 63 |
+
| **Repo-level Q&A** | High | Stable multi-file reasoning |
|
| 64 |
+
| **Persian Instruction Following** | **Excellent** | Optimized bilingual performance |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
|
| 66 |
---
|
| 67 |
|
| 68 |
+
# 📦 Model Comparison (Prototype Benchmarks)
|
| 69 |
+
|
| 70 |
+
| Model | Params | HumanEval | Multi-file Reasoning | Persian Support | Speed (tok/s) | Availability |
|
| 71 |
+
|------|--------|-----------|------------------------|------------------|----------------|--------------|
|
| 72 |
+
| **JumpLander Coder 32B** | 32B | **72%** | ✔ Strong | **Excellent** | 34 | Local-only via app |
|
| 73 |
+
| Qwen2.5-Coder 32B | 32B | 75% | Medium | Weak | 32 | Open-source |
|
| 74 |
+
| DeepSeek-Coder 33B | 33B | 79% | Strong | Weak | 29 | Open-source |
|
| 75 |
+
| StarCoder2 15B | 15B | 63% | Limited | Weak | **45** | Open-source |
|
| 76 |
+
| Llama-3.1 70B | 70B | **82%** | Strong | Weak | 20 | Open-source |
|
| 77 |
|
| 78 |
---
|
| 79 |
|
| 80 |
+
# 💡 Why JumpLander Coder 32B?
|
| 81 |
+
|
| 82 |
+
> ### 🧠 Multi-file reasoning
|
| 83 |
+
> Designed for architecture-level understanding and full-repository analysis.
|
| 84 |
+
|
| 85 |
+
> ### 🇮🇷 Persian-optimized workflow
|
| 86 |
+
> Tuned for real Persian programming scenarios and instruction patterns.
|
| 87 |
+
|
| 88 |
+
> ### 🛡️ Secure-by-design outputs
|
| 89 |
+
> Refactoring logic, patch suggestions, and safe coding guidelines included.
|
| 90 |
+
|
| 91 |
+
> ### ⚡ Developer-focused ecosystem
|
| 92 |
+
> Future SDK, CLI tools, and integrated analysis modules.
|
| 93 |
|
| 94 |
---
|
| 95 |
|
| 96 |
+
## 🗂 Local Execution (Official Software Only)
|
| 97 |
+
|
| 98 |
+
Local execution of the model will be provided through the **JumpLander App**, enabling:
|
| 99 |
+
|
| 100 |
+
- Secure local model loading
|
| 101 |
+
- Offline and online operation modes
|
| 102 |
+
- Integrated coding environment
|
| 103 |
+
- Automatic model updates
|
| 104 |
+
- Full repository understanding features
|
| 105 |
+
|
| 106 |
+
**Note:**
|
| 107 |
+
Weights will *not* be downloadable manually.
|
| 108 |
+
They are packaged, encrypted, and tied to the official software.
|
| 109 |
|
| 110 |
---
|
| 111 |
|
| 112 |
+
## 🎯 Use Cases
|
| 113 |
+
|
| 114 |
+
- Application scaffolding
|
| 115 |
+
- Repository-wide refactoring
|
| 116 |
+
- Debugging & architecture inspection
|
| 117 |
+
- Documentation and API specification
|
| 118 |
+
- Programming education (EN + FA)
|
| 119 |
|
| 120 |
---
|
| 121 |
|
| 122 |
+
## 🛠 Planned Capabilities
|
| 123 |
+
|
| 124 |
- Repository-wide code generation
|
| 125 |
- Multi-language support: Python, JS/TS, Go, Rust, Java, C/C++, Bash, SQL
|
| 126 |
+
- Long-context reasoning (hundreds of thousands of tokens)
|
| 127 |
+
- Test generation: unit, integration, regression
|
| 128 |
+
- IDE extensions (VS Code + JetBrains)
|
| 129 |
+
- Full SDK + CLI tools
|
|
|
|
| 130 |
|
| 131 |
---
|
| 132 |
|
| 133 |
+
## 💻 Example Usage (Future API)
|
| 134 |
+
|
| 135 |
```python
|
| 136 |
from jumplander_sdk import JumplanderClient
|
| 137 |
|
| 138 |
client = JumplanderClient(api_key="YOUR_KEY")
|
| 139 |
|
| 140 |
+
# Scaffold a FastAPI app
|
| 141 |
project = client.scaffold(
|
| 142 |
+
"Create a FastAPI service with JWT and PostgreSQL",
|
| 143 |
language="python"
|
| 144 |
)
|
| 145 |
project.save("./generated_app")
|
| 146 |
|
| 147 |
# Refactor an existing repository
|
| 148 |
+
patches = client.refactor("./myrepo", intent="improve structure")
|
| 149 |
client.apply_patches(patches)
|