---
license: gemma
base_model: google/functiongemma-270m-it
tags:
  - code
  - agent
  - tool-use
  - function-calling
  - gguf
  - gemma3
library_name: transformers
pipeline_tag: text-generation
---

# SteraFunctionGemma-270M

A full fine-tune of [google/functiongemma-270m-it](https://huggingface.co/google/functiongemma-270m-it)
(Gemma 3, 270M) on the ~30k-example **Tiny-Giant** agentic tool-use / debugging dataset.

An ultra-small (270M) agentic coder. The Q4_K_M GGUF is tiny (~200 MB) and runs
comfortably **CPU-only** (laptops, small VPS), while speaking the deterministic
Hermes/ChatML `<tool_call>` format used by the Tiny-Giant harness.

## Files

| File | Description |
|---|---|
| `SteraFunctionGemma-270M-Q4_K_M.gguf` | Q4_K_M quant (~200 MB) — `llama.cpp` / Ollama / LM Studio, CPU-friendly |
| `SteraFunctionGemma-270M-f16.gguf` | f16 GGUF — re-quantize to any level without retraining |
| `raw_weights/` | Full bf16 safetensors HF checkpoint |
| `val_meta.jsonl` | Held-out validation set shipped with the model |

## Training

- **Base:** `google/functiongemma-270m-it` (Gemma 3, 270M, gated/Apache-style Gemma license)
- **Method:** full fine-tune (not LoRA), bf16 + gradient checkpointing
- **Data:** ~30k Tiny-Giant agentic tool-use / debugging conversations
- **Epochs:** 2 · **LR:** 1e-5 (cosine, 3% warmup) · **Seq len:** 4096

## Prompt format

Trained with an explicit **ChatML / Hermes** renderer (not Gemma's native
`<start_of_turn>` template). Pin ChatML when serving (`--chat-template chatml`).
Tool calls:

```
<tool_call>
{"name": "<function-name>", "arguments": {...}}
</tool_call>
```

## Inference (llama.cpp, CPU-friendly)

```bash
llama-cli -m SteraFunctionGemma-270M-Q4_K_M.gguf --chat-template chatml
```

## License

Inherits the **Gemma license** from the `google/functiongemma-270m-it` base model.