Instructions to use basically-ai/Pebble-10M-Chat with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use basically-ai/Pebble-10M-Chat with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="basically-ai/Pebble-10M-Chat", trust_remote_code=True)

# Load model directly
from transformers import Pebble10MLM
model = Pebble10MLM.from_pretrained("basically-ai/Pebble-10M-Chat", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use basically-ai/Pebble-10M-Chat with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "basically-ai/Pebble-10M-Chat"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "basically-ai/Pebble-10M-Chat",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/basically-ai/Pebble-10M-Chat

SGLang

How to use basically-ai/Pebble-10M-Chat with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "basically-ai/Pebble-10M-Chat" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "basically-ai/Pebble-10M-Chat",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "basically-ai/Pebble-10M-Chat" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "basically-ai/Pebble-10M-Chat",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use basically-ai/Pebble-10M-Chat with Docker Model Runner:
```
docker model run hf.co/basically-ai/Pebble-10M-Chat
```

Holy sh, a small language model that is mamab!

by FlameF0X - opened 2 days ago

Discussion

FlameF0X

2 days ago

Peak. You rarely see non transformers model that are so small.

Hoglet-33

basically AI org 1 day ago

Yeah! I kinda got bored with all of the small models being pure transformers. So the mamba/transformer hybrid is what I've settled on for now.

Datdanboi25

1 day ago

Love the diversity!

Hoglet-33

basically AI org 1 day ago

Love the diversity!

Thanks! Once I get the other Pebbles released i'm going to do some experimenting with the transformer layers and try some new stuff out...

Datdanboi25

1 day ago

Love the diversity!

Thanks! Once I get the other Pebbles released i'm going to do some experimenting with the transformer layers and try some new stuff out...

Sounds exciting, hell yeah!

ContextReq

1 day ago

Love the diversity!

Thanks! Once I get the other Pebbles released i'm going to do some experimenting with the transformer layers and try some new stuff out...

If I recall correctly, Mamba is currently not known at scale?

Hoglet-33

basically AI org 1 day ago

I created a new org https://huggingface.co/basically-experimental and thats probably where i will publish all of the dumb ideas.

so expect to wait a few more days.

Datdanboi25

1 day ago

I created a new org https://huggingface.co/basically-experimental and thats probably where i will publish all of the dumb ideas.

Also here is how the training of Pebble 25M is going in case anyone cares: step 86180/190734 | loss 2.1485 | ppl 8.57 | grad 0.233 | lr 0.633x | 0.13M tkn/s | eta 29.7h

so expect to wait a few more days.

Sweet will chuck you a follow!

FlameF0X

1 day ago

Yeah! I kinda got bored with all of the small models being pure transformers. So the mamba/transformer hybrid is what I've settled on for now.

Epic twin. Epic.

FlameF0X

1 day ago

•

edited 1 day ago

Love the diversity!

Thanks! Once I get the other Pebbles released i'm going to do some experimenting with the transformer layers and try some new stuff out...

If I recall correctly, Mamba is currently not known at scale?

Mamba is relatively recent in comparation with RWKV or other linear attention models.

FlameF0X changed discussion status to closed 1 day ago

FlameF0X changed discussion status to open 1 day ago

FlameF0X

1 day ago

oh, wait, i misread the message, sorry

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment