mistralrs_quant/

safetensors.rs

use candle_core::{DType, Device, Error, Result, Shape, Tensor, WithDType};
use candle_nn::var_builder::{Backend, SimpleBackend, VarBuilderArgs};
use float8::F8E4M3;
use regex::Regex;
use safetensors::tensor as st;
use safetensors::tensor::SafeTensors;
use std::collections::HashMap;
use std::path::Path;
use std::sync::Arc;

fn convert_slice<T: WithDType>(data: &[u8], shape: &[usize], device: &Device) -> Result<Tensor> {
    let size_in_bytes = T::DTYPE.size_in_bytes();
    let elem_count = data.len() / size_in_bytes;
    if (data.as_ptr() as usize) % size_in_bytes == 0 {
        // SAFETY This is safe because we just checked that this
        // was correctly aligned.
        let data: &[T] =
            unsafe { std::slice::from_raw_parts(data.as_ptr() as *const T, elem_count) };
        Tensor::from_slice(data, shape, device)
    } else {
        // XXX: We need to specify `T` here, otherwise the compiler will infer u8 because of the following cast
        // Making this vector too small to fit a full f16/f32/f64 weights, resulting in out-of-bounds access
        let mut c: Vec<T> = Vec::with_capacity(elem_count);
        // SAFETY: We just created c, so the allocated memory is necessarily
        // contiguous and non overlapping with the view's data.
        // We're downgrading the `c` pointer from T to u8, which removes alignment
        // constraints.
        unsafe {
            std::ptr::copy_nonoverlapping(data.as_ptr(), c.as_mut_ptr() as *mut u8, data.len());
            c.set_len(elem_count)
        }
        Tensor::from_slice(&c, shape, device)
    }
}

fn convert_slice_with_cast<T: Sized + Copy, U: WithDType, F: Fn(T) -> Result<U>>(
    data: &[u8],
    shape: &[usize],
    device: &Device,
    conv: F,
) -> Result<Tensor> {
    let size_in_bytes = std::mem::size_of::<T>();
    let elem_count = data.len() / size_in_bytes;
    if (data.as_ptr() as usize) % size_in_bytes == 0 {
        // SAFETY This is safe because we just checked that this
        // was correctly aligned.
        let data: &[T] =
            unsafe { std::slice::from_raw_parts(data.as_ptr() as *const T, elem_count) };
        let data = data.iter().map(|t| conv(*t)).collect::<Result<Vec<_>>>()?;
        Tensor::from_vec(data, shape, device)
    } else {
        // XXX: We need to specify `T` here, otherwise the compiler will infer u8 because of the following cast
        // Making this vector too small to fit a full f16/f32/f64 weights, resulting in out-of-bounds access
        let mut c: Vec<T> = Vec::with_capacity(elem_count);
        // SAFETY: We just created c, so the allocated memory is necessarily
        // contiguous and non overlapping with the view's data.
        // We're downgrading the `c` pointer from T to u8, which removes alignment
        // constraints.
        unsafe {
            std::ptr::copy_nonoverlapping(data.as_ptr(), c.as_mut_ptr() as *mut u8, data.len());
            c.set_len(elem_count)
        }
        let c = c.into_iter().map(conv).collect::<Result<Vec<_>>>()?;
        Tensor::from_vec(c, shape, device)
    }
}

fn convert_with_cast_<T: Sized + Copy, U: WithDType, F: Fn(T) -> Result<U>>(
    view: &st::TensorView<'_>,
    device: &Device,
    conv: F,
) -> Result<Tensor> {
    convert_slice_with_cast::<T, U, F>(view.data(), view.shape(), device, conv)
}

fn convert_<T: WithDType>(view: &st::TensorView<'_>, device: &Device) -> Result<Tensor> {
    convert_slice::<T>(view.data(), view.shape(), device)
}

pub trait Load {
    fn load(&self, device: &Device, dtype: Option<DType>) -> Result<Tensor>;
}

impl Load for st::TensorView<'_> {
    fn load(&self, device: &Device, dtype: Option<DType>) -> Result<Tensor> {
        convert(self, device, dtype)
    }
}

fn convert(
    view: &st::TensorView<'_>,
    device: &Device,
    cast_dtype: Option<DType>,
) -> Result<Tensor> {
    match (view.dtype(), cast_dtype) {
        (st::Dtype::U8, _) => convert_::<u8>(view, device),
        (st::Dtype::U16, _) => {
            let conv = |x| Ok(u32::from(x));
            convert_with_cast_::<u16, u32, _>(view, device, conv)
        }
        (st::Dtype::U32, _) => convert_::<u32>(view, device),
        (st::Dtype::I16, _) => convert_::<i16>(view, device),
        (st::Dtype::I32, _) => convert_::<i32>(view, device),
        (st::Dtype::I64, _) => convert_::<i64>(view, device),
        (st::Dtype::BF16, None | Some(DType::BF16)) => convert_::<half::bf16>(view, device),
        (st::Dtype::F16, None | Some(DType::F16)) => convert_::<half::f16>(view, device),
        (st::Dtype::F32, _) => convert_::<f32>(view, device),
        (st::Dtype::F64, _) => convert_::<f64>(view, device),
        (st::Dtype::F8_E4M3, _) => convert_::<F8E4M3>(view, device),

        (st::Dtype::BF16, Some(DType::F16)) => {
            let conv = |x: half::bf16| Ok(half::f16::from_f32(x.to_f32()));
            convert_with_cast_::<half::bf16, half::f16, _>(view, device, conv)
        }
        (st::Dtype::BF16, Some(DType::F32)) => {
            let conv = |x: half::bf16| Ok(x.to_f32());
            convert_with_cast_::<half::bf16, f32, _>(view, device, conv)
        }
        (st::Dtype::F16, Some(DType::BF16)) => {
            let conv = |x: half::f16| Ok(half::bf16::from_f32(x.to_f32()));
            convert_with_cast_::<half::f16, half::bf16, _>(view, device, conv)
        }
        (st::Dtype::F16, Some(DType::F32)) => {
            let conv = |x: half::f16| Ok(x.to_f32());
            convert_with_cast_::<half::f16, f32, _>(view, device, conv)
        }
        (dtype, _) => Err(Error::UnsupportedSafeTensorDtype(dtype)),
    }
}

#[derive(yoke::Yokeable)]
struct SafeTensors_<'a>(SafeTensors<'a>);

pub struct MmapedSafetensors {
    safetensors: Vec<yoke::Yoke<SafeTensors_<'static>, memmap2::Mmap>>,
    routing: Option<HashMap<String, usize>>,
}

impl MmapedSafetensors {
    /// Creates a wrapper around a memory mapped file and deserialize the safetensors header.
    ///
    /// # Safety
    ///
    /// The unsafe is inherited from [`memmap2::MmapOptions`].
    pub unsafe fn new<P: AsRef<Path>>(p: P) -> Result<Self> {
        let p = p.as_ref();
        let file = std::fs::File::open(p).map_err(|e| Error::from(e).with_path(p))?;
        let file = memmap2::MmapOptions::new()
            .map(&file)
            .map_err(|e| Error::from(e).with_path(p))?;
        let safetensors = yoke::Yoke::<SafeTensors_<'static>, memmap2::Mmap>::try_attach_to_cart(
            file,
            |data: &[u8]| {
                let st = safetensors::SafeTensors::deserialize(data)
                    .map_err(|e| Error::from(e).with_path(p))?;
                Ok::<_, Error>(SafeTensors_(st))
            },
        )?;
        Ok(Self {
            safetensors: vec![safetensors],
            routing: None,
        })
    }

    /// Creates a wrapper around multiple memory mapped file and deserialize the safetensors headers.
    ///
    /// If a tensor name appears in multiple files, the last entry is returned.
    ///
    /// # Safety
    ///
    /// The unsafe is inherited from [`memmap2::MmapOptions`].
    pub unsafe fn multi<P: AsRef<Path>>(paths: &[P]) -> Result<Self> {
        let mut routing = HashMap::new();
        let mut safetensors = vec![];
        for (index, p) in paths.iter().enumerate() {
            let p = p.as_ref();
            let file = std::fs::File::open(p).map_err(|e| Error::from(e).with_path(p))?;
            let file = memmap2::MmapOptions::new()
                .map(&file)
                .map_err(|e| Error::from(e).with_path(p))?;
            let data = yoke::Yoke::<SafeTensors_<'static>, memmap2::Mmap>::try_attach_to_cart(
                file,
                |data: &[u8]| {
                    let st = safetensors::SafeTensors::deserialize(data)
                        .map_err(|e| Error::from(e).with_path(p))?;
                    Ok::<_, Error>(SafeTensors_(st))
                },
            )?;
            for k in data.get().0.names() {
                routing.insert(k.to_string(), index);
            }
            safetensors.push(data)
        }
        Ok(Self {
            safetensors,
            routing: Some(routing),
        })
    }

    pub fn load(&self, name: &str, dev: &Device, dtype: Option<DType>) -> Result<Tensor> {
        self.get(name)?.load(dev, dtype)
    }

    pub fn tensors(&self) -> Vec<(String, st::TensorView<'_>)> {
        let mut tensors = vec![];
        for safetensors in self.safetensors.iter() {
            tensors.push(safetensors.get().0.tensors())
        }
        tensors.into_iter().flatten().collect()
    }

    pub fn get(&self, name: &str) -> Result<st::TensorView<'_>> {
        let index = match &self.routing {
            None => 0,
            Some(routing) => {
                let index = routing.get(name).ok_or_else(|| {
                    Error::CannotFindTensor {
                        path: name.to_string(),
                    }
                    .bt()
                })?;
                *index
            }
        };
        Ok(self.safetensors[index].get().0.tensor(name)?)
    }
}

impl SimpleBackend for MmapedSafetensors {
    fn get(
        &self,
        s: Shape,
        name: &str,
        _: candle_nn::Init,
        dtype: DType,
        dev: &Device,
    ) -> Result<Tensor> {
        let tensor = self.get_unchecked(name, dtype, dev)?;
        if tensor.shape() != &s {
            Err(candle_core::Error::UnexpectedShape {
                msg: format!("shape mismatch for {name}"),
                expected: s,
                got: tensor.shape().clone(),
            }
            .bt())?
        }
        Ok(tensor)
    }

    fn get_unchecked(&self, name: &str, dtype: DType, dev: &Device) -> Result<Tensor> {
        self.load(name, dev, Some(dtype))
    }

    fn contains_tensor(&self, name: &str) -> bool {
        self.get(name).is_ok()
    }
}

pub enum ShardedSafeTensors<'a> {
    Sharded {
        b: MmapedSafetensors,
        make_dummy_regexes: Option<Arc<Vec<Regex>>>,
    },
    SimpleBackend(Box<dyn SimpleBackend + 'a>),
}

pub type ShardedVarBuilder<'a> = VarBuilderArgs<'a, ShardedSafeTensors<'a>>;

impl<'a> ShardedSafeTensors<'a> {
    /// Initializes a `VarBuilder` that retrieves tensors stored in a collection of safetensors
    /// files and make them usable in a sharded way.
    ///
    /// # Safety
    ///
    /// The unsafe is inherited from [`memmap2::MmapOptions`].
    pub unsafe fn sharded<P: AsRef<std::path::Path>>(
        paths: &[P],
        dtype: DType,
        dev: &Device,
        make_dummy_regexes: Option<Arc<Vec<Regex>>>,
    ) -> Result<ShardedVarBuilder<'a>> {
        let tensors = MmapedSafetensors::multi(paths)?;
        let backend = ShardedSafeTensors::Sharded {
            b: tensors,
            make_dummy_regexes,
        };
        Ok(VarBuilderArgs::new_with_args(backend, dtype, dev))
    }

    pub fn wrap(
        backend: Box<dyn SimpleBackend + 'a>,
        dtype: DType,
        dev: Device,
    ) -> ShardedVarBuilder<'a> {
        VarBuilderArgs::new_with_args(Self::SimpleBackend(backend), dtype, &dev)
    }
}

#[derive(Debug, Clone, Copy, Eq, PartialEq)]
pub struct Shard {
    pub dim: usize,
    pub rank: usize,
    pub world_size: usize,
}

impl Default for Shard {
    fn default() -> Self {
        Self {
            dim: 0,
            rank: 0,
            world_size: 1,
        }
    }
}

/// Get part of a tensor, typically used to do Tensor Parallelism sharding.
///
/// If the tensor is of size (1024, 1024).
///
/// `dim` corresponds to the dimension to slice into
/// `rank` is the rank of the current process
/// `world_size` is the total number of ranks in the process group
///
/// `get_sharded("tensor", 0, 0, 2)` means `tensor.i((..512))`
/// `get_sharded("tensor", 0, 1, 2)` means `tensor.i((512..))`
/// `get_sharded("tensor", 1, 0, 2)` means `tensor.i((.., ..512))`
impl Backend for ShardedSafeTensors<'_> {
    type Hints = Shard;

    fn get(
        &self,
        target_shape: Shape,
        path: &str,
        h: Self::Hints,
        dtype: DType,
        dev: &Device,
    ) -> Result<Tensor> {
        if h.world_size == 1 {
            // There is no sharding to be applied here so we use the default backend to speed
            // things up.
            match self {
                Self::Sharded {
                    b,
                    make_dummy_regexes,
                } => {
                    if let Some(make_dummy_regexes) = make_dummy_regexes {
                        if make_dummy_regexes.iter().any(|x| x.is_match(path)) {
                            return Err(Error::CannotFindTensor {
                                path: path.to_string(),
                            });
                        }
                    }
                    return SimpleBackend::get(
                        b,
                        target_shape,
                        path,
                        Default::default(),
                        dtype,
                        dev,
                    );
                }
                Self::SimpleBackend(b) => {
                    return SimpleBackend::get(
                        b.as_ref(),
                        target_shape,
                        path,
                        Default::default(),
                        dtype,
                        dev,
                    )
                }
            }
        }

        let Shard {
            dim,
            rank,
            world_size,
        } = h;
        match self {
            Self::Sharded {
                b,
                make_dummy_regexes,
            } => {
                use safetensors::slice::IndexOp;

                if let Some(make_dummy_regexes) = make_dummy_regexes {
                    if make_dummy_regexes.iter().any(|x| x.is_match(path)) {
                        return Err(Error::CannotFindTensor {
                            path: path.to_string(),
                        });
                    }
                }

                let view = b.get(path)?;
                let view_dtype = view.dtype();
                let mut shape = view.shape().to_vec();
                let size = shape[dim];

                if size % world_size != 0 {
                    return Err(Error::ShapeMismatchSplit {
                        shape: shape.into(),
                        dim,
                        n_parts: world_size,
                    });
                }
                let block_size = size / world_size;
                let start = rank * block_size;
                let stop = (rank + 1) * block_size;

                // Everything is expressed in tensor dimension
                // bytes offsets is handled automatically for safetensors.

                let iterator = if dim == 0 {
                    view.slice(start..stop).map_err(|_| {
                        Error::Msg(format!(
                    "Cannot slice tensor {path} ({shape:?} along dim {dim} with {start}..{stop}"
                ))
                    })?
                } else if dim == 1 {
                    view.slice((.., start..stop)).map_err(|_| {
                        Error::Msg(format!(
                    "Cannot slice tensor {path} ({shape:?} along dim {dim} with {start}..{stop}"
                ))
                    })?
                } else {
                    candle_core::bail!("Got sharded on dimensions != 0 or 1")
                };

                shape[dim] = block_size;

                let view_dtype: DType = view_dtype.try_into()?;
                let raw: Vec<u8> = iterator.into_iter().flatten().cloned().collect();
                Tensor::from_raw_buffer(&raw, view_dtype, &shape, dev)?.to_dtype(dtype)
            }
            Self::SimpleBackend(b) => {
                use candle_core::IndexOp;
                let tensor = b.get(target_shape, path, Default::default(), dtype, dev)?;

                let size = tensor.dim(dim)?;
                let shape = tensor.dims().to_vec();

                if size % world_size != 0 {
                    return Err(Error::ShapeMismatchSplit {
                        shape: shape.into(),
                        dim,
                        n_parts: world_size,
                    });
                }
                let block_size = size / world_size;
                let start = rank * block_size;
                let stop = (rank + 1) * block_size;

                if dim == 0 {
                    tensor.i((start..stop, ..))
                } else if dim == 1 {
                    tensor.i((.., start..stop))
                } else {
                    candle_core::bail!("Got sharded on dimensions != 0 or 1")
                }
            }
        }
    }

    fn get_unchecked(&self, name: &str, dtype: DType, dev: &Device) -> Result<Tensor> {
        match self {
            Self::Sharded {
                b,
                make_dummy_regexes,
            } => {
                if let Some(make_dummy_regexes) = make_dummy_regexes {
                    if make_dummy_regexes.iter().any(|x| x.is_match(name)) {
                        return Err(Error::CannotFindTensor {
                            path: name.to_string(),
                        });
                    }
                }
                <MmapedSafetensors as SimpleBackend>::get_unchecked(b, name, dtype, dev)
            }
            Self::SimpleBackend(b) => b.as_ref().get_unchecked(name, dtype, dev),
        }
    }

    fn contains_tensor(&self, name: &str) -> bool {
        match self {
            Self::Sharded {
                b,
                make_dummy_regexes,
            } => {
                if let Some(make_dummy_regexes) = make_dummy_regexes {
                    if make_dummy_regexes.iter().any(|x| x.is_match(name)) {
                        return false;
                    }
                }
                b.get(name).is_ok()
            }
            Self::SimpleBackend(b) => b.as_ref().contains_tensor(name),
        }
    }
}