turboderp / exllamav2

A fast inference library for running LLMs locally on modern consumer-class GPUs
MIT License
3.69k stars 283 forks source link

[BUG] AttributeError: module 'exllamav2_ext' has no attribute 'safetensors_free_pinned_buffer' #648

Closed Katehuuh closed 1 month ago

Katehuuh commented 1 month ago

OS

Windows

GPU Library

CUDA 11.8

Python version

3.10

Pytorch version

2.3.1

Model

No response

Describe the bug

>pip install https://github.com/turboderp/exllamav2/releases/download/v0.2.3/exllamav2-0.2.3+cu118.torch2.3.1-cp310-cp310-win_amd64.whl
(venv) C:\exllamav2>python convert.py -i "C:\exllamav2\Qwen2.5-32B-Instruct-abliterated-pass2" -o "C:\myTempQ" -cf "C:\Qwen2.5-32B-Instruct-abliterated-pass2-4.7bpw-h8-exl2" -b 4.7 -hb 8
 -- Resuming job
 !! Note: Overriding options with settings from existing job
 -- Input: C:\exllamav2\Qwen2.5-32B-Instruct-abliterated-pass2
 -- Output: C:\myTempQ
 -- Using default calibration dataset
 -- Target bits per weight: 4.125 (decoder), 6 (head)
 -- Max shard size: 8192 MB
 -- Full model will be compiled to: C:\Qwen2.5-32B-Instruct-abliterated-pass2-4.7bpw-h8-exl2
Traceback (most recent call last):
  File "C:\exllamav2\convert.py", line 1, in <module>
    import exllamav2.conversion.convert_exl2
  File "C:\exllamav2\exllamav2\conversion\convert_exl2.py", line 209, in <module>
    model.load(lazy = True)
  File "C:\exllamav2\exllamav2\model.py", line 256, in load
    for item in f:
  File "C:\exllamav2\exllamav2\model.py", line 294, in load_gen
    cleanup_stfiles()
  File "C:\exllamav2\exllamav2\fasttensors.py", line 40, in cleanup_stfiles
    ext_c.safetensors_free_pinned_buffer()
AttributeError: module 'exllamav2_ext' has no attribute 'safetensors_free_pinned_buffer'

Reproduction steps

Previous version run fine like pip install https://github.com/turboderp/exllamav2/releases/download/v0.2.2/exllamav2-0.2.2+cu118.torch2.3.1-cp310-cp310-win_amd64.whl

Expected behavior

.

Logs

No response

Additional context

No response

Acknowledgements

turboderp commented 1 month ago

It looks like you updated the Python library but you're still running from an old version of the repo. So do a git pull or clone the repo again.