Update requirements.txt

[CryingN]

The current pip library does not provide version 2.4.1 'touch' and version 3.0.0 'triton', and the 'requirements.txt' file has been updated to a minimum to meet the current pip installation requirements

[musvaage]

cf: #410

this and the foregoing are subject matter for an Issue not for a Pull Request

[mowentian]

if you update requirements.txt, please make sure it works and update readme, thanks

[peti562]

Need to update readme here:

DeepSeek-V3/README.md

Line 275 in 1d7d440

torch==2.4.1

[CryingN]

Thank you for the reminder. Recently, some security incidents have made it impossible for me to take time to verify every file. I am slowly sorting out the information related to the changes in this branch.

[musvaage]

creating a virtual environment, and upgrading pip in that virtual environment to 23.0.1

in that virtual environment then running pip install -r requirements.txt produced this output

didn't let download complete however as the computer I ran the command on was running python 3.9.21

torch-2.4.1-cp39-cp39-manylinux1_x86_64.whl.metadata (26 kB)
triton-3.0.0-1-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl.metadata (1.3 kB)

[CryingN]

if you update requirements.txt, please make sure it works and update readme, thanks

Currently, I can only say that the branch can meet the basic requirements and run locally on my machine. I am trying to explain from a code perspective why it is suitable to update to the current version. I will reply again when I finish.

[CryingN]

creating a virtual environment, and upgrading pip in that virtual environment to 23.0.1

in that virtual environment then running produced this outputpip install -r requirements.txt

didn't let download complete however as the computer I ran the command on was running python 3.9.21

torch-2.4.1-cp39-cp39-manylinux1_x86_64.whl.metadata (26 kB) triton-3.0.0-1-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl.metadata (1.3 kB)

In this branch, I have made the following updates:

torch==2.6.0
triton==3.2.0

to meet the requirements for Python 3.13. Considering that the README.md file:

Note

Linux with Python 3.10 only. Mac and Windows are not supported.

I have not yet tested the version 3.9.21. However, these two versions of the environment clearly exist in the Tsinghua repository. In this branch, you can try switching to the following two wheel files for installation:

torch-2.6.0-cp39-cp39-manylinux1_x86_64.whl
triton-3.2.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

[musvaage]

This pull was opened on Feb 4, 2025.

I'm not trying on my end to install dependencies on an older than Python 3.10.

checked that if the torch version on this requirement file is 2.4.1,
when you run 'pip install -r requirements.txt', ...

The wording in #410 (above) seemed a bit vague to me and the thought crossed my mind that the user just ran the command pip install torch, which returned on Jan 28, 2025, that package's version 2.5.1, more recently, on Jan 29, 2025, that package's version 2.6.0 was released.

On Sep 4, 2024, that package's version 2.4.1 was released; the version currently listed in the README.md file.

I'm not thinking personally along the lines of newest known version but known to work version, which is why above I deferred the matter to an Issue. But then perhaps I'm missing something.

[mowentian]

I'm not thinking personally along the lines of newest known version but known to work version, which is why above I deferred the matter to an Issue. But then perhaps I'm missing something.

i think a work version is just fine, no need to follow the newest... but we should make sure it works fine (for Linux with Python 3.10 only)

[musvaage]

@piratagit
@CryingN

As it stands neither user has been forthcoming with for example an excerpt of shell output running Python 3.10 showing failure of torch==2.4.1 and triton==3.0.0 to complete installation, under that circumstance I therefore think it is appropriate to close both Pull Requests.

should make sure it works fine

As I wrote before the subject of updating [the python version] and dependencies' versions might favourably be raised in an Issue which states the reasons for the request and furthermore demonstrates that the user has confirmed before opening the Issue that a proposed updated interpreter version and/or dependencies' versions work with the specified software [and will pass tests, if and where appropriate]. This additionally might be a policy consideration as applied Organization wide.

[CryingN]

@piratagit @CryingN

As it stands neither user has been forthcoming with for example an excerpt of shell output running Python 3.10 showing failure of torch==2.4.1 and triton==3.0.0 to complete installation, under that circumstance I therefore think it is appropriate to close both Pull Requests.

should make sure it works fine

As I wrote before the subject of updating [the python version] and dependencies' versions might favourably be raised in an Issue which states the reasons for the request and furthermore demonstrates that the user has confirmed before opening the Issue that a proposed updated interpreter version and/or dependencies' versions work with the specified software [and will pass tests, if and where appropriate]. This additionally might be a policy consideration as applied Organization wide.

Yes, Torch==2.4.1 and Triton==3.0.0 can run normally in Python 3.10 version. If Deepseek does not need to consider upward compatibility issues, this branch can be closed.

There are: issues#478, issues#325I believe the root cause of multiple issues caused by the Python language environment itself is due to some older libraries not keeping up with the environment after the major version update of Python 3.13, resulting in the failure of the latest Python deployment. This can be addressed through Qinghua Source As an example, Torch==2.6.0 and Triton==3.2.0 are compatible with versions from Python 3.9 to Python 3.13. I have only provided update suggestions based on my own needs, and whether forward compatibility is necessary still needs to be decided by the responsible person.
If possible, I am comparing the update issues of various deepseek functions between different versions to provide evidence:

should make sure it works fine

But this will take time.

[musvaage]

cf: #325

[Triton]... package is not available for windows

cf: #478

我是Mac pro m1

They're both running an unsupported Operating System.

Neither Issue clarified which interpreter version was utilised.

I point out that some users in their Issues are sparing in the information they provide.

A former Bell Labs employee once commented that if you write a program you have to maintain the libraries it uses.

forward compatibility

I'm unqualified to respond on that matter.

---

A Python environment is a context in which you run Python code and includes global, virtual, and conda environments. An environment consists of an interpreter, a library (typically the Python Standard Library), and a set of installed packages. These components together determine valid language constructs and syntax, operating-system functionality that you can access, and packages you can use.

© Microsoft 2025