huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-19 03:54:56 +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

12
candle-core/src/quantized/gguf_file.rs
View File

@ -7,7 +7,7 @@ use crate::Result;
use byteorder::{LittleEndian, ReadBytesExt};
use std::collections::HashMap;
pub const DEFAULT_ALIGNMENT: usize = 32;
pub const DEFAULT_ALIGNMENT: u64 = 32;
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum Magic {
@ -208,7 +208,15 @@ impl Content {
);
}
let position = reader.stream_position()?;
let alignment = DEFAULT_ALIGNMENT as u64;
let alignment = match metadata.get("general.alignment") {
Some(Value::U8(v)) => *v as u64,
Some(Value::U16(v)) => *v as u64,
Some(Value::U32(v)) => *v as u64,
Some(Value::I8(v)) if *v >= 0 => *v as u64,
Some(Value::I16(v)) if *v >= 0 => *v as u64,
Some(Value::I32(v)) if *v >= 0 => *v as u64,
_ => DEFAULT_ALIGNMENT,
};
let tensor_data_offset = (position + alignment - 1) / alignment * alignment;
Ok(Self {
magic,