[BUG] nexa convert fails: missing imports

[iwr-redmond]

Issue Description

Running nexa convert results in:

Error during conversion: name 'GGML_TYPE_BF16' is not defined

This appears to be resolvable by adding the missing types to llama_cpp.py#L106

GGML_TYPE_BF16 = 32
GGML_TYPE_Q4_0_4_4 = 33
GGML_TYPE_Q4_0_4_8 = 34
GGML_TYPE_Q4_0_8_8 = 35

I have obtained these values from line 382.

Steps to Reproduce

1. Install nexaai[convert]==0.0.98 --extra-index-url https://github.nexa.ai/whl/cu124
2. Run nexa convert [some-hf-repo] q4_1 /your/model.gguf
3. Observe the error 😢
4. Add the lines shown above
5. Observe the successful conversion 😄

OS

Ubuntu 22.04

Python Version

3.10

Nexa SDK Version

0.0.98-cu124

GPU (if using one)

NVIDIA RTX 3060

[xsxszab]

Thanks for reporting this issue, A PR has been posted to address it

[keithem]

Thank you for addressing the reported issue and for sharing the PR link. I have reviewed it and agree with the proposed changes. Please proceed with the fix. Your prompt attention to this matter is greatly appreciated. Thank you for your ongoing support.

…

On Wed, Feb 5, 2025 at 2:44 PM Yifei Wang ***@***.***> wrote: Thanks for reporting this issue, A PR <#377> has been posted to address it — Reply to this email directly, view it on GitHub <#372 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BN5OQI73PINDBGRH55NCXTL2OJZ2FAVCNFSM6AAAAABWP3ZMK6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMMZXHE4DGMBYGE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[iwr-redmond]

I've tested the fix in ea46950 and confirmed it works. Thank you!