[Bug]: download of onnx all-MiniLM-L6-v2 fails in chromadb\utils\embedding_functions\onnx_mini_lm_l6_v2.py

[bmerkle]

What happened?

I do a fresh setup of chroma, want to compute embeddings with all-MiniLM-L6-v2

the following code results in a timeout exception:
from chromadb.utils import embedding_functions default_ef = embedding_functions.DefaultEmbeddingFunction() val = default_ef(["foo"])

call which fails is:
with httpx.stream("GET", url) as resp:

chroma/chromadb/utils/embedding_functions/onnx_mini_lm_l6_v2.py

Line 100 in ff69c4d

with httpx.stream("GET", url) as resp:

my fix is to pass the timeout parameter to the httpx.stream() function call, to avoid the timeout, so
with httpx.stream("GET", url, timeout=None) as resp:
fixed the problem.

I will submit this as PR.

Please consider that in corporate networks there maybe proxies, firewalls, which cause a delay, which then results in a failed download of the model. I do not know about a default timeout of httpx but the fix/PR avoids the problem.

I had to roll back to Chroma 0.5.3 because of #2534 and vanna-ai/vanna#584

Versions

Chroma 0.5.3, Python 3.11.9, Windows 10

I had to roll back to Chroma 0.5.3 because of #2534 and vanna-ai/vanna#584

Relevant log output

No response