ChromaDB插入数据报错：无法用神经网络模型计算预测

运行以下代码向ChromaDB集合添加数据时出现CoreML执行错误：

import ollama, chromadb, time

chroma_client = chromadb.Client()
#chroma_client = chromadb.HttpClient(host="localhost", port=8000)

collectionname = "my_collection"

print(chroma_client.list_collections())
if any(collection.name == collectionname for collection in chroma_client.list_collections()):
    print("deleting collection")
    chroma_client.delete_collection("my_collection")

collection = chroma_client.get_or_create_collection(name=collectionname, metadata={"hnsw:space": "cosine"})

collection.upsert(
    documents=[
        "This is a document about machine learning",
        "This is another document about data science",
        "A third document about artificial intelligence"
    ],
    metadatas=[
        {"source": "test1"},
        {"source": "test2"},
        {"source": "test3"}
    ],
    ids=[
        "id1",
        "id2",
        "id3"
    ]
)

错误信息：

2024-07-13 17:31:45.534848 [E:onnxruntime:, sequential_executor.cc:516 ExecuteKernel] Non-zero status code returned while running CoreML_16713284627214834184_1 node. Name:'CoreMLExecutionProvider_CoreML_16713284627214834184_1_1' Status Message: Error executing model: Unable to compute the prediction using a neural network model. It can be an invalid input data or broken/unsupported model (error code: -1).
Traceback (most recent call last):
  File "/Users/joe/development/python/gettings_started_chromadb.py", line 15, in <module>
    collection.upsert(
  File "/Users/joe/development/python/venv/dev/lib/python3.11/site-packages/chromadb/api/models/Collection.py", line 296, in upsert
    ) = self._validate_and_prepare_upsert_request(
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/joe/development/python/venv/dev/lib/python3.11/site-packages/chromadb/api/models/CollectionCommon.py", line 531, in _validate_and_prepare_upsert_request
    embeddings = self._embed(input=documents)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/joe/development/python/venv/dev/lib/python3.11/site-packages/chromadb/api/models/CollectionCommon.py", line 568, in _embed
    return self._embedding_function(input=input)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/joe/development/python/venv/dev/lib/python3.11/site-packages/chromadb/api/types.py", line 211, in __call__
    result = call(self, input)
             ^^^^^^^^^^^^^^^^^
  File "/Users/joe/development/python/venv/dev/lib/python3.11/site-packages/chromadb/utils/embedding_functions/onnx_mini_lm_l6_v2.py", line 200, in __call__
    return cast(Embeddings, self._forward(input).tolist())
                            ^^^^^^^^^^^^^^^^^^^^
  File "/Users/joe/development/python/venv/dev/lib/python3.11/site-packages/chromadb/utils/embedding_functions/onnx_mini_lm_l6_v2.py", line 143, in _forward
    model_output = self.model.run(None, onnx_input)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/joe/development/python/venv/dev/lib/python3.11/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 220, in run
    return self._sess.run(output_names, input_feed, run_options)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running CoreML_16713284627214834184_1 node. Name:'CoreMLExecutionProvider_CoreML_16713284627214834184_1_1' Status Message: Error executing model: Unable to compute the prediction using a neural network model. It can be an invalid input data or broken/unsupported model (error code: -1).

已尝试清理Python虚拟环境、Docker部署，问题仍存在。

• 禁用CoreML执行器
  强制ONNXRuntime使用CPU执行，避免CoreML兼容性问题。在代码开头添加：

  import os
  os.environ["ONNXRUNTIME_EXECUTION_PROVIDERS"] = "CPUExecutionProvider"
  

• 更换嵌入模型
  改用Ollama提供的嵌入函数替代Chroma默认的ONNX模型，修改集合创建代码：

  from chromadb.utils.embedding_functions import OllamaEmbeddingFunction
  
  embedding_function = OllamaEmbeddingFunction(
      model_name="llama3",  # 使用本地已拉取的Ollama模型
      base_url="http://localhost:11434"
  )
  
  collection = chroma_client.get_or_create_collection(
      name=collectionname,
      metadata={"hnsw:space": "cosine"},
      embedding_function=embedding_function
  )
  

  注意：需确保Ollama已启动，且指定的模型已通过ollama pull llama3拉取到本地。

• 调整ONNXRuntime版本
  当前版本的ONNXRuntime可能存在CoreML兼容bug，尝试安装稳定版：

  pip install onnxruntime==1.16.3 --force-reinstall
  

• 检查系统CoreML支持
  若使用旧版macOS（低于10.15 Catalina），CoreML对部分ONNX模型支持有限，需升级系统至符合要求的版本。

内容的提问来源于stack exchange，提问作者Joe Forrester