updated tch-rs to 0.13.0 (guillaume-be#380)

* updated tch-rs to 0.13.0 find replaced of_slice to from_slice as per https://github.com/LaurentMazare/tch-rs/blob/008fff6cc0a91ecff274d1549c2dff25b8ca57a4/CHANGELOG.md * fixed formatting * Add download feature and update CI * add build script, update CI * updated chanelog, readme, convert script * fixed wrong position for build script * added libtorch download to dependencies download test script * args reordering --------- Co-authored-by: josephhajduk <[email protected]> Co-authored-by: Guillaume Becquin <[email protected]>
ShabbirHasan1 · May 21, 2023 · 2bff63b · 2bff63b
1 parent 5f9500c
commit 2bff63b
Show file tree

Hide file tree

Showing 50 changed files with 178 additions and 128 deletions.
diff --git a/.github/workflows/continuous-integration.yml b/.github/workflows/continuous-integration.yml
@@ -20,6 +20,7 @@ jobs:
       - uses: actions-rs/cargo@v1
         with:
           command: build
+          args: --features download-libtorch
 
   build-no-defaults:
     name: Build no defaults
@@ -34,7 +35,7 @@ jobs:
       - uses: actions-rs/cargo@v1
         with:
           command: build
-          args: --no-default-features
+          args: --no-default-features --features download-libtorch
 
   build-windows:
     name: Build Windows
@@ -49,6 +50,7 @@ jobs:
       - uses: actions-rs/cargo@v1
         with:
           command: build
+          args: --features download-libtorch
 
   build-mac-os:
     name: Build macOS
@@ -63,6 +65,7 @@ jobs:
       - uses: actions-rs/cargo@v1
         with:
           command: build
+          args: --features download-libtorch
 
   test-batch-0:
     name: Integration tests (batch 0)
@@ -89,6 +92,7 @@ jobs:
             --test fnet
             --test deberta
             --test deberta_v2
+            --features download-libtorch
 
   test-batch-1:
     name: Integration tests (batch 1)
@@ -114,6 +118,7 @@ jobs:
             --test longformer
             --test pegasus
             --test gpt_neo
+            --features download-libtorch
 
   test-batch-2:
     name: Integration tests (batch 2)
@@ -133,6 +138,7 @@ jobs:
             --test longt5
             --test gpt_j
             --test nllb
+            --features download-libtorch
 
   convert-model:
     name: Model conversion test

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -11,7 +11,7 @@ All notable changes to this project will be documented in this file. The format
 ## Changed
 - Bumped the tokenizers dependency from 7.x to 8.x, exposing additional options for special token mapping and adding the NLLBTokenizer.
 - (BREAKING) Simplified the generation traits (removal of LMHeadModel and elimination of unnecessary specification for LanguageGenerator)
-- Upgraded to `torch` 2.0 (via `tch` 0.12.0).
+- (BREAKING) Upgraded to `torch` 2.0 (via `tch` 0.13.0). The process to automatically download the dependencies have changed, it must now be enabled via the `download-libtorch` feature flag.
 
 ## Fixed
 - MIN/MAX computation for float-like (was set to infinity instead of min/max)

diff --git a/Cargo.toml b/Cargo.toml
@@ -8,6 +8,7 @@ repository = "https://github.com/guillaume-be/rust-bert"
 documentation = "https://docs.rs/rust-bert"
 license = "Apache-2.0"
 readme = "README.md"
+build = "build.rs"
 keywords = [
     "nlp",
     "deep-learning",
@@ -64,13 +65,14 @@ default = ["remote"]
 doc-only = ["tch/doc-only"]
 all-tests = []
 remote = ["cached-path", "dirs", "lazy_static"]
+download-libtorch = ["torch-sys/download-libtorch"]
 
 [package.metadata.docs.rs]
 features = ["doc-only"]
 
 [dependencies]
 rust_tokenizers = "8.1"
-tch = "0.12.0"
+tch = "0.13.0"
 serde_json = "1"
 serde = { version = "1", features = ["derive"] }
 ordered-float = "3"
@@ -88,6 +90,6 @@ anyhow = "1"
 csv = "1"
 criterion = "0.4"
 tokio = { version = "1.24", features = ["sync", "rt-multi-thread", "macros"] }
-torch-sys = "0.12.0"
+torch-sys = "0.13.0"
 tempfile = "3"
 itertools = "0.10"
diff --git a/README.md b/README.md
@@ -95,7 +95,7 @@ $Env:Path += ";X:\path\to\libtorch\lib"
 
 ### Automatic installation
 
-Alternatively, you can let the `build` script automatically download the `libtorch` library for you.
+Alternatively, you can let the `build` script automatically download the `libtorch` library for you. The `download-libtorch` feature flag needs to be enabled.
 The CPU version of libtorch will be downloaded by default. To download a CUDA version, please set the environment variable `TORCH_CUDA_VERSION` to `cu118`.
 Note that the libtorch library is large (order of several GBs for the CUDA-enabled version) and the first build may therefore take several minutes to complete.
 

diff --git a/build.rs b/build.rs
@@ -0,0 +1,29 @@
+// Copyright 2023 Laurent Mazare
+// https://github.com/LaurentMazare/diffusers-rs/blob/main/build.rs
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//     http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+fn main() {
+    let os = std::env::var("CARGO_CFG_TARGET_OS").expect("Unable to get TARGET_OS");
+    match os.as_str() {
+        "linux" | "windows" => {
+            if let Some(lib_path) = std::env::var_os("DEP_TCH_LIBTORCH_LIB") {
+                println!(
+                    "cargo:rustc-link-arg=-Wl,-rpath={}",
+                    lib_path.to_string_lossy()
+                );
+            }
+            println!("cargo:rustc-link-arg=-Wl,--no-as-needed");
+            println!("cargo:rustc-link-arg=-Wl,--copy-dt-needed-entries");
+            println!("cargo:rustc-link-arg=-ltorch");
+        }
+        _ => {}
+    }
+}
diff --git a/examples/natural_language_inference_deberta.rs b/examples/natural_language_inference_deberta.rs
@@ -63,7 +63,7 @@ fn main() -> anyhow::Result<()> {
             input.extend(vec![0; max_len - input.len()]);
             input
         })
-        .map(|input| Tensor::of_slice(&(input)))
+        .map(|input| Tensor::from_slice(&(input)))
         .collect::<Vec<_>>();
     let input_tensor = Tensor::stack(tokenized_input.as_slice(), 0).to(device);
 

diff --git a/src/bart/bart_model.rs b/src/bart/bart_model.rs
@@ -1032,7 +1032,7 @@ impl BartGenerator {
         let impossible_tokens: Vec<i64> = (0..self.get_vocab_size())
             .filter(|pos| !token_ids.contains(pos))
             .collect();
-        let impossible_tokens = Tensor::of_slice(&impossible_tokens).to_device(scores.device());
+        let impossible_tokens = Tensor::from_slice(&impossible_tokens).to_device(scores.device());
         let _ = scores.index_fill_(1, &impossible_tokens, f64::NEG_INFINITY);
     }
 }
@@ -1207,7 +1207,7 @@ impl PrivateLanguageGenerator for BartGenerator {
                 input.extend(temp);
                 input
             })
-            .map(|tokens| Tensor::of_slice(&tokens).to(self.get_var_store().device()))
+            .map(|tokens| Tensor::from_slice(&tokens).to(self.get_var_store().device()))
             .collect::<Vec<Tensor>>();
 
         Tensor::stack(&token_ids, 0)

diff --git a/src/distilbert/embeddings.rs b/src/distilbert/embeddings.rs
@@ -42,7 +42,7 @@ where
                 );
             }
         }
-        let temp_vec = Tensor::of_slice(&temp_vec);
+        let temp_vec = Tensor::from_slice(&temp_vec);
         sinusoidal_embedding.push(temp_vec);
     }
     let sinusoidal_embedding = Tensor::stack(&sinusoidal_embedding, 0)

diff --git a/src/gpt_neo/attention.rs b/src/gpt_neo/attention.rs
@@ -173,7 +173,7 @@ impl GptNeoSelfAttention {
 
         let mut attention_weights = attention_weights.where_self(
             causal_mask,
-            &Tensor::of_slice(&[-1e9f32]).to_device(attention_weights.device()),
+            &Tensor::from_slice(&[-1e9f32]).to_device(attention_weights.device()),
         );
         if let Some(attention_mask_value) = attention_mask {
             attention_weights = attention_weights + attention_mask_value;

diff --git a/src/lib.rs b/src/lib.rs
@@ -105,7 +105,7 @@
 //!
 //! ### Automatic installation
 //!
-//! Alternatively, you can let the `build` script automatically download the `libtorch` library for you.
+//! Alternatively, you can let the `build` script automatically download the `libtorch` library for you. The `download-libtorch` feature flag needs to be enabled.
 //! The CPU version of libtorch will be downloaded by default. To download a CUDA version, please set the environment variable `TORCH_CUDA_VERSION` to `cu118`.
 //! Note that the libtorch library is large (order of several GBs for the CUDA-enabled version) and the first build may therefore take several minutes to complete.
 //!

diff --git a/src/longt5/longt5_model.rs b/src/longt5/longt5_model.rs
@@ -775,7 +775,7 @@ impl PrivateLanguageGenerator for LongT5Generator {
                 input.extend(temp);
                 input
             })
-            .map(|tokens| Tensor::of_slice(&tokens).to(self.get_var_store().device()))
+            .map(|tokens| Tensor::from_slice(&tokens).to(self.get_var_store().device()))
             .collect::<Vec<Tensor>>();
 
         Tensor::stack(&token_ids, 0)

diff --git a/src/m2m_100/m2m_100_model.rs b/src/m2m_100/m2m_100_model.rs
@@ -578,7 +578,7 @@ impl M2M100Generator {
         let impossible_tokens: Vec<i64> = (0..self.get_vocab_size())
             .filter(|pos| !token_ids.contains(pos))
             .collect();
-        let impossible_tokens = Tensor::of_slice(&impossible_tokens).to_device(scores.device());
+        let impossible_tokens = Tensor::from_slice(&impossible_tokens).to_device(scores.device());
         let _ = scores.index_fill_(1, &impossible_tokens, f64::NEG_INFINITY);
     }
 }
@@ -750,7 +750,7 @@ impl PrivateLanguageGenerator for M2M100Generator {
                 input.extend(temp);
                 input
             })
-            .map(|tokens| Tensor::of_slice(&tokens).to(self.get_var_store().device()))
+            .map(|tokens| Tensor::from_slice(&tokens).to(self.get_var_store().device()))
             .collect::<Vec<Tensor>>();
 
         Tensor::stack(&token_ids, 0)

diff --git a/src/marian/marian_model.rs b/src/marian/marian_model.rs
@@ -799,7 +799,7 @@ impl MarianGenerator {
         let impossible_tokens: Vec<i64> = (0..self.get_vocab_size())
             .filter(|pos| !token_ids.contains(pos))
             .collect();
-        let impossible_tokens = Tensor::of_slice(&impossible_tokens).to_device(scores.device());
+        let impossible_tokens = Tensor::from_slice(&impossible_tokens).to_device(scores.device());
         let _ = scores.index_fill_(1, &impossible_tokens, f64::NEG_INFINITY);
     }
 }
@@ -895,7 +895,7 @@ impl PrivateLanguageGenerator for MarianGenerator {
     ) {
         let _ = scores.index_fill_(
             1,
-            &Tensor::of_slice(&[self.get_pad_id().unwrap()])
+            &Tensor::from_slice(&[self.get_pad_id().unwrap()])
                 .to_kind(Kind::Int64)
                 .to_device(scores.device()),
             f64::NEG_INFINITY,
@@ -975,7 +975,7 @@ impl PrivateLanguageGenerator for MarianGenerator {
                 input.extend(temp);
                 input
             })
-            .map(|tokens| Tensor::of_slice(&tokens).to(self.get_var_store().device()))
+            .map(|tokens| Tensor::from_slice(&tokens).to(self.get_var_store().device()))
             .collect::<Vec<Tensor>>();
 
         Tensor::stack(&token_ids, 0)

diff --git a/src/mbart/mbart_model.rs b/src/mbart/mbart_model.rs
@@ -831,7 +831,7 @@ impl MBartGenerator {
         let impossible_tokens: Vec<i64> = (0..self.get_vocab_size())
             .filter(|pos| !token_ids.contains(pos))
             .collect();
-        let impossible_tokens = Tensor::of_slice(&impossible_tokens).to_device(scores.device());
+        let impossible_tokens = Tensor::from_slice(&impossible_tokens).to_device(scores.device());
         let _ = scores.index_fill_(1, &impossible_tokens, f64::NEG_INFINITY);
     }
 }
@@ -1004,7 +1004,7 @@ impl PrivateLanguageGenerator for MBartGenerator {
                 input.extend(temp);
                 input
             })
-            .map(|tokens| Tensor::of_slice(&tokens).to(self.get_var_store().device()))
+            .map(|tokens| Tensor::from_slice(&tokens).to(self.get_var_store().device()))
             .collect::<Vec<Tensor>>();
 
         Tensor::stack(&token_ids, 0)

diff --git a/src/pegasus/embeddings.rs b/src/pegasus/embeddings.rs
@@ -66,7 +66,7 @@ impl SinusoidalPositionalEmbedding {
                     temp_vec.push(base_value.cos());
                 }
             }
-            let temp_vec = Tensor::of_slice(&temp_vec);
+            let temp_vec = Tensor::from_slice(&temp_vec);
 
             sinusoidal_embedding.push(temp_vec);
         }

diff --git a/src/pegasus/pegasus_model.rs b/src/pegasus/pegasus_model.rs
@@ -540,7 +540,7 @@ impl PegasusConditionalGenerator {
         let impossible_tokens: Vec<i64> = (0..self.get_vocab_size())
             .filter(|pos| !token_ids.contains(pos))
             .collect();
-        let impossible_tokens = Tensor::of_slice(&impossible_tokens).to_device(scores.device());
+        let impossible_tokens = Tensor::from_slice(&impossible_tokens).to_device(scores.device());
         let _ = scores.index_fill_(
             1,
             &impossible_tokens,
@@ -716,7 +716,7 @@ impl PrivateLanguageGenerator for PegasusConditionalGenerator {
                 input.extend(temp);
                 input
             })
-            .map(|tokens| Tensor::of_slice(&tokens).to(self.get_var_store().device()))
+            .map(|tokens| Tensor::from_slice(&tokens).to(self.get_var_store().device()))
             .collect::<Vec<Tensor>>();
 
         Tensor::stack(&token_ids, 0)

diff --git a/src/pipelines/conversation.rs b/src/pipelines/conversation.rs
@@ -1027,7 +1027,7 @@ impl ConversationModel {
                 padded_input.extend(input);
                 padded_input
             })
-            .map(|tokens| Tensor::of_slice(&tokens).to(self.device))
+            .map(|tokens| Tensor::from_slice(&tokens).to(self.device))
             .collect::<Vec<Tensor>>();
 
         (Tensor::stack(&concatenated_inputs, 0), attention_mask)