ainsert()的时候维度始终不对

[learningpro]

使用源代码中的examples来进行尝试，不论是使用ollama，还是openai，或者是openai兼容的接口，都会报同样的错误：
提示：
/LightRAG/lightrag/lightrag.py", line 162, in insert
return loop.run_until_complete(self.ainsert(string_or_strings))

all the input array dimensions except for the concatenation axis must match exactly, but along dimension 1, the array at index 0 has size 768 and the array at index 1 has size 1024

[LarFii]

可以尝试新建一个工作路径再次运行

[amenhere]

可以尝试新建一个工作路径再次运行

不行，还是这个错误，是有哪里设置为1024了吗

[LarFii]

给我看看log吧，还有具体的代码

[Kilimajaro]

同样的报错，清除工作路径缓存无效：(cail) (base) dell@dell-PowerEdge-R750:~/test/LexiLaw-main/demo$ /home/dell/anaconda3/envs/cail/bin/python /home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/lightrag_openai_compatible_demo.py
INFO:lightrag:Logger initialized for working directory: /home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/new
DEBUG:lightrag:LightRAG init with param:
working_dir = /home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/new,
chunk_token_size = 1200,
chunk_overlap_token_size = 100,
tiktoken_model_name = gpt-4o-mini,
entity_extract_max_gleaning = 1,
entity_summary_to_max_tokens = 500,
node_embedding_algorithm = node2vec,
node2vec_params = {'dimensions': 1536, 'num_walks': 10, 'walk_length': 40, 'window_size': 2, 'iterations': 3, 'random_seed': 3},
embedding_func = {'embedding_dim': 4096, 'max_token_size': 8192, 'func': <function embedding_func at 0x73bddf912170>},
embedding_batch_num = 32,
embedding_func_max_async = 16,
llm_model_func = <function llm_model_func at 0x73bddf911fc0>,
llm_model_name = meta-llama/Llama-3.2-1B-Instruct,
llm_model_max_token_size = 32768,
llm_model_max_async = 16,
key_string_value_json_storage_cls = <class 'lightrag.storage.JsonKVStorage'>,
vector_db_storage_cls = <class 'lightrag.storage.NanoVectorDBStorage'>,
vector_db_storage_cls_kwargs = {},
graph_storage_cls = <class 'lightrag.storage.NetworkXStorage'>,
enable_llm_cache = True,
addon_params = {},
convert_response_to_json_func = <function convert_response_to_json at 0x73bc8848a8c0>

INFO:lightrag:Load KV full_docs with 0 data
INFO:lightrag:Load KV text_chunks with 0 data
INFO:lightrag:Load KV llm_response_cache with 0 data
INFO:nano-vectordb:Init {'embedding_dim': 4096, 'metric': 'cosine', 'storage_file': '/home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/new/vdb_entities.json'} 0 data
INFO:nano-vectordb:Init {'embedding_dim': 4096, 'metric': 'cosine', 'storage_file': '/home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/new/vdb_relationships.json'} 0 data
INFO:nano-vectordb:Init {'embedding_dim': 4096, 'metric': 'cosine', 'storage_file': '/home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/new/vdb_chunks.json'} 0 data
INFO:lightrag:Creating a new event loop in a sub-thread.
INFO:lightrag:[New Docs] inserting 1 docs
INFO:lightrag:[New Chunks] inserting 1 chunks
INFO:lightrag:Inserting 1 vectors to chunks
INFO:httpx:HTTP Request: POST https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings "HTTP/1.1 200 OK"
32842 32842
INFO:lightrag:Writing graph with 0 nodes, 0 edges
Traceback (most recent call last):
File "/home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/lightrag_openai_compatible_demo.py", line 59, in
rag.insert(f.read())
File "/home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/lightrag/lightrag.py", line 162, in insert
return loop.run_until_complete(self.ainsert(string_or_strings))
File "/home/dell/anaconda3/envs/cail/lib/python3.10/asyncio/base_events.py", line 649, in run_until_complete
return future.result()
File "/home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/lightrag/lightrag.py", line 206, in ainsert
await self.chunks_vdb.upsert(inserting_chunks)
File "/home/dell/test/LexiLaw-main/demo/LightRAG-main/examples/lightrag/storage.py", line 98, in upsert
results = self._client.upsert(datas=list_data)
File "/home/dell/anaconda3/envs/cail/lib/python3.10/site-packages/nano_vectordb/dbs.py", line 108, in upsert
self.__storage["matrix"] = np.vstack([self.__storage["matrix"], new_matrix])
File "/home/dell/anaconda3/envs/cail/lib/python3.10/site-packages/numpy/core/shape_base.py", line 289, in vstack
return _nx.concatenate(arrs, 0, dtype=dtype, casting=casting)
ValueError: all the input array dimensions except for the concatenation axis must match exactly, but along dimension 1, the array at index 0 has size 4096 and the array at index 1 has size 1536

[amenhere]

给我看看log吧，还有具体的代码

INFO:httpx:HTTP Request: POST http://192.168.31.100:3001/v1/chat/completions "HTTP/1.1 200 OK"
llm_model_func: I'm just a program, so I don't have feelings, but I'm here and ready to help you! How can I assist you today?
INFO:httpx:HTTP Request: POST http://192.168.31.100:21434/v1/embeddings "HTTP/1.1 200 OK"
embedding_func: [[-0.02448499 0.06301917 0.02070389 ... 0.02287727 0.05131468
-0.03886142]]
INFO:lightrag:Logger initialized for working directory: ./aka
DEBUG:lightrag:LightRAG init with param:
working_dir = ./aka,
chunk_token_size = 1200,
chunk_overlap_token_size = 100,
tiktoken_model_name = gpt-4o-mini,
entity_extract_max_gleaning = 1,
entity_summary_to_max_tokens = 500,
node_embedding_algorithm = node2vec,
node2vec_params = {'dimensions': 1536, 'num_walks': 10, 'walk_length': 40, 'window_size': 2, 'iterations': 3, 'random_seed': 3},
embedding_func = {'embedding_dim': 1792, 'max_token_size': 8192, 'func': <function embedding_func at 0x7fa653b6b490>},
embedding_batch_num = 32,
embedding_func_max_async = 16,
llm_model_func = <function llm_model_func at 0x7fa751ba7d90>,
llm_model_name = meta-llama/Llama-3.2-1B-Instruct,
llm_model_max_token_size = 32768,
llm_model_max_async = 16,
key_string_value_json_storage_cls = <class 'lightrag.storage.JsonKVStorage'>,
vector_db_storage_cls = <class 'lightrag.storage.NanoVectorDBStorage'>,
vector_db_storage_cls_kwargs = {},
graph_storage_cls = <class 'lightrag.storage.NetworkXStorage'>,
enable_llm_cache = True,
addon_params = {},
convert_response_to_json_func = <function convert_response_to_json at 0x7fa6544d6b00>

INFO:lightrag:Load KV full_docs with 0 data
INFO:lightrag:Load KV text_chunks with 0 data
INFO:lightrag:Load KV llm_response_cache with 0 data
INFO:nano-vectordb:Init {'embedding_dim': 1792, 'metric': 'cosine', 'storage_file': './aka/vdb_entities.json'} 0 data
INFO:nano-vectordb:Init {'embedding_dim': 1792, 'metric': 'cosine', 'storage_file': './aka/vdb_relationships.json'} 0 data
INFO:nano-vectordb:Init {'embedding_dim': 1792, 'metric': 'cosine', 'storage_file': './aka/vdb_chunks.json'} 0 data
INFO:lightrag:Creating a new event loop in a sub-thread.
INFO:lightrag:[New Docs] inserting 1 docs
INFO:lightrag:[New Chunks] inserting 2 chunks
INFO:lightrag:Inserting 2 vectors to chunks
INFO:httpx:HTTP Request: POST http://192.168.31.100:21434/v1/embeddings "HTTP/1.1 200 OK"
INFO:lightrag:Writing graph with 0 nodes, 0 edges
Traceback (most recent call last):
File "/root/LightRAG/examples/lightrag_openai_compatible_demo.py", line 61, in
rag.insert(f.read())
File "/root/LightRAG/lightrag/lightrag.py", line 162, in insert
return loop.run_until_complete(self.ainsert(string_or_strings))
File "/root/anaconda3/envs/lightrag/lib/python3.10/asyncio/base_events.py", line 641, in run_until_complete
return future.result()
File "/root/LightRAG/lightrag/lightrag.py", line 206, in ainsert
await self.chunks_vdb.upsert(inserting_chunks)
File "/root/LightRAG/lightrag/storage.py", line 98, in upsert
results = self._client.upsert(datas=list_data)
File "/root/anaconda3/envs/lightrag/lib/python3.10/site-packages/nano_vectordb/dbs.py", line 108, in upsert
self.__storage["matrix"] = np.vstack([self.__storage["matrix"], new_matrix])
File "/root/anaconda3/envs/lightrag/lib/python3.10/site-packages/numpy/core/shape_base.py", line 289, in vstack
return _nx.concatenate(arrs, 0, dtype=dtype, casting=casting)
ValueError: all the input array dimensions except for the concatenation axis must match exactly, but along dimension 1, the array at index 0 has size 1792 and the array at index 1 has size 1024

[amenhere]

import os
import asyncio
from lightrag import LightRAG, QueryParam
from lightrag.llm import openai_complete_if_cache, openai_embedding
from lightrag.utils import EmbeddingFunc
import numpy as np
from dotenv import load_dotenv
load_dotenv()

WORKING_DIR = "./aka"

if not os.path.exists(WORKING_DIR):
os.mkdir(WORKING_DIR)

async def llm_model_func(
prompt, system_prompt=None, history_messages=[], **kwargs
) -> str:
return await openai_complete_if_cache(
"gpt-4o-mini-2024-07-18",
prompt,
system_prompt=system_prompt,
history_messages=history_messages,
api_key=os.getenv("AZURE_OPENAI_API_KEY"),
base_url="http://192.168.31.100:3001/v1/",
**kwargs,
)

async def embedding_func(texts: list[str]) -> np.ndarray:
return await openai_embedding(
texts,
model="aerok/xiaobu-embedding-v2:latest", # ZNV-Embedding
api_key=os.getenv("AZURE_OPENAI_API_KEY"),
base_url="http://192.168.31.100:21434/v1/",
)

function test

async def test_funcs():
result = await llm_model_func("How are you?")
print("llm_model_func: ", result)

result = await embedding_func(["How are you?"])
print("embedding_func: ", result)

asyncio.run(test_funcs())

exit()

rag = LightRAG(
working_dir=WORKING_DIR,
llm_model_func=llm_model_func,
embedding_func=EmbeddingFunc(
embedding_dim=1792, max_token_size=8192, func=embedding_func
),
) #1792 #4096

with open("./book.txt") as f:
rag.insert(f.read())

Perform naive search

print(
rag.query("讲了一个什么故事，主题是什么？", param=QueryParam(mode="naive"))
)

Perform local search

print(
rag.query("讲了一个什么故事，主题是什么？", param=QueryParam(mode="local"))
)

Perform global search

print(
rag.query("讲了一个什么故事，主题是什么？", param=QueryParam(mode="global"))
)

Perform hybrid search

print(
rag.query("讲了一个什么故事，主题是什么？", param=QueryParam(mode="hybrid"))
)

[Kilimajaro]

类似报错，求大佬解答:《

[nicklhy]

+1

[nicklhy]

Solved this problem by modify the embedding_dim in lightrag_openai_compatible_demo.py.

In my case, I changed the default embedding model to bge-m3 and set the corresponding embedding_dim parameter as below

rag = LightRAG(
    working_dir=WORKING_DIR,
    llm_model_func=llm_model_func,
    embedding_func=EmbeddingFunc(
        embedding_dim=1024, max_token_size=8192, func=embedding_func
    ),
)

[Keito654]

I resolved this problem by changing the embedding_dim to match the number following the array at index 1 has size in the error message.
For example, if you encounter an error like:

ValueError: all the input array dimensions except for the concatenation axis must match exactly, but along dimension 1, the array at index 0 has size 1792 and the array at index 1 has size 1024

In this case, change embedding_dim to 1024.

[TianyuFan0504]

可能是embedding_dim没设置对，查询时的emb_dim和你的本地存储的dim不一样