| | --- |
| | language: [en] |
| | license: mit |
| | tags: |
| | - software-engineering |
| | - programming |
| | - algorithms |
| | - system-design |
| | - slm |
| | - llama-style |
| | - rope |
| | - 1m-context |
| | - from-scratch |
| | - 1b-params |
| | pipeline_tag: text-generation |
| | --- |
| | |
| | # Software Engineer-SLM: Role-Based Small Language Model |
| |
|
| | A **LLaMA-style transformer** (~989.9M params, ~0.99B) trained from scratch for the **Software Engineer** role. |
| | Supports up to **1M token context** via RoPE with gradient checkpointing. |
| |
|
| | ## Architecture |
| | | Component | Value | |
| | |-----------|-------| |
| | | Architecture | LLaMA-style (RoPE + RMSNorm + SwiGLU) | |
| | | Parameters | ~989.9M (~0.99B) | |
| | | Layers | 32 | |
| | | Heads | 20 | |
| | | Embedding | 1600 | |
| | | Max Context | 100,000,000,000 tokens | |
| | | Max Output | 1,000,000 tokens | |
| | | Vocab | 2,180 BPE | |
| | | Model Size | ~4 GB (fp32) | |
| |
|
| | ## Training |
| | - Best eval loss: 0.301249697804451 |
| | - Trained with gradient checkpointing on Apple M4 (MPS) |
| | - 5 epochs, batch_size=1, grad_accum=16 |
| |
|
| | ## Usage |
| | ```python |
| | from huggingface_hub import hf_hub_download |
| | from tokenizers import Tokenizer |
| | |
| | model_path = hf_hub_download("sathishphdai/software-engineer-slm-1m", "model.safetensors") |
| | tokenizer_path = hf_hub_download("sathishphdai/software-engineer-slm-1m", "software_engineer_tokenizer.json") |
| | tokenizer = Tokenizer.from_file(tokenizer_path) |
| | ``` |
| |
|
