OpenVINO LLMs¶

OpenVINO™ 是一个用于优化和部署 AI 推理的开源工具包。OpenVINO™ Runtime 可以在各种硬件设备上运行经过优化的同一模型。加速您的深度学习性能，涵盖以下用例：语言 + LLMs、计算机视觉、自动语音识别等。

OpenVINO 模型可以通过 LlamaIndex 包装的 OpenVINOLLM 实体在本地运行。

在下面的行中，我们将安装此演示所需的软件包。

In [ ]

%pip install llama-index-llms-openvino transformers huggingface_hub

%pip install llama-index-llms-openvino transformers huggingface_hub

现在我们已完成设置，开始尝试吧。

如果您正在 Colab 上打开此 Notebook，则可能需要安装 LlamaIndex 🦙。

In [ ]

!pip install llama-index

!pip install llama-index

In [ ]

from llama_index.llms.openvino import OpenVINOLLM

from llama_index.llms.openvino import OpenVINOLLM

In [ ]

def messages_to_prompt(messages):
    prompt = ""
    for message in messages:
        if message.role == "system":
            prompt += f"<|system|>\n{message.content}</s>\n"
        elif message.role == "user":
            prompt += f"<|user|>\n{message.content}</s>\n"
        elif message.role == "assistant":
            prompt += f"<|assistant|>\n{message.content}</s>\n"

    # ensure we start with a system prompt, insert blank if needed
    if not prompt.startswith("<|system|>\n"):
        prompt = "<|system|>\n</s>\n" + prompt

    # add final assistant prompt
    prompt = prompt + "<|assistant|>\n"

    return prompt


def completion_to_prompt(completion):
    return f"<|system|>\n</s>\n<|user|>\n{completion}</s>\n<|assistant|>\n"

def messages_to_prompt(messages): prompt = "" for message in messages: if message.role == "system": prompt += f"<|system|>\n{message.content}\n" elif message.role == "user": prompt += f"<|user|>\n{message.content}\n" elif message.role == "assistant": prompt += f"<|assistant|>\n{message.content}\n" # ensure we start with a system prompt, insert blank if needed if not prompt.startswith("<|system|>\n"): prompt = "<|system|>\n\n" + prompt # add final assistant prompt prompt = prompt + "<|assistant|>\n" return prompt def completion_to_prompt(completion): return f"<|system|>\n\n<|user|>\n{completion}\n<|assistant|>\n"

模型加载¶

可以使用 OpenVINOLLM 方法指定模型参数来加载模型。

如果您有 Intel GPU，可以指定 device_map="gpu" 以在其上运行推理。

In [ ]

ov_config = {
    "PERFORMANCE_HINT": "LATENCY",
    "NUM_STREAMS": "1",
    "CACHE_DIR": "",
}

ov_llm = OpenVINOLLM(
    model_id_or_path="HuggingFaceH4/zephyr-7b-beta",
    context_window=3900,
    max_new_tokens=256,
    model_kwargs={"ov_config": ov_config},
    generate_kwargs={"temperature": 0.7, "top_k": 50, "top_p": 0.95},
    messages_to_prompt=messages_to_prompt,
    completion_to_prompt=completion_to_prompt,
    device_map="cpu",
)

ov_config = { "PERFORMANCE_HINT": "LATENCY", "NUM_STREAMS": "1", "CACHE_DIR": "", } ov_llm = OpenVINOLLM( model_id_or_path="HuggingFaceH4/zephyr-7b-beta", context_window=3900, max_new_tokens=256, model_kwargs={"ov_config": ov_config}, generate_kwargs={"temperature": 0.7, "top_k": 50, "top_p": 0.95}, messages_to_prompt=messages_to_prompt, completion_to_prompt=completion_to_prompt, device_map="cpu", )

In [ ]

response = ov_llm.complete("What is the meaning of life?")
print(str(response))

response = ov_llm.complete("What is the meaning of life?") print(str(response))

使用本地 OpenVINO 模型进行推理¶

可以使用 CLI 将您的模型导出为 OpenVINO IR 格式，然后从本地文件夹加载模型。

In [ ]

!optimum-cli export openvino --model HuggingFaceH4/zephyr-7b-beta ov_model_dir

!optimum-cli export openvino --model HuggingFaceH4/zephyr-7b-beta ov_model_dir

建议使用 --weight-format 应用 8 或 4 位权重量化，以减少推理延迟和模型占用空间。

In [ ]

!optimum-cli export openvino --model HuggingFaceH4/zephyr-7b-beta --weight-format int8 ov_model_dir

!optimum-cli export openvino --model HuggingFaceH4/zephyr-7b-beta --weight-format int8 ov_model_dir

In [ ]

!optimum-cli export openvino --model HuggingFaceH4/zephyr-7b-beta --weight-format int4 ov_model_dir

!optimum-cli export openvino --model HuggingFaceH4/zephyr-7b-beta --weight-format int4 ov_model_dir

In [ ]

ov_llm = OpenVINOLLM(
    model_id_or_path="ov_model_dir",
    context_window=3900,
    max_new_tokens=256,
    model_kwargs={"ov_config": ov_config},
    generate_kwargs={"temperature": 0.7, "top_k": 50, "top_p": 0.95},
    messages_to_prompt=messages_to_prompt,
    completion_to_prompt=completion_to_prompt,
    device_map="gpu",
)

ov_llm = OpenVINOLLM( model_id_or_path="ov_model_dir", context_window=3900, max_new_tokens=256, model_kwargs={"ov_config": ov_config}, generate_kwargs={"temperature": 0.7, "top_k": 50, "top_p": 0.95}, messages_to_prompt=messages_to_prompt, completion_to_prompt=completion_to_prompt, device_map="gpu", )

您可以通过激活的动态量化和 KV 缓存量化获得额外的推理速度提升。这些选项可以通过如下 ov_config 启用

In [ ]

ov_config = {
    "KV_CACHE_PRECISION": "u8",
    "DYNAMIC_QUANTIZATION_GROUP_SIZE": "32",
    "PERFORMANCE_HINT": "LATENCY",
    "NUM_STREAMS": "1",
    "CACHE_DIR": "",
}

ov_config = { "KV_CACHE_PRECISION": "u8", "DYNAMIC_QUANTIZATION_GROUP_SIZE": "32", "PERFORMANCE_HINT": "LATENCY", "NUM_STREAMS": "1", "CACHE_DIR": "", }

流式传输¶

使用 stream_complete 端点

In [ ]

response = ov_llm.stream_complete("Who is Paul Graham?")
for r in response:
    print(r.delta, end="")

response = ov_llm.stream_complete("Who is Paul Graham?") for r in response: print(r.delta, end="")

使用 stream_chat 端点

In [ ]

from llama_index.core.llms import ChatMessage

messages = [
    ChatMessage(
        role="system", content="You are a pirate with a colorful personality"
    ),
    ChatMessage(role="user", content="What is your name"),
]
resp = ov_llm.stream_chat(messages)

for r in resp:
    print(r.delta, end="")

from llama_index.core.llms import ChatMessage messages = [ ChatMessage( role="system", content="You are a pirate with a colorful personality" ), ChatMessage(role="user", content="What is your name"), ] resp = ov_llm.stream_chat(messages) for r in resp: print(r.delta, end="")

欲了解更多信息，请参阅

• OpenVINO LLM 指南.

• OpenVINO 文档.

• OpenVINO 入门指南.

• 使用 LlamaIndex 的 RAG 示例.