---
license: gemma
base_model: unsloth/gemma-3-1b-it-unsloth-bnb-4bit
library_name: llama.cpp
pipeline_tag: text-generation
tags:
- gemma
- gemma-3
- gguf
- text-generation
- telegram
- trading-signals
- structured-output
- json
datasets:
- Sigrex/telegram-signal-extract
---

# Telegram Signal Extract — Gemma 3 1B

A fine-tuned **Gemma 3 1B Instruct** model for extracting a compact trading-signal payload from Telegram channel messages. It is intended to return either a payload such as:

```json
{"symbol":"ONEUSDT","side":"BUY"}
```

or the literal JSON value:

```json
null
```

The model was trained with supervised examples pairing Telegram messages with a signal payload or `null`. For best observed output, use **temperature 0** and the task system prompt below. Fine-tuning alone does not guarantee valid JSON or correct classifications; validate model output before using it.

## System prompt

```text
You extract trade signals from Telegram channel messages. For an actionable open/close signal, reply with ONLY a JSON object like {"symbol": "BTCUSDT", "side": "BUY"} where symbol is the uppercase trading pair and side is BUY or SELL. If the message is not an actionable signal, reply with exactly: null
```

## Files

| File | Quantization / precision | Approx. size |
|---|---:|---:|
| `gemma-3-1b-it.Q4_K_M.gguf` | Q4_K_M | 769 MiB |
| `gemma-3-1b-it.Q5_K_M.gguf` | Q5_K_M | 812 MiB |
| `gemma-3-1b-it.Q6_K.gguf` | Q6_K | 965 MiB |
| `gemma-3-1b-it.Q8_0.gguf` | Q8_0 | 1020 MiB |
| `gemma-3-1b-it.F16.gguf` | F16 | 1.87 GiB |
| `gemma-3-1b-it.BF16.gguf` | BF16 | 1.87 GiB |
| Other `IQ*` and `Q2`/`Q3` GGUFs | Smaller quantizations | See file sizes on the Files tab |

The repository is large because it includes **all exported variants**. For most local use, start with `Q4_K_M`; use a higher precision variant if you have enough memory and want to compare quality.

## Task and output

- **Input:** one raw Telegram message.
- **Positive output:** `{"symbol":"<UPPERCASE_PAIR>","side":"BUY"}` or `{"symbol":"<UPPERCASE_PAIR>","side":"SELL"}`.
- **Negative output:** `null`.
- The payload contains `symbol` and `side` only. It does not estimate trade size, profitability, or risk.

## Intended use and limitations

This is an experimental information-extraction model, not a trading system or investment recommendation. It can miss signals, misread tickers/direction, or return invalid output. **Do not automatically place orders from its responses.** Parse the output as JSON, reject anything outside the expected schema, and independently validate symbols and trading actions.

## Base model and license

This model is derived from `unsloth/gemma-3-1b-it-unsloth-bnb-4bit`, a 4-bit quantization of `google/gemma-3-1b-it`. Gemma is **not** Apache-2.0 or MIT: it is governed by the **Gemma Terms of Use** and the **Gemma Prohibited Use Policy**, so this model is marked `gemma`. If you distribute it, pass the Gemma Terms of Use along and ship the notice text file Google requires — *"Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms"*. See the [Gemma 3 1B model page](https://huggingface.co/google/gemma-3-1b-it).