semantic-kernel调用huggingface本地模型

引言

semantic-kernel主要支持OpenAI和Azure OpenAI，由于国内原因无法使用。想调用本地模型发现资料很少，遂有下文：
示例不全，想快速上手推荐直接看仓库代码
官方文档：https://learn.microsoft.com/zh-cn/semantic-kernel/get-started/quick-start-guide?pivots=programming-language-python
下文来自源码库中示例：07-hugging-face-for-plugins.ipynb
官方仓库：https://github.com/microsoft/semantic-kernel.git

Using Hugging Face With Plugins

In this notebook, we demonstrate using Hugging Face models for Plugins using both SemanticMemory and text completions.

SK supports downloading models from the Hugging Face that can perform the following tasks: text-generation, text2text-generation, summarization, and sentence-similarity. You can search for models by task at https://huggingface.co/models.

# Note: if using a Poetry virtual environment, do not run this cell
%pip install semantic-kernel[hugging_face]==1.4.0

from services import Service

# Select a service to use for this notebook (available services: OpenAI, AzureOpenAI, HuggingFace)
selectedService = Service.HuggingFace
print(f"Using service type: {selectedService}")

First, we will create a kernel and add both text completion and embedding services.

For text completion, we are choosing GPT2. This is a text-generation model. (Note: text-generation will repeat the input in the output, text2text-generation will not.)
For embeddings, we are using sentence-transformers/all-MiniLM-L6-v2. Vectors generated for this model are of length 384 (compared to a length of 1536 from OpenAI ADA).

The following step may take a few minutes when run for the first time as the models will be downloaded to your local machine.

from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.hugging_face import HuggingFaceTextCompletion, HuggingFaceTextEmbedding
from semantic_kernel.core_plugins import TextMemoryPlugin
from semantic_kernel.memory import SemanticTextMemory, VolatileMemoryStore

kernel = Kernel()

# Configure LLM service
if selectedService == Service.HuggingFace:
    # Feel free to update this model to any other model available on Hugging Face
    text_service_id = "HuggingFaceM4/tiny-random-LlamaForCausalLM"
    kernel.add_service(
        service=HuggingFaceTextCompletion(
            service_id=text_service_id, ai_model_id=text_service_id, task="text-generation"
        ),
    )
    embed_service_id = "sentence-transformers/all-MiniLM-L6-v2"
    embedding_svc = HuggingFaceTextEmbedding(service_id=embed_service_id, ai_model_id=embed_service_id)
    kernel.add_service(
        service=embedding_svc,
    )
    memory = SemanticTextMemory(storage=VolatileMemoryStore(), embeddings_generator=embedding_svc)
    kernel.add_plugin(TextMemoryPlugin(memory), "TextMemoryPlugin")

Add Memories and Define a plugin to use them

Most models available on huggingface.co are not as powerful as OpenAI GPT-3+. Your plugins will likely need to be simpler to accommodate this.

from semantic_kernel.connectors.ai.hugging_face import HuggingFacePromptExecutionSettings
from semantic_kernel.prompt_template import PromptTemplateConfig

collection_id = "generic"

await memory.save_information(collection=collection_id, id="info1", text="Sharks are fish.")
await memory.save_information(collection=collection_id, id="info2", text="Whales are mammals.")
await memory.save_information(collection=collection_id, id="info3", text="Penguins are birds.")
await memory.save_information(collection=collection_id, id="info4", text="Dolphins are mammals.")
await memory.save_information(collection=collection_id, id="info5", text="Flies are insects.")

# Define prompt function using SK prompt template language
my_prompt = """I know these animal facts: 
- {{recall 'fact about sharks'}}
- {{recall 'fact about whales'}} 
- {{recall 'fact about penguins'}} 
- {{recall 'fact about dolphins'}} 
- {{recall 'fact about flies'}}
Now, tell me something about: {{$request}}"""

execution_settings = HuggingFacePromptExecutionSettings(
    service_id=text_service_id,
    ai_model_id=text_service_id,
    max_tokens=45,
    temperature=0.5,
    top_p=0.5,
)

prompt_template_config = PromptTemplateConfig(
    template=my_prompt,
    name="text_complete",
    template_format="semantic-kernel",
    execution_settings=execution_settings,
)

my_function = kernel.add_function(
    function_name="text_complete",
    plugin_name="TextCompletionPlugin",
    prompt_template_config=prompt_template_config,
)

Let’s now see what the completion looks like! Remember, “gpt2” is nowhere near as large as ChatGPT, so expect a much simpler answer.

output = await kernel.invoke(
    my_function,
    request="What are whales?",
)

output = str(output).strip()

query_result1 = await memory.search(
    collection=collection_id, query="What are sharks?", limit=1, min_relevance_score=0.3
)

print(f"The queried result for 'What are sharks?' is {query_result1[0].text}")

print(f"{text_service_id} completed prompt with: '{output}'")