Impossible to install on windows

[secretlycarl]

EDIT: was able to get it running. To do that,

I skipped the [.train] and flash attn commands

I had to set 8bit CPU offload in some config file

And now it works

Maybe the readme should be updated to clarify that the commands to install training dependencies arent needed

---

Hi, I'm on windows and trying to run this through WSL/conda. I previously got this version of LLaVa installed in the same environment and it works. But for this video implementation, when I run

pip install -e ".[train]"

I get

...
  Downloading deepspeed-0.9.5.tar.gz (809 kB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 809.9/809.9 kB 6.4 MB/s eta 0:00:00
  Preparing metadata (setup.py) ... error
  error: subprocess-exited-with-error

  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> [15 lines of output]
      test.c
      LINK : fatal error LNK1181: cannot open input file 'aio.lib'
      Traceback (most recent call last):
        File "<string>", line 2, in <module>
        File "<pip-setuptools-caller>", line 34, in <module>
        File "C:\Users\wmwal\AppData\Local\Temp\pip-install-ek_3t9mt\deepspeed_ddf8566e4bda48f28faa3b67d7e7bb99\setup.py", line 163, in <module>
          abort(f"Unable to pre-compile {op_name}")
        File "C:\Users\wmwal\AppData\Local\Temp\pip-install-ek_3t9mt\deepspeed_ddf8566e4bda48f28faa3b67d7e7bb99\setup.py", line 51, in abort
          assert False, msg
      AssertionError: Unable to pre-compile async_io
      DS_BUILD_OPS=1
       [WARNING]  async_io requires the dev libaio .so object and headers but these were not found.
       [WARNING]  If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
       [WARNING]  One can disable async_io with DS_BUILD_AIO=0
       [ERROR]  Unable to pre-compile async_io
      [end of output]

I found that there's some issue with deepspeed and Windows. This guide for deepspeed says only 0.8.3 works on windows, while the training reqs for this require 0.9.5. I'm not doing any training. Is there a way around this?

Any help is appreciated. I'm trying to incorporate Video-LLaVa into a video sorting script, where each video has the same question associated with it, and the videos are sorted by the model's answers.