Instructions to use yujiepan/mamba-tiny-random with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use yujiepan/mamba-tiny-random with Transformers:

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

pipe = pipeline("text-generation", model="yujiepan/mamba-tiny-random")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("yujiepan/mamba-tiny-random")
model = AutoModelForCausalLM.from_pretrained("yujiepan/mamba-tiny-random")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use yujiepan/mamba-tiny-random with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/yujiepan/mamba-tiny-random

SGLang

How to use yujiepan/mamba-tiny-random 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 "yujiepan/mamba-tiny-random" \
    --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": "yujiepan/mamba-tiny-random",
		"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 "yujiepan/mamba-tiny-random" \
        --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": "yujiepan/mamba-tiny-random",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use yujiepan/mamba-tiny-random with Docker Model Runner:
```
docker model run hf.co/yujiepan/mamba-tiny-random
```

mamba-tiny-random / README.md

yujiepan

Upload folder using huggingface_hub

6891e83 verified over 1 year ago

preview code

raw

history blame contribute delete

1.83 kB

metadata

pipeline_tag: text-generation
inference: true
widget:
  - text: Hello!
    example_title: Hello world
    group: Python
library_name: transformers

This model is randomly initialized, using the config from state-spaces/mamba-2.8b-hf but with smaller size.

Codes:

import os

import torch

import transformers
from huggingface_hub import create_repo, upload_folder

source_model_id = 'state-spaces/mamba-2.8b-hf'
tiny_random_name = 'mamba-tiny-random'
save_path = f'/tmp/yujiepan/{tiny_random_name}'
repo_id = f'yujiepan/{tiny_random_name}'

config = transformers.AutoConfig.from_pretrained(
    source_model_id, trust_remote_code=True)
config.hidden_size = 8
config.expand = 4
config.intermediate_size = 32
config.state_size = 8
config.num_hidden_layers = 2
config.n_layer = 2
config.torch_dtype = torch.bfloat16

model = transformers.AutoModelForCausalLM.from_config(
    config, torch_dtype=torch.bfloat16,
    trust_remote_code=True,
)
model.generation_config = transformers.GenerationConfig.from_pretrained(
    source_model_id,
    trust_remote_code=True,
)

transformers.set_seed(42)
with torch.no_grad():
    for name, p in sorted(model.named_parameters()):
        print(name, p.shape)
        torch.nn.init.uniform_(p, -0.5, 0.5)

model.save_pretrained(save_path)
tokenizer = transformers.AutoTokenizer.from_pretrained(
    source_model_id, trust_remote_code=True)

result = transformers.pipelines.pipeline(
    'text-generation',
    model=model, tokenizer=tokenizer,
    device='cuda',
    max_new_tokens=16,
)('Hello')
print(result)

model.save_pretrained(save_path)
tokenizer.save_pretrained(save_path)

os.system(f'ls -alh {save_path}')
create_repo(repo_id, exist_ok=True)
upload_folder(repo_id=repo_id, folder_path=save_path)