orign 0.1.25

A Python client for Orign

orign-py

A Python client for Orign

Installation

pip install orign

Install the Orign CLI

curl -fsSL -H "Cache-Control: no-cache" https://storage.googleapis.com/orign/releases/install.sh | bash

Login to Orign

$ orign login

Usage

Get a list of available models

$ orign get models

Chat

Define which model we would like to use

from orign import ChatModel

model = ChatModel(model="allenai/Molmo-7B-D-0924", provider="vllm")

Open a socket connection to the model

model.connect()

Chat with the model

model.chat(msg="What's in this image?", image="https://tinyurl.com/2fz6ms35")

Stream tokens from the model

for response in model.chat(msg="What is the capital of France?", stream_tokens=True):
    print(response)

Send a thread of messages to the model

model.chat(prompt=[
    {"role": "user", "content": "What is the capital of France?"},
    {"role": "assistant", "content": "Paris"},
    {"role": "user", "content": "When was it built?"}
])

Send a batch of threads to the model

model.chat(batch=[
    [{"role": "user", "content": "What is the capital of France?"}, {"role": "assistant", "content": "Paris"}, {"role": "user", "content": "When was it built?"}],
    [{"role": "user", "content": "What is the capital of Spain?"}, {"role": "assistant", "content": "Madrid"}, {"role": "user", "content": "When was it built?"}]
]):

Use the async API

from orign import AsyncChatModel

model = AsyncChatModel(model="allenai/Molmo-7B-D-0924", provider="vllm")
await model.connect()

async for response in model.chat(
    msg="What is the capital of france?", stream_tokens=True
):
    print(response)

Embeddings

Define which model we would like to use

from orign import EmbeddingModel

model = EmbeddingModel(provider="sentence-tf", model="clip-ViT-B-32")

Embed a text

model.embed(text="What is the capital of France?")

Embed an image

model.embed(image="https://example.com/image.jpg")

Embed text and image

model.embed(text="What is the capital of France?", image="https://example.com/image.jpg")

Use the async API

from orign import AsyncEmbeddingModel

model = AsyncEmbeddingModel(provider="sentence-tf", model="clip-ViT-B-32")
await model.connect()

await model.embed(text="What is the capital of France?")

OCR

Define which model we would like to use

from orign import OCRModel

model = OCRModel(provider="easyocr")

Detect text in an image

model.detect(image="https://example.com/image.jpg")

Use the async API

from orign import AsyncOCRModel

model = AsyncOCRModel(provider="doctr")
await model.connect()

await model.detect(image="https://example.com/image.jpg")

Replay Buffer

Replay buffers offer a means of training models in an online fashion.

from orign import ReplayBuffer, V1MSSwiftBufferParams

params = V1MSSwiftBufferParams(
    model="Qwen/Qwen2-VL-7B-Instruct",
    model_type="qwen2_vl",
    train_type="lora",
    deepspeed="zero3",
    torch_dtype="bfloat16",
    max_length=16384,
    val_split_ratio=0.95,
    num_train_epochs=3,
    eval_strategy="epoch",
    save_strategy="epoch",
    save_total_limit=3,
    lora_rank=64,
    lora_alpha=128,
    size_factor=28,
    max_pixels=1025000,
    freeze_vit=True,
)

buffer = ReplayBuffer(
    name="sql-adapter",
    vram_request="40Gi",
    dtype="bfloat16",
    train_every=50,
    sample_n=100,
    sample_strategy="Random",
    ms_swift_params=params,
)

Then send examples to the buffer

buffer.send(
    [
        {
            "messages": [
                {"role": "user", "content": "what's in this image <image>"},
                {"role": "assistant", "content": "Well its a penguin of course"},
            ],
            "images": [
                "https://cdn.britannica.com/77/81277-050-2A6A35B2/Adelie-penguin.jpg"
            ],
        },
    ]
)

Examples

See the examples directory for more usage examples.