37 lines
956 B
Markdown
37 lines
956 B
Markdown
---
|
|
datasets:
|
|
- IlyaGusev/ru_turbo_alpaca
|
|
- IlyaGusev/ru_turbo_saiga
|
|
- IlyaGusev/ru_sharegpt_cleaned
|
|
- IlyaGusev/oasst1_ru_main_branch
|
|
- IlyaGusev/ru_turbo_alpaca_evol_instruct
|
|
- lksy/ru_instruct_gpt4
|
|
language:
|
|
- ru
|
|
inference: false
|
|
pipeline_tag: conversational
|
|
license: llama2
|
|
---
|
|
|
|
Llama.cpp compatible versions of an original [7B model](https://huggingface.co/IlyaGusev/saiga2_7b_lora).
|
|
|
|
Download one of the versions, for example `model-q4_K.gguf`.
|
|
```
|
|
wget https://huggingface.co/IlyaGusev/saiga2_7b_gguf/resolve/main/model-q4_K.gguf
|
|
```
|
|
|
|
Download [interact_llamacpp.py](https://raw.githubusercontent.com/IlyaGusev/rulm/master/self_instruct/src/interact_llamacpp.py)
|
|
```
|
|
wget https://raw.githubusercontent.com/IlyaGusev/rulm/master/self_instruct/src/interact_llamacpp.py
|
|
```
|
|
|
|
How to run:
|
|
```
|
|
pip install llama-cpp-python fire
|
|
|
|
python3 interact_llamacpp.py model-q4_K.gguf
|
|
```
|
|
|
|
System requirements:
|
|
* 10GB RAM for q8_0 and less for smaller quantizations
|