|
|
--- |
|
|
library_name: transformers |
|
|
pipeline_tag: text-generation |
|
|
--- |
|
|
|
|
|
|
|
|
# Wasm-Coder-8B-Instruct-V1 |
|
|
|
|
|
**Wasm-Coder-8B-Instruct-V1** is an 8-billion parameter instruction-tuned language model developed by [wasmdashai](https://huggingface.co/wasmdashai), , code generation, and technical reasoning. It is designed to help developers working on edge computing, browser-based runtimes, and low-level systems programming. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 Introduction |
|
|
|
|
|
`Wasm-Coder-8B-Instruct-V1` is part of the Wasm-Coder family—models specifically tailored for tasks involving WebAssembly, Rust, C/C++, and embedded systems programming. The model has been instruction-tuned on a diverse dataset combining code, documentation, compiler logs, and structured code reasoning tasks. |
|
|
|
|
|
### Key Features: |
|
|
|
|
|
* Strong performance in **code synthesis**, **bug fixing**, and **code explanation**, especially for Rust and WebAssembly projects. |
|
|
* Efficient for **edge devices**, **browsers**, and **serverless runtimes**. |
|
|
* Based on a powerful transformer architecture with performance enhancements such as RoPE and SwiGLU. |
|
|
* Trained with instruction-following datasets for natural conversations and multi-turn reasoning. |
|
|
* Supports **long-context processing** (up to 32,768 tokens) with optional rope scaling. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧠 Model Details |
|
|
|
|
|
* **Architecture**: Decoder-only transformer |
|
|
* **Parameters**: 8B |
|
|
* **Training**: Pretrained + Instruction fine-tuning |
|
|
* **Supported Context Length**: 32,768 tokens |
|
|
* **Specialization**: WebAssembly, Rust, C/C++, Systems Programming |
|
|
* **Components**: |
|
|
|
|
|
* RoPE (Rotary Positional Embeddings) |
|
|
* SwiGLU activation |
|
|
* RMSNorm |
|
|
* QKV Attention Bias |
|
|
|
|
|
--- |
|
|
|
|
|
## 💻 Quickstart |
|
|
|
|
|
Install dependencies: |
|
|
|
|
|
```bash |
|
|
pip install --upgrade transformers |
|
|
``` |
|
|
|
|
|
Example code to load and run the model: |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
|
|
model_name = "wasmdashai/Wasm-Coder-8B-Instruct-V1" |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name) |
|
|
model = AutoModelForCausalLM.from_pretrained( |
|
|
model_name, |
|
|
torch_dtype="auto", |
|
|
device_map="auto" |
|
|
) |
|
|
|
|
|
prompt = "Write a Rust function that compiles to WebAssembly and adds two numbers." |
|
|
|
|
|
inputs = tokenizer(prompt, return_tensors="pt").to(model.device) |
|
|
outputs = model.generate(**inputs, max_new_tokens=512) |
|
|
result = tokenizer.decode(outputs[0], skip_special_tokens=True) |
|
|
|
|
|
print(result) |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## 📚 Long-Context Support |
|
|
|
|
|
To process long inputs (e.g., full source files or compiler traces), use **YaRN-based RoPE scaling**: |
|
|
|
|
|
Add this to `config.json`: |
|
|
|
|
|
```json |
|
|
{ |
|
|
"rope_scaling": { |
|
|
"type": "yarn", |
|
|
"factor": 4.0, |
|
|
"original_max_position_embeddings": 32768 |
|
|
} |
|
|
} |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔧 Use Cases |
|
|
|
|
|
* WebAssembly code generation and debugging |
|
|
* Rust/C++ code explanation and transformation |
|
|
* Embedded/IoT code support |
|
|
* Smart contract logic for blockchain environments using Wasm |
|
|
* Code agents and assistants running in browsers |
|
|
|
|
|
--- |
|
|
|
|
|
## 📬 Contact |
|
|
|
|
|
📧 For questions, collaborations, or commercial licensing: |
|
|
|
|
|
**[[email protected]](mailto:[email protected])** |
|
|
|
|
|
--- |
|
|
|
|
|
|
