huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 18:48:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

Handle GGUF files in tensor-tools. (#558)

Browse Source
This commit is contained in:
Laurent Mazare
2023-08-23 06:32:07 +01:00
committed by GitHub
parent 6a30ecefad
commit 0764741cc4
2 changed files with 30 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
candle-core/examples/tensor-tools.rs
View File

@ -6,6 +6,7 @@ enum Format {
Safetensors,
Npz,
Ggml,
Gguf,
Pth,
Pickle,
}
@ -16,11 +17,12 @@ impl Format {
.extension()
.and_then(|e| e.to_str())
.and_then(|e| match e {
// We don't infer any format for .bin as it can be used for ggml or pytorch.
// We don't infer any format for .bin as it can be used for ggml/gguf or pytorch.
"safetensors" | "safetensor" => Some(Self::Safetensors),
"npz" => Some(Self::Npz),
"pth" | "pt" => Some(Self::Pth),
"ggml" => Some(Self::Ggml),
"gguf" => Some(Self::Gguf),
_ => None,
})
}
@ -121,6 +123,23 @@ fn run_ls(file: &std::path::PathBuf, format: Option<Format>, verbose: bool) -> R
println!("{name}: [{:?}; {:?}]", qtensor.shape(), qtensor.dtype());
}
}
Format::Gguf => {
let mut file = std::fs::File::open(file)?;
let content = candle_core::quantized::gguf_file::Content::read(&mut file)?;
if verbose {
let mut metadata = content.metadata.into_iter().collect::<Vec<_>>();
metadata.sort_by(|a, b| a.0.cmp(&b.0));
println!("metadata entries ({})", metadata.len());
for (key, value) in metadata.iter() {
println!(" {key}: {value:?}");
}
}
let mut tensors = content.tensor_infos.into_iter().collect::<Vec<_>>();
tensors.sort_by(|a, b| a.0.cmp(&b.0));
for (name, info) in tensors.iter() {
println!("{name}: [{:?}; {:?}]", info.shape, info.ggml_dtype);
}
}
}
Ok(())
}