---
license: apache-2.0
library_name: transformers
tags:
  - gpt2
  - solana
  - clawd
  - code
  - text-generation
language:
  - en
pipeline_tag: text-generation
---

# deepsol-clawd-code

Merged GPT-2 checkpoint from the Solana Clawd AI training stack (`deepsol-clawd-code-merged`).

## Model details

| | |
|---|---|
| Architecture | `GPT2LMHeadModel` |
| Layers | 12 |
| Hidden size | 768 |
| Heads | 12 |
| Context | 1024 |
| Vocab | 50257 (GPT-2 tokenizer) |
| Weights dtype | float16 (`model.safetensors`) |
| Size | ~237 MB |

## Files

- `model.safetensors` — merged weights
- `config.json` / `generation_config.json`
- `tokenizer.json` / `tokenizer_config.json`

## Quick start

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "ordlibrary/deepsol-clawd-code"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo)

prompt = "def transfer_sol("
inputs = tok(prompt, return_tensors="pt")
out = model.generate(**inputs, max_new_tokens=64)
print(tok.decode(out[0], skip_special_tokens=True))
```

## Intended use

Research / experimentation around Solana-oriented code and tooling assistants in the Clawd training pipeline. This is a small GPT-2-scale model, not a production 7B+ coder.

## Limitations

- Small capacity vs modern LLMs; expect weak long-context and complex reasoning.
- Training data and merge recipe are project-internal; evaluate before any production use.
- Do not rely on it for financial advice or unsigned transaction construction without human review.

## Citation

```bibtex
@misc{deepsol-clawd-code,
  title = {deepsol-clawd-code},
  author = {ordlibrary},
  year = {2026},
  howpublished = {\url{https://huggingface.co/ordlibrary/deepsol-clawd-code}}
}
```