help！"No valid XML" when using ollama

[ZdengdengZZ]

2025-10-31 13:26:19,732 - INFO - api.api - api.py:591 - Scanning for project cache files in: /home/tcci/.adalflow/wikicache
2025-10-31 13:26:19,736 - INFO - api.api - api.py:629 - Found 0 processed project entries.
2025-10-31 13:26:19,760 - INFO - api.api - api.py:591 - Scanning for project cache files in: /home/tcci/.adalflow/wikicache
2025-10-31 13:26:19,761 - INFO - api.api - api.py:629 - Found 0 processed project entries.
2025-10-31 13:26:20,594 - INFO - api.api - api.py:179 - Fetching model configurations
2025-10-31 13:26:20,617 - INFO - api.api - api.py:179 - Fetching model configurations
2025-10-31 13:26:28,054 - INFO - api.api - api.py:476 - Attempting to retrieve wiki cache for AsyncFuncAI/deepwiki-open (github), lang: zh
2025-10-31 13:26:28,054 - INFO - api.api - api.py:483 - Wiki cache not found for AsyncFuncAI/deepwiki-open (github), lang: zh
2025-10-31 13:26:32,214 - INFO - api.simple_chat - simple_chat.py:85 - Request size: 8755 tokens
2025-10-31 13:26:32,214 - WARNING - api.simple_chat - simple_chat.py:87 - Request exceeds recommended token limit (8755 > 7500)
2025-10-31 13:26:32,223 - INFO - api.ollama_patch - ollama_patch.py:48 - Ollama model 'nomic-embed-text' is available
2025-10-31 13:26:32,289 - INFO - adalflow.core.prompt_builder - prompt_builder.py:74 - Prompt has variables: ['schema']
2025-10-31 13:26:32,290 - INFO - adalflow.core.prompt_builder - prompt_builder.py:74 - Prompt has variables: ['schema']
2025-10-31 13:26:32,290 - INFO - api.rag - rag.py:74 - Dialog turns list exists but is empty
2025-10-31 13:26:32,290 - INFO - api.rag - rag.py:88 - Returning 0 dialog turns from memory
2025-10-31 13:26:32,345 - INFO - adalflow.optim.grad_component - grad_component.py:79 - EvalFnToTextLoss: No backward engine provided. Creating one using model_client and model_kwargs.
2025-10-31 13:26:32,346 - INFO - adalflow.core.generator - generator.py:188 - Generator Generator initialized.
2025-10-31 13:26:32,346 - INFO - api.data_pipeline - data_pipeline.py:780 - Preparing repo storage for https://github.com/AsyncFuncAI/deepwiki-open...
2025-10-31 13:26:32,346 - INFO - api.data_pipeline - data_pipeline.py:790 - Extracted repo name: AsyncFuncAI_deepwiki-open
2025-10-31 13:26:32,346 - INFO - api.data_pipeline - data_pipeline.py:799 - Repository already exists at /home/tcci/.adalflow/repos/AsyncFuncAI_deepwiki-open. Using existing repository.
2025-10-31 13:26:32,346 - INFO - api.data_pipeline - data_pipeline.py:813 - Repo paths: {'save_repo_dir': '/home/tcci/.adalflow/repos/AsyncFuncAI_deepwiki-open', 'save_db_file': '/home/tcci/.adalflow/databases/AsyncFuncAI_deepwiki-open.pkl'}
2025-10-31 13:26:32,346 - INFO - api.data_pipeline - data_pipeline.py:843 - Loading existing database...
2025-10-31 13:26:32,381 - INFO - adalflow.core.component - component.py:335 - Restoring class using from_dict TextSplitter, {'type': 'TextSplitter', 'data': {'_components': {'_ordered_dict': True, 'data': []}, '_parameters': {'_ordered_dict': True, 'data': []}, 'training': False, 'teacher_mode': False, 'tracing': False, 'name': 'TextSplitter', '_init_args': {'split_by': 'word', 'chunk_size': 800, 'chunk_overlap': 200, 'batch_size': 1000, 'separators': {'page': '\x0c', 'passage': '\n\n', 'word': ' ', 'sentence': '.', 'token': ''}}, 'split_by': 'word', 'separators': {'page': '\x0c', 'passage': '\n\n', 'word': ' ', 'sentence': '.', 'token': ''}, 'chunk_size': 350, 'chunk_overlap': 100, 'batch_size': 1000}}
2025-10-31 13:26:32,381 - INFO - adalflow.core.component - component.py:335 - Restoring class using from_dict OllamaDocumentProcessor, {'type': 'OllamaDocumentProcessor', 'data': {'_components': {'_ordered_dict': True, 'data': [('embedder', {'type': 'Embedder', 'data': {'_components': {'_ordered_dict': True, 'data': [('model_client', {'type': 'OllamaClient', 'data': {'_components': {'_ordered_dict': True, 'data': []}, '_parameters': {'_ordered_dict': True, 'data': []}, 'training': False, 'teacher_mode': False, 'tracing': False, 'name': 'OllamaClient', '_init_args': {'host': None}, '_host': 'http://localhost:11434', 'generate': False}})]}, '_parameters': {'_ordered_dict': True, 'data': []}, 'training': False, 'teacher_mode': False, 'tracing': False, 'name': 'Embedder', '_init_args': {'model_client': None, 'model_kwargs': {'model': 'nomic-embed-text'}, 'output_processors': None}, 'model_kwargs': {'model': 'nomic-embed-text'}, 'output_processors': None}})]}, '_parameters': {'_ordered_dict': True, 'data': []}, 'training': False, 'teacher_mode': False, 'tracing': False, 'name': 'OllamaDocumentProcessor', '_init_args': {'embedder': None}}}
2025-10-31 13:26:32,381 - INFO - adalflow.core.component - component.py:335 - Restoring class using from_dict Embedder, {'type': 'Embedder', 'data': {'_components': {'_ordered_dict': True, 'data': [('model_client', {'type': 'OllamaClient', 'data': {'_components': {'_ordered_dict': True, 'data': []}, '_parameters': {'_ordered_dict': True, 'data': []}, 'training': False, 'teacher_mode': False, 'tracing': False, 'name': 'OllamaClient', '_init_args': {'host': None}, '_host': 'http://localhost:11434', 'generate': False}})]}, '_parameters': {'_ordered_dict': True, 'data': []}, 'training': False, 'teacher_mode': False, 'tracing': False, 'name': 'Embedder', '_init_args': {'model_client': None, 'model_kwargs': {'model': 'nomic-embed-text'}, 'output_processors': None}, 'model_kwargs': {'model': 'nomic-embed-text'}, 'output_processors': None}}
2025-10-31 13:26:32,382 - INFO - adalflow.core.component - component.py:335 - Restoring class using from_dict OllamaClient, {'type': 'OllamaClient', 'data': {'_components': {'_ordered_dict': True, 'data': []}, '_parameters': {'_ordered_dict': True, 'data': []}, 'training': False, 'teacher_mode': False, 'tracing': False, 'name': 'OllamaClient', '_init_args': {'host': None}, '_host': 'http://localhost:11434', 'generate': False}}
2025-10-31 13:26:32,435 - INFO - api.data_pipeline - data_pipeline.py:848 - Loaded 978 documents from existing database
2025-10-31 13:26:32,436 - INFO - api.rag - rag.py:372 - Loaded 978 documents for retrieval
2025-10-31 13:26:32,436 - INFO - api.rag - rag.py:380 - Using 978 documents with valid embeddings for retrieval
2025-10-31 13:26:32,436 - INFO - adalflow.optim.grad_component - grad_component.py:79 - EvalFnToTextLoss: No backward engine provided. Creating one using model_client and model_kwargs.
2025-10-31 13:26:32,456 - WARNING - adalflow.components.retriever.faiss_retriever - faiss_retriever.py:191 - Embeddings are not normalized, normalizing the embeddings
2025-10-31 13:26:32,460 - INFO - adalflow.components.retriever.faiss_retriever - faiss_retriever.py:197 - Index built with 978 chunks
2025-10-31 13:26:32,460 - INFO - api.rag - rag.py:391 - FAISS retriever created successfully
2025-10-31 13:26:32,460 - INFO - api.simple_chat - simple_chat.py:114 - Retriever prepared for https://github.com/AsyncFuncAI/deepwiki-open
2025-10-31 13:26:32,460 - INFO - api.rag - rag.py:74 - Dialog turns list exists but is empty
2025-10-31 13:26:32,460 - INFO - api.rag - rag.py:88 - Returning 0 dialog turns from memory
2025-10-31 13:26:32,460 - INFO - api.simple_chat - simple_chat.py:324 - No context available from RAG
2025-10-31 13:27:43,401 - INFO - httpx - _client.py:1786 - HTTP Request: POST http://localhost:11434/api/chat "HTTP/1.1 200 OK"

[ZILosoft]

same

[toBeInTheDocument]

same

[hawk9821]

wikicache not created.

2025-11-05 11:03:05,974 - INFO - api.websocket_wiki - websocket_wiki.py:202 - Retrieved 20 documents
2025-11-05 11:03:05,977 - INFO - api.rag - rag.py:74 - Dialog turns list exists but is empty
2025-11-05 11:03:06,066 - INFO - api.rag - rag.py:88 - Returning 0 dialog turns from memory
2025-11-05 11:03:06,075 - DEBUG - adalflow.components.model_client.ollama_client - ollama_client.py:308 - Using host: http://localhost:11434
2025-11-05 11:03:09,540 - DEBUG - httpcore.connection - _trace.py:87 - connect_tcp.started host='127.0.0.1' port=7890 local_address=None timeout=None socket_options=None
INFO:     127.0.0.1:55929 - "GET /auth/status HTTP/1.1" 200 OK
INFO:     127.0.0.1:62161 - "GET /auth/status HTTP/1.1" 200 OK
2025-11-05 11:03:09,545 - DEBUG - httpcore.connection - _trace.py:87 - connect_tcp.complete return_value=<httpcore._backends.anyio.AnyIOStream object at 0x000001BEB6F19E20>
2025-11-05 11:03:09,546 - DEBUG - httpcore.http11 - _trace.py:87 - send_request_headers.started request=<Request [b'POST']>
2025-11-05 11:03:09,547 - DEBUG - httpcore.http11 - _trace.py:87 - send_request_headers.complete
2025-11-05 11:03:09,547 - DEBUG - httpcore.http11 - _trace.py:87 - send_request_body.started request=<Request [b'POST']>
2025-11-05 11:03:09,548 - DEBUG - httpcore.http11 - _trace.py:87 - send_request_body.complete
2025-11-05 11:03:09,548 - DEBUG - httpcore.http11 - _trace.py:87 - receive_response_headers.started request=<Request [b'POST']>
2025-11-05 11:03:13,923 - DEBUG - httpcore.http11 - _trace.py:87 - receive_response_headers.complete return_value=(b'HTTP/1.1', 200, b'OK', [(b'Connection', b'close'), (b'Transfer-Encoding', b'chunked'), (b'Content-Type', b'application/x-ndjson'), (b'Date', b'Wed, 05 Nov 2025 03:03:13 GMT')])
2025-11-05 11:03:13,923 - INFO - httpx - _client.py:1740 - HTTP Request: POST http://localhost:11434/api/chat "HTTP/1.1 200 OK"
2025-11-05 11:03:13,923 - DEBUG - httpcore.http11 - _trace.py:87 - receive_response_body.started request=<Request [b'POST']>
2025-11-05 11:03:46,984 - DEBUG - httpcore.http11 - _trace.py:87 - receive_response_body.complete
2025-11-05 11:03:46,984 - DEBUG - httpcore.http11 - _trace.py:87 - response_closed.started
2025-11-05 11:03:46,985 - DEBUG - httpcore.http11 - _trace.py:87 - response_closed.complete

[utopiaprince]

I get the same problem.

[loon-hejw]

same

[loon-hejw]

1. Currently, I've solved the problem. First, if you're using it in a container or Kubernetes, you need to pay attention to the UI WebSocket address.
2. Second, you need to modify the code in the api/websocket_wiki.py file where it says request.provider equals "ollama".

But it's still in the testing phase currently.

         if request.provider == "ollama":
                # Get the response and handle it properly using the previously created api_kwargs
                response = await model.acall(api_kwargs=api_kwargs, model_type=ModelType.LLM)
                # Handle streaming response from Ollama
                async for chunk in response:
                    text = None
                    if isinstance(chunk, dict):
                        text = chunk.get("message", {}).get("content") if isinstance(chunk.get("message"), dict) else chunk.get("message")
                    else:
                        message = getattr(chunk, "message", None)
                        if message is not None:
                            if isinstance(message, dict):
                                text = message.get("content")
                            else:
                                text = getattr(message, "content", None)

                    if not text:
                        text = getattr(chunk, 'response', None) or getattr(chunk, 'text', None)

                    if not text and hasattr(chunk, "__dict__"):
                        message = chunk.__dict__.get("message")
                        if isinstance(message, dict):
                            text = message.get("content")

                    if isinstance(text, str) and text and not text.startswith('model=') and not text.startswith('created_at='):
                        clean_text = text.replace('<think>', '').replace('</think>', '')
                        await websocket.send_text(clean_text)
                # Explicitly close the WebSocket connection after the response is complete
                await websocket.close()

[NuoLanC]

我也遇到相同的问题了，不懂如何解决

[loon-hejw]

我也遇到相同的问题了，不懂如何解决

试一下我的法子那

[NuoLanC]

我没用ollama,直接按官网的文档部署的docker 的。也是这个提示呢，

[johnpwhite]

1. Currently, I've solved the problem. First, if you're using it in a container or Kubernetes, you need to pay attention to the UI WebSocket address.
2. Second, you need to modify the code in the api/websocket_wiki.py file where it says request.provider equals "ollama".

But it's still in the testing phase currently.

         if request.provider == "ollama":
                # Get the response and handle it properly using the previously created api_kwargs
                response = await model.acall(api_kwargs=api_kwargs, model_type=ModelType.LLM)
                # Handle streaming response from Ollama
                async for chunk in response:
                    text = None
                    if isinstance(chunk, dict):
                        text = chunk.get("message", {}).get("content") if isinstance(chunk.get("message"), dict) else chunk.get("message")
                    else:
                        message = getattr(chunk, "message", None)
                        if message is not None:
                            if isinstance(message, dict):
                                text = message.get("content")
                            else:
                                text = getattr(message, "content", None)

                    if not text:
                        text = getattr(chunk, 'response', None) or getattr(chunk, 'text', None)

                    if not text and hasattr(chunk, "__dict__"):
                        message = chunk.__dict__.get("message")
                        if isinstance(message, dict):
                            text = message.get("content")

                    if isinstance(text, str) and text and not text.startswith('model=') and not text.startswith('created_at='):
                        clean_text = text.replace('<think>', '').replace('</think>', '')
                        await websocket.send_text(clean_text)
                # Explicitly close the WebSocket connection after the response is complete
                await websocket.close()

I made this change where i think it should be in that file (there's a couple places with if request.provider == "ollama":), but i get the same XML error. The logs don't move on from this line

2025-11-14 22:23:40,436 - INFO - httpx - _client.py:1740 - HTTP Request: POST http://192.168.1.8:11434/api/chat "HTTP/1.1 200 OK"

I can't read the chinese responses :-(

[loon-hejw]

The original code:

if request.provider == "ollama":
    # Get the response and handle it properly using the previously created api_kwargs
    response = await model.acall(api_kwargs=api_kwargs, model_type=ModelType.LLM)
    # Handle streaming response from Ollama
    async for chunk in response:
        text = getattr(chunk, 'response', None) or getattr(chunk, 'text', None) or str(chunk)
        if text and not text.startswith('model=') and not text.startswith('created_at='):
            text = text.replace('<think>', '').replace('</think>', '')
            await websocket.send_text(text)
    # Explicitly close the WebSocket connection after the response is complete
    await websocket.close()

is the section that was replaced.

[loon-hejw]

1. Currently, I've solved the problem. First, if you're using it in a container or Kubernetes, you need to pay attention to the UI WebSocket address.
2. Second, you need to modify the code in the api/websocket_wiki.py file where it says request.provider equals "ollama".

But it's still in the testing phase currently.

         if request.provider == "ollama":
                # Get the response and handle it properly using the previously created api_kwargs
                response = await model.acall(api_kwargs=api_kwargs, model_type=ModelType.LLM)
                # Handle streaming response from Ollama
                async for chunk in response:
                    text = None
                    if isinstance(chunk, dict):
                        text = chunk.get("message", {}).get("content") if isinstance(chunk.get("message"), dict) else chunk.get("message")
                    else:
                        message = getattr(chunk, "message", None)
                        if message is not None:
                            if isinstance(message, dict):
                                text = message.get("content")
                            else:
                                text = getattr(message, "content", None)

                    if not text:
                        text = getattr(chunk, 'response', None) or getattr(chunk, 'text', None)

                    if not text and hasattr(chunk, "__dict__"):
                        message = chunk.__dict__.get("message")
                        if isinstance(message, dict):
                            text = message.get("content")

                    if isinstance(text, str) and text and not text.startswith('model=') and not text.startswith('created_at='):
                        clean_text = text.replace('<think>', '').replace('</think>', '')
                        await websocket.send_text(clean_text)
                # Explicitly close the WebSocket connection after the response is complete
                await websocket.close()

I made this change where i think it should be in that file (there's a couple places with if request.provider == "ollama":), but i get the same XML error. The logs don't move on from this line

2025-11-14 22:23:40,436 - INFO - httpx - _client.py:1740 - HTTP Request: POST http://192.168.1.8:11434/api/chat "HTTP/1.1 200 OK"

I can't read the chinese responses :-(

---

Since I’m not very familiar with Python, I only made some rough modifications, and I'm not sure whether the official team would consider this an acceptable approach. However, it’s currently working on my side.
If you have any further questions, feel free to reply. I’ll do my best to help.

[loon-hejw]

If the modified code runs, try executing it first. If issues still occur, check the browser’s WebSocket requests.
When running inside a container, there’s a high chance the WebSocket may be connecting to the wrong address, such as localhost:8001. Make sure this address matches the actual Python service endpoint.

[johnpwhite]

Hi @loon-hejw , How do i check the websocket value, and how do i change it if it is wrong?

[seydel1847]

Guys, i am not software developer but I know where is the mess.
There is hint for you where to look for problem.

embedder.py
See that commented code below:
def get_embedder(is_local_ollama: bool = False, use_google_embedder: bool = False, embedder_type: str = None) -> adal.Embedder:
"""Get embedder based on configuration or parameters.

Args:
    is_local_ollama: Legacy parameter for Ollama embedder
    use_google_embedder: Legacy parameter for Google embedder  
    embedder_type: Direct specification of embedder type ('ollama', 'google', 'openai')

Returns:
    adal.Embedder: Configured embedder instance
"""
# Determine which embedder config to use

print("RagE-1")
print(embedder_type)
print(is_local_ollama)
print(configs)
# print(configs["embedder_ollama"])

embedder_config = configs["embedder"]

# if embedder_type:
#     if embedder_type == 'ollama':
#         embedder_config = configs["embedder_ollama"]
#     elif embedder_type == 'google':
#         embedder_config = configs["embedder_google"]
#     else:  # default to openai
#         embedder_config = configs["embedder"]
# elif is_local_ollama:
#     embedder_config = configs["embedder_ollama"]
# elif use_google_embedder:
#     embedder_config = configs["embedder_google"]
# else:
#     # Auto-detect based on current configuration
#     current_type = get_embedder_type()
#     print("current_type")
#     print(current_type)
#     if current_type == 'ollama':
#         embedder_config = configs["embedder_ollama"]
#     elif current_type == 'google':
#         embedder_config = configs["embedder_google"]
#     else:
#         embedder_config = configs["embedder"]

JSON for confihs loooks like this one:
{
"default_provider": "google",
"providers": {
...
},
"embedder": {
"client_class": "OllamaClient",
"model_kwargs": {
"model": "nomic-embed-text"
},
"model_client": <adalflow.utils.lazy_import.LazyImport object at 0x0000022838596EA0>
},

So as you can see there is configs['embedder'] and there is no configs["embedder_ollama"].
The logic seems to be fucked here and errors reported to output are not really meaningful because 'No valid XML' is very far away from the real source of the problem.
You may do this very ugly fix done by non-developer and it will worrk like a hell :)))
Happy coding and troubleshooting to you all ;)

BTW:
It may happen that there are many more errors hidden in this code and error reporting si so bad that all errors are under 'No valid XML'.
In this case please refactor code for better handling of exceprtions ...