plamo-2-1b / README.md
mitmul's picture
Upload folder using huggingface_hub
3600941 verified
|
Raw
History Blame
2.62 kB
---
license: apache-2.0
language:
- en
- ja
pipeline_tag: text-generation
library_name: transformers
base_model: pfnet/plamo-2-1b
tags:
- mlx
---
# mlx-community/plamo-2-1b
The Model [mlx-community/plamo-2-1b](https://huggingface.co/mlx-community/plamo-2-1b) was
converted to MLX format from [pfnet/plamo-2-1b](https://huggingface.co/pfnet/plamo-2-1b)
using mlx-lm version **0.21.0**.
## Use with mlx
```bash
pip install mlx numba # numba is required for the new PLaMo tokenizer
pip install "git+https://github.com/mitmul/mlx-examples.git@mitmul/add-plamo2-1b-support#egg=mlx-lm&subdirectory=llms"
```
```bash
python -m mlx_lm.generate \
--model mlx-community/plamo-2-1b \
--prompt '็พŽๅ‘ณใ—ใ„ใ‚ซใƒฌใƒผใฎไฝœใ‚Šๆ–น ใฎใƒฌใ‚ทใƒ”ใ‚’็ดนไป‹ใ—ใพใ™ใ€‚' \
--max-tokens 1024 \
--extra-eos-token '<|plamo:bos|>' \
--ignore-chat-template \
--temp 0
Fetching 7 files: 100%|โ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆ| 7/7 [00:00<00:00, 116508.44it/s]
==========
## ็พŽๅ‘ณใ—ใ„ใ‚ซใƒฌใƒผใฎไฝœใ‚Šๆ–น
### ๆๆ–™
- ็މใญใŽ ๏ผ‘ๅ€‹
- ใซใ‚“ใ˜ใ‚“ ๏ผ‘ๆœฌ
- ใ˜ใ‚ƒใŒใ„ใ‚‚ ๏ผ‘ๅ€‹
- ่ฑšใฒใ่‚‰ ๏ผ‘๏ผ๏ผ๏ฝ‡
- ใซใ‚“ใซใ ๏ผ‘ใ‹ใ‘
- ใ—ใ‚‡ใ†ใŒ ๏ผ‘ใ‹ใ‘
- ใ‚ซใƒฌใƒผ็ฒ‰ ๅคงใ•ใ˜๏ผ‘
- ๅกฉ ๅฐใ•ใ˜๏ผ‘
- ๆฐด ๏ผ‘๏ผ๏ผ๏ฝƒ๏ฝƒ
- ใƒˆใƒžใƒˆ็ผถ ๏ผ‘็ผถ
- ๆฐด ๏ผ‘๏ผ๏ผ๏ฝƒ๏ฝƒ
- ใ‚ตใƒฉใƒ€ๆฒน ๅคงใ•ใ˜๏ผ‘
- ๅกฉ ๅฐ‘ใ€…
- ใ“ใ—ใ‚‡ใ† ๅฐ‘ใ€…
### ไฝœใ‚Šๆ–น
- ็މใญใŽใ€ใซใ‚“ใ˜ใ‚“ใ€ใ˜ใ‚ƒใŒใ„ใ‚‚ใ€ใซใ‚“ใซใใ€ใ—ใ‚‡ใ†ใŒใ‚’ใฟใ˜ใ‚“ๅˆ‡ใ‚Šใซใ—ใพใ™ใ€‚
- ใƒ•ใƒฉใ‚คใƒ‘ใƒณใซใ‚ตใƒฉใƒ€ๆฒนใ‚’็†ฑใ—ใ€ใซใ‚“ใซใใ€ใ—ใ‚‡ใ†ใŒใ‚’็‚’ใ‚ใพใ™ใ€‚
- ้ฆ™ใ‚ŠใŒๅ‡บใฆใใŸใ‚‰ใ€็މใญใŽใ€ใซใ‚“ใ˜ใ‚“ใ€ใ˜ใ‚ƒใŒใ„ใ‚‚ใ€่ฑšใฒใ่‚‰ใ€ๅกฉใ€ใ“ใ—ใ‚‡ใ†ใ€ๆฐดใ€ใƒˆใƒžใƒˆ็ผถใ€ๆฐดใ€ใ‚ซใƒฌใƒผ็ฒ‰ใ‚’ๅ…ฅใ‚Œใฆ็‚’ใ‚ใพใ™ใ€‚
- ้‡Ž่œใซ็ซใŒ้€šใฃใŸใ‚‰ใ€ใ‚ซใƒฌใƒผ็ฒ‰ใ‚’ๆบถใ‹ใ—ใพใ™ใ€‚
- ๆฐดใ€ๅกฉใ€ใ“ใ—ใ‚‡ใ†ใงๅ‘ณใ‚’่ชฟใˆใพใ™ใ€‚
- ๆœ€ๅพŒใซใƒˆใƒžใƒˆ็ผถใ‚’ๅŠ ใˆใฆใ€็…ฎ่พผใ‚“ใงๅฎŒๆˆใงใ™ใ€‚
## ใพใจใ‚
็พŽๅ‘ณใ—ใ„ใ‚ซใƒฌใƒผใฎไฝœใ‚Šๆ–นใฎใƒฌใ‚ทใƒ”ใ‚’็ดนไป‹ใ—ใพใ—ใŸใ€‚
ใœใฒๅ‚่€ƒใซใ—ใฆใฟใฆใใ ใ•ใ„ใ€‚
==========
Prompt: 8 tokens, 114.293 tokens-per-sec
Generation: 234 tokens, 67.785 tokens-per-sec
Peak memory: 2.727 GB
```
You can also write your code to use this model like this:
```python
from mlx_lm import load, generate
model, tokenizer = load("mlx-community/plamo-2-1b")
prompt = "็พŽๅ‘ณใ—ใ„ใ‚ซใƒฌใƒผใฎไฝœใ‚Šๆ–นใฎใƒฌใ‚ทใƒ”ใ‚’็ดนไป‹ใ—ใพใ™ใ€‚"
response = generate(model, tokenizer, prompt=prompt, verbose=True)
```