oomtig 1 Issue: Rate Limiting When Down DeepSeek-V3-0324
I’m experiencing consistent rate limiting issues when trying to download the DeepSeek-V3-0324 large model.
What I’ve tried:
Git over HTTPS:
- Direct download attempts result in “Rate limit exceeded” errors
Hugging Face CLI:
- Created and provided a token with write permissions
- Error 429 (Too Many Requests) for the
/whoami
endpoint
- Set HF_TOKEN environment variable
Git over SSH:
- Set up SSH authentication
- Still receiving rate limit errors when pulling LFS objects
Curl request:
Specific error messages:
batch response: Rate limit exceeded: https://huggingface.co/deepseek-ai/DeepSeek-V3-0324.git/info/lfs/objects/batch
Failed to fetch some objects from 'https://hf.co/deepseek-ai/DeepSeek-V3-0324.git/info/lfs'
git over ssh with skip(configured correctly)
GIT_LFS_SKIP_SMUDE=1 git clone git@hf.co:deepseek-ai/DeepSeek-V3-0324
Cloning into 'DeepSeek-V3-0324'...
remote: Enumerating objects: 203, done.
remote: Counting objects: 100% (199/199), done.
remote: Compressing objects: 100% (198/198), done.
remote: Total 203 (delta 13), reused 0 (delta 0), pack-reused 4 (from 1)
Receiving objects: 100% (203/203), 2.41 MiB | 6.35 MiB/s, done.
Resolving deltas: 100% (13/13), done.
Down model-00001-of-000163.safetensors (5.2 GB)
Error down object: model-00001-of-000163.safetensors (134f51f): Smudge error: Error down model-00001-of-000163.safetensors (134f51f4642dde3f3d63aacc8e4605e6355d27aceb4ee1a1f914e30464623f10): batch response: Rate limit exceeded: https://huggingface.co/deepseek-ai/DeepSeek-V3-0324.git/info/lfs/objects/batch
Huggingface Cli error
Traceback (most recent call last):
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_http.py", line 409, in hf_raise_for_status
response.raise_for_status()
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/requests/models.py", line 1024, in raise_for_status
raise HTTPError(http_error_msg, response=self)
requests.exceptions.HTTPError: 429 Client Error: Too Many Requests for url: https://huggingface.co/api/whoami-v2
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/hf_api.py", line 1737, in whoami
hf_raise_for_status(r)
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_http.py", line 482, in hf_raise_for_status
raise _format(HfHubHTTPError, str(e), response) from e
huggingface_hub.errors.HfHubHTTPError: 429 Client Error: Too Many Requests for url: https://huggingface.co/api/whoami-v2
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_deprecation.py", line 101, in inner_f
return f(*args, **kwargs)
^^^^^^^^^^^^^^^^^^
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_deprecation.py", line 31, in inner_f
return f(*args, **kwargs)
^^^^^^^^^^^^^^^^^^
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/_login.py", line 130, in login
interpreter_login(new_session=new_session)
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_deprecation.py", line 101, in inner_f
return f(*args, **kwargs)
^^^^^^^^^^^^^^^^^^
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_deprecation.py", line 31, in inner_f
return f(*args, **kwargs)
^^^^^^^^^^^^^^^^^^
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/_login.py", line 291, in interpreter_login
_login(token=token, add_to_git_credential=add_to_git_credential)
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/_login.py", line 405, in _login
token_info = whoami(token)
Is anyone else experiencing issues with DeepSeek-V3-0324? Are there alternative download methods available?
Any assistance would be appreciated.
There have been a lot of reports of 429 errors for about two weeks, so it’s possible that the rate limit has been raised. Also, as a method of down, there is the following method that uses the Python huggingface_hub library.
Just for testing, I’ve reattempted down from scratch, with SSH authentication, git ssh and extremely limited download bandwidth(about 10-20 Gigs an hour).
I particularly met the rate limiting error after I downloaded the 100th safetensor so can’t download anything over 100
This is the error
18:11:10.393098 trace git-lfs: tq: sending batch of size 1 [0/775]
18:11:10.393267 trace git-lfs: ssh cache expired: git@hf.co git-lfs-authenticate deepseek-ai/DeepSeek-V3-0324 download
18:11:10.393307 trace git-lfs: run_command: sh -c ssh -v git@hf.co 'git-lfs-authenticate deepseek-ai/DeepSeek-V3-0324 download'
18:11:10.393361 trace git-lfs: exec: sh '-c' 'ssh -v git@hf.co 'git-lfs-authenticate deepseek-ai/DeepSeek-V3-0324 download''
18:11:10.757466 trace git-lfs: api: batch 1 files
18:11:10.757848 trace git-lfs: HTTP: POST https://huggingface.co/deepseek-ai/DeepSeek-V3-0324.git/info/lfs/objects/batch
> POST /deepseek-ai/DeepSeek-V3-0324.git/info/lfs/objects/batch HTTP/1.1
> Host: huggingface.co
> Accept: application/vnd.git-lfs+json
> Authorization: Basic * * * * *
> Content-Length: 233
> Content-Type: application/vnd.git-lfs+json; charset=utf-8
> User-Agent: git-lfs/3.3.0 (GitHub; linux amd64; go 1.19.8)
>
{"operation":"download","objects":[{"oid":"9976dae2a44b4407a3a39dc69104ccce67001a307b1b49e966c98ab246b371f2","size":4302317817}],"transfers":["lfs-standalone-file","basic","ssh"],"ref":{"nam
e":"refs/heads/main"},"hash_algo":"sha256"}18:11:10.883177 trace git-lfs: HTTP: 429
< HTTP/2.0 429 Too Many Requests
< Content-Length: 3134
< Content-Type: text/html
< Date: Tue, 01 Apr 2025 18:11:10 GMT
< Server: CloudFront
< Via: 1.1 2f51a381830e231f6bc2b46fda74f69e.cloudfront.net (CloudFront)
< X-Amz-Cf-Id: UAy76RQ23ZYFRXxVpysyAzg765cXgMIF16yeb8vec4LgJ9M_kBvgVQ==
< X-Amz-Cf-Pop: ORD56-P7
< X-Cache: Error from cloudfront
<
18:11:10.883437 trace git-lfs: api error: Rate limit exceeded: https://huggingface.co/deepseek-ai/DeepSeek-V3-0324.git/info/lfs/objects/batch
Error down object: model-00101-of-000163.safetensors (9976dae): Smudge error: Error down model-00101-of-000163.safetensors (9976dae2a44b4407a3a39dc69104ccce67001a307b1b49e966c9
8ab246b371f2): batch response: Rate limit exceeded: https://huggingface.co/deepseek-ai/DeepSeek-V3-0324.git/info/lfs/objects/batch
Errors logged to '/mnt/deepseek/deepseek_model/DeepSeek-V3-0324/.git/lfs/logs/20250401T181110.883602204.log'.
Use `git lfs logs last` to view the log.
error: external filter 'git-lfs filter-process' failed
fatal: model-00101-of-000163.safetensors: smudge filter lfs failed
warning: Clone succeeded, but checkout failed.
You can inspect what was checked out with 'git status'
and retry with 'git restore --source=HEAD :/'
Following your recommendation of using huggingface_hub - I get the same error, at the same endpoint for this particular file(I can download other small files)
>>> hf_hub_download(repo_id='deepseek-ai/DeepSeek-V3-0324', filename ='model-00101-of-000163.safetensors') [4/833]
Traceback (most recent call last):
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_http.py", line 409, in hf_raise_for_status
response.raise_for_status() File "/mnt/deepseek/.venv/lib/python3.11/site-packages/requests/models.py", line 1024, in raise_for_status
raise HTTPError(http_error_msg, response=self)
requests.exceptions.HTTPError: 429 Client Error: Too Many Requests for url: https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/resolve/main/model-00101-of-000163.safetensors
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/file_download.py", line 1484, in _get_metadata_or_catch_error
metadata = get_hf_file_metadata(
^^^^^^^^^^^^^^^^^^^^^
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/file_download.py", line 1401, in get_hf_file_metadata
r = _request_wrapper(
^^^^^^^^^^^^^^^^^
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/file_download.py", line 285, in _request_wrapper
response = _request_wrapper(
^^^^^^^^^^^^^^^^^
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/file_download.py", line 309, in _request_wrapper
hf_raise_for_status(response)
File "/mnt/deepseek/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_http.py", line 482, in hf_raise_for_status
raise _format(HfHubHTTPError, str(e), response) from e
huggingface_hub.errors.HfHubHTTPError: 429 Client Error: Too Many Requests for url: https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/resolve/main/model-00101-of-000163.safetensors
The above exception was the direct cause of the following exception:
The bugs referred has been raised 21st November, so it seems the model at the moment is not accessible.
Inspecting the download rates - I can also notice in the download counter that the downloads in the month started to go in a rapid decline.
I didn’t find a recent post on this issue.
Is this a bug of some kind…? @Wauplin @pierric
oomtig 5 Just to clean up and summarize experience:
Using git downloads more than 1T data, possibly leading to IP ban,
Using a custom cloud provider - the assigned IP address might be in the range of banned addresses, resulting in 429 rejection
It seems the best approach is a targeted download of the tensorfiles with huggingface_hub cli as indicated
I can approve - this seems to work, if the rate limit has not been hit.
Have you tried shutting off your vpn? Same happened to me then my dumb butt realized it was my vpn
I would like to download the DeepSeek R1 Distill 70B model as well as several others. I am running Debian 12, have python 3.11.2, have all standard build tools for gcc, g++, gfortran, gnu Octave, etc. I have not installed git and git-lfs as one AI recommended: though I would ask here (AI often goofs). I want a download method that can pick up fron where it left should a glitch or interruption occur. I see huggingface_hub cli recommended, any instructions for using it? I am very familiar with standard Bash CLI, and would usually use wget or curl, or other standard tool. No rush, so down a small chunk at a time is fine, as is throttling the download rate.