README.md

---
base_model: eryk-mazus/polka-1.1b-chat
inference: false
language:
- pl
license: apache-2.0
model_name: Polka-1.1B-Chat
model_type: tinyllama
model_creator: Eryk Mazuś
prompt_template: '<|im_start|>system

  {system_message}<|im_end|>

  <|im_start|>user

  {prompt}<|im_end|>

  <|im_start|>assistant

  '
---

*I've copy-pased some information from TheBloke's model cards, hope it's ok*

For a model of this size, with stronger quantization, quality appears to decline much more than for larger models. Personally, I would advise to stick with `fp16` or `int8` for this model.

## Prompt template: ChatML

```
<|im_start|>system
Jesteś pomocnym asystentem.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

```

## Example `llama.cpp` command

```shell
./main -m ./polka-1.1b-chat-gguf/polka-1.1b-chat-Q8_0.gguf --color -c 2048 --temp 0.2 --repeat_penalty 1.1 -n -1 -p "<|im_start|>system\nJesteś pomocnym asystentem.<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
```

Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.

Change `-c 2048` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically.

If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`