]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commitdiff
requirements : fix PyTorch uint64 compatibility (#15134)
authorDaniel Bevenius <redacted>
Thu, 7 Aug 2025 03:31:48 +0000 (05:31 +0200)
committerGitHub <redacted>
Thu, 7 Aug 2025 03:31:48 +0000 (05:31 +0200)
This commit addresses an issue with the convert_hf_to_gguf script
which is currently failing with:
```console
AttributeError: module 'torch' has no attribute 'uint64'
```

This occurred because safetensors expects torch.uint64 to be available
in the public API, but PyTorch 2.2.x only provides limited support for
unsigned types beyond uint8 it seems. The torch.uint64 dtype exists but
is not exposed in the standard torch namespace
(see pytorch/pytorch#58734).

PyTorch 2.4.0 properly exposes torch.uint64 in the public API, resolving
the compatibility issue with safetensors. This also required torchvision
to updated to =0.19.0 for compatibility.

Refs: https://huggingface.co/spaces/ggml-org/gguf-my-repo/discussions/186#68938de803e47d990aa087fb
Refs: https://github.com/pytorch/pytorch/issues/58734

requirements/requirements-convert_hf_to_gguf.txt
tools/mtmd/requirements.txt

index fd21ec479541f4581f19f789590f87b430a1c526..766745f42f7ea14224c57c4eec1e882936dd96dc 100644 (file)
@@ -2,7 +2,7 @@ mistral-common>=1.8.3
 
 -r ./requirements-convert_legacy_llama.txt
 --extra-index-url https://download.pytorch.org/whl/cpu
-torch~=2.2.1; platform_machine != "s390x"
+torch~=2.4.0; platform_machine != "s390x"
 
 # torch s390x packages can only be found from nightly builds
 --extra-index-url https://download.pytorch.org/whl/nightly
index ad069f774566fdc447e3b354fae63b3d362c9673..a9d788f2653ad012c9f1777b619abdb1428d91f2 100644 (file)
@@ -1,5 +1,5 @@
 -r ../../requirements/requirements-convert_legacy_llama.txt
 --extra-index-url https://download.pytorch.org/whl/cpu
 pillow~=11.3.0
-torch~=2.2.1
-torchvision~=0.17.1
+torch~=2.4.0
+torchvision~=0.19.1