burn_tensor/tensor/backend/
base.rs

1use alloc::string::String;
2
3use crate::tensor::Element;
4use crate::TensorMetadata;
5use crate::{ops::*, quantization::QTensorPrimitive};
6
7use super::DeviceOps;
8
9/// This trait defines all types and functions needed for a backend to be used with burn.
10///
11/// ## Design
12///
13/// This trait aims to be as unopinionated as possible and allows implementations to define
14/// their own types and patterns. Therefore, there are few pre-defined abstractions baked
15/// into this trait.
16///
17/// Backends must define their own tensor types for each data type: `float`, `int`, and `bool`.
18/// Since we minimize assumptions, we chose to separate these types, as they are used in
19/// different contexts. However, some backends may have a generic tensor type that is used
20/// for all data types.
21///
22/// ### Eager Mode
23///
24/// Because burn supports dynamic graphs, the backend trait is designed around kernel
25/// implementations that can be called without any mutable context or graph. This may not be
26/// ideal for backends that want to configure their computational graphs and execute them
27/// multiple times.
28///
29/// To implement this kind of backend, channels could be used to communicate with a backend
30/// server thread to build the computation graphs and re-execute the ones that are repeated,
31/// with some form of cache. Once that pattern has matured, a graph mode backend trait could
32/// be extracted from it, allowing other backends of the same kind to be quickly integrated
33/// with burn. This pattern could also be used to create an operation fusion trait, which
34/// allows backends to define what kind of graph structures can be fused into one operation.
35///
36/// ### Multi-Threaded
37///
38/// Backend tensor types are all `Clone` + `Send`, which allows them to be safely
39/// sent between threads. It is recommended to wrap tensors with [Arc](alloc::sync::Arc),
40/// which avoids copying the tensor's buffer. Note that it is still possible to mutate and
41/// reuse tensors' buffer without locking; see the next section on the Mutable API.
42///
43/// ### Mutable API
44///
45/// There is no mutable or inplace operation API to implement, but that does not mean that
46/// backends cannot support them. Using [try_unwrap](alloc::sync::Arc::try_unwrap) and
47/// [get_mut](alloc::sync::Arc::get_mut) allows backends to have access to an owned or mutable
48/// reference to their tensor buffer data structure if the tensor is not shared. In that case,
49/// backends can dispatch to their owned inplace operations for better performance.
50///
51/// ## Documentation
52///
53/// Most of the documentation for each function can be found on the user API [tensor struct](crate::Tensor).
54/// For modules, public functions are often created, which can be used by `burn-core` modules.
55pub trait Backend:
56    FloatTensorOps<Self>
57    + BoolTensorOps<Self>
58    + IntTensorOps<Self>
59    + ModuleOps<Self>
60    + ActivationOps<Self>
61    + QTensorOps<Self>
62    + TransactionOps<Self>
63    + Clone
64    + Default
65    + Sized
66    + Send
67    + Sync
68    + core::fmt::Debug
69    + 'static
70{
71    /// Device type.
72    type Device: DeviceOps;
73
74    /// Tensor primitive to be used for all float operations.
75    type FloatTensorPrimitive: TensorMetadata + 'static;
76    /// Default float element type.
77    type FloatElem: Element;
78
79    /// Tensor primitive to be used for all int operations.
80    type IntTensorPrimitive: TensorMetadata + 'static;
81    /// Int element type.
82    type IntElem: Element;
83
84    /// Tensor primitive to be used for all bool operations.
85    type BoolTensorPrimitive: TensorMetadata + 'static;
86    /// Tensor primitive to be used for all bool operations.
87    type BoolElem: Element;
88
89    /// Tensor primitive to be used for all quantized operations.
90    type QuantizedTensorPrimitive: TensorMetadata + QTensorPrimitive + 'static;
91    /// Quantized tensor encoding type.
92    type QuantizedEncoding: Element;
93
94    /// If autodiff is enabled.
95    fn ad_enabled() -> bool {
96        false
97    }
98
99    /// Name of the backend.
100    fn name() -> String;
101
102    /// Seed the backend.
103    fn seed(seed: u64);
104
105    /// Sync the backend, ensure that all computation are finished.
106    fn sync(_device: &Self::Device) {}
107}
108
109/// Trait that allows a backend to support autodiff.
110pub trait AutodiffBackend: Backend {
111    /// The inner backend type.
112    type InnerBackend: Backend<
113        Device = Self::Device,
114        FloatElem = Self::FloatElem,
115        IntElem = Self::IntElem,
116    >;
117
118    /// Gradients type.
119    type Gradients: Send;
120
121    /// Backward pass.
122    ///
123    /// # Arguments
124    ///
125    /// * `tensor` - The tensor is the last node of computational graph where the gradients are computed.
126    ///
127    /// # Returns
128    ///
129    /// The gradients.
130    fn backward(tensor: FloatTensor<Self>) -> Self::Gradients;
131
132    /// Returns the gradients of a tensor.
133    ///
134    /// # Arguments
135    ///
136    /// * `tensor` - The tensor to extract the gradients from.
137    ///
138    /// # Returns
139    ///
140    /// An optional tensor containing the gradient.
141    fn grad(
142        tensor: &FloatTensor<Self>,
143        grads: &Self::Gradients,
144    ) -> Option<FloatTensor<Self::InnerBackend>>;
145
146    /// Pops the gradients of a tensor and returns them.
147    ///
148    /// # Arguments
149    ///
150    /// * `tensor` - The tensor to pop the gradients from.
151    /// * `grads` - The gradients.
152    ///
153    /// # Returns
154    ///
155    /// An optional tensor containing the given gradients.
156    fn grad_remove(
157        tensor: &FloatTensor<Self>,
158        grads: &mut Self::Gradients,
159    ) -> Option<FloatTensor<Self::InnerBackend>>;
160
161    /// Replace the gradients of a tensor with the one provided.
162    ///
163    /// If no gradient existed for the provided tensor, register it.
164    ///
165    /// # Arguments
166    ///
167    /// * `tensor` - The tensor to pop the gradients from.
168    /// * `grads` - The gradients.
169    /// * `grad` - The updated grad tensor.
170    fn grad_replace(
171        tensor: &FloatTensor<Self>,
172        grads: &mut Self::Gradients,
173        grad: FloatTensor<Self::InnerBackend>,
174    );
175
176    /// Returns the tensor with inner backend type.
177    ///
178    /// # Arguments
179    ///
180    /// * `tensor` - The tensor to get the inner backend tensor for.
181    ///
182    /// # Returns
183    ///
184    /// The inner backend tensor.
185    fn inner(tensor: FloatTensor<Self>) -> FloatTensor<Self::InnerBackend>;
186
187    /// Returns the tensor with inner backend type.
188    ///
189    /// # Arguments
190    ///
191    /// * `tensor` - The tensor to get the inner backend tensor for.
192    ///
193    /// # Returns
194    ///
195    /// The inner backend tensor.
196    fn int_inner(tensor: IntTensor<Self>) -> IntTensor<Self::InnerBackend>;
197
198    /// Returns the tensor with inner backend type.
199    ///
200    /// # Arguments
201    ///
202    /// * `tensor` - The tensor to get the inner backend tensor for.
203    ///
204    /// # Returns
205    ///
206    /// The inner backend tensor.
207    fn bool_inner(tensor: BoolTensor<Self>) -> BoolTensor<Self::InnerBackend>;
208
209    /// Returns the tensor with inner backend type.
210    ///
211    /// # Arguments
212    ///
213    /// * `tensor` - The tensor to get the inner backend tensor for.
214    ///
215    /// # Returns
216    ///
217    /// The inner backend tensor.
218    fn q_inner(tensor: QuantizedTensor<Self>) -> QuantizedTensor<Self::InnerBackend>;
219
220    /// Converts the inner backend tensor to the autodiff backend tensor.
221    ///
222    /// # Arguments
223    ///
224    /// * `tensor` - The inner backend tensor to convert.
225    ///
226    ///
227    /// # Returns
228    ///
229    /// The autodiff backend tensor.
230    fn from_inner(tensor: FloatTensor<Self::InnerBackend>) -> FloatTensor<Self>;
231
232    /// Converts the inner backend tensor to the autodiff backend tensor.
233    ///
234    /// # Arguments
235    ///
236    /// * `tensor` - The inner backend tensor to convert.
237    ///
238    ///
239    /// # Returns
240    ///
241    /// The autodiff backend tensor.
242    fn int_from_inner(tensor: IntTensor<Self::InnerBackend>) -> IntTensor<Self>;
243
244    /// Converts the inner backend tensor to the autodiff backend tensor.
245    ///
246    /// # Arguments
247    ///
248    /// * `tensor` - The inner backend tensor to convert.
249    ///
250    ///
251    /// # Returns
252    ///
253    /// The autodiff backend tensor.
254    fn bool_from_inner(tensor: BoolTensor<Self::InnerBackend>) -> BoolTensor<Self>;
255
256    /// Converts the inner backend tensor to the autodiff backend tensor.
257    ///
258    /// # Arguments
259    ///
260    /// * `tensor` - The inner backend tensor to convert.
261    ///
262    ///
263    /// # Returns
264    ///
265    /// The autodiff backend tensor.
266    fn q_from_inner(tensor: QuantizedTensor<Self::InnerBackend>) -> QuantizedTensor<Self>;
267}