Pytorch 2 - OpenNMT-py v3.3 (#2418)

* Pytorch 2 - OpenNMT-py v3.3

.github/workflows/push.yml

@@ -374,7 +374,7 @@ jobs:
           -ban_unk_token \
           -length_penalty none \
           -out /tmp/gen
-        diff data/data_lm/gen-nucleus-sampling-sol.txt /tmp/gen && rm /tmp/gen
+        diff data/data_lm/gen-nucleus-sampling-sol$(python -c "import torch; print(torch.__version__[0])").txt /tmp/gen && rm /tmp/gen
     - name: Test LM generation with random sampling multi-beams
       run: |
         python translate.py -model onmt/tests/test_model_lm.pt \
@@ -389,7 +389,7 @@ jobs:
           -ban_unk_token \
           -min_length 5 \
           -out /tmp/gen
-        diff data/data_lm/gen-sampling-beams-sol.txt /tmp/gen && rm /tmp/gen
+        diff data/data_lm/gen-sampling-beams-sol$(python -c "import torch; print(torch.__version__[0])").txt /tmp/gen && rm /tmp/gen
     - name: Test extract_vocabulary tool
       run: |
         python tools/extract_vocabulary.py \

CHANGELOG.md

@@ -2,7 +2,18 @@
 **Notes on versioning**
 
 ## [Unreleased]
-## [3.2.0](https://github.com/OpenNMT/OpenNMT-py/tree/3.2.0) (2023-06-XX)
+
+## [3.3.0](https://github.com/OpenNMT/OpenNMT-py/tree/3.2.0) (2023-06-21)
+
+* Switch to pytorch 2.0.1
+* Eval LLM with MMLU benchmark
+* Fix Falcon 40B conversion / finetuning / inference
+* Plugin encoder/decoder thanks @kleag / @n2oblife
+* Safetensors for model storage (beta)
+* finetuning config templates for supported LLMs
+
+
+## [3.2.0](https://github.com/OpenNMT/OpenNMT-py/tree/3.2.0) (2023-06-07)
 * Skip init during model build (way faster building)
 * Enable quantization of LoRA layers
 * Enable 4bit quantization from bitsandbytes (NF4 / FP4)

data/data_lm/gen-nucleus-sampling-sol2.txt

@@ -0,0 +1,7 @@
+like your equipment !
+in your Presidency or registering erratic full detailed table since Waddington , as well handled separately .
+the importance 's future .
+there is survivors , under new public beforehand ?
+bankers sit on the honour and the old , accusations by Cubase for political leadership in the fifth generation of 0.5 , 1 January in serving the old , 1 .
+can do planet what Mr Titley , should like What is really do so done guarantee - ™ your Gore .
+we can restore confidence in Switzerland .

data/data_lm/gen-sampling-beams-sol2.txt

@@ -0,0 +1,7 @@
+you ! your staff !
+in German Presidency or rather than Andalusia .
+a moment , one century .
+" s operational from 06: 00 on weekdays , clothes hang thick from the and 00 until about 00 until the sea .
+fine words has underlined the fine vegetarian ...
+service are buying any of my property .
+we are much running behind the facts .

onmt/__init__.py

@@ -21,4 +21,4 @@
     onmt.modules,
 ]
 
-__version__ = "3.2.0"
+__version__ = "3.3.0"

onmt/tests/pull_request_chk.sh

@@ -491,7 +491,7 @@ ${PYTHON} translate.py -model ${TEST_DIR}/test_model_lm.pt  \
             -ban_unk_token \
             -length_penalty none \
             -out $TMP_OUT_DIR/gen_sampling  >> ${LOG_FILE} 2>&1
-diff ${DATA_DIR}/data_lm/gen-nucleus-sampling-sol.txt $TMP_OUT_DIR/gen_sampling
+diff ${DATA_DIR}/data_lm/gen-nucleus-sampling-sol$(python -c "import torch; print(torch.__version__[0])").txt $TMP_OUT_DIR/gen_sampling
 [ "$?" -eq 0 ] || error_exit
 echo "Succeeded" | tee -a ${LOG_FILE}
 rm $TMP_OUT_DIR/gen_sampling
@@ -509,7 +509,7 @@ ${PYTHON} translate.py -model ${TEST_DIR}/test_model_lm.pt  \
             -ban_unk_token \
             -min_length 5 \
             -out $TMP_OUT_DIR/gen_sampling  >> ${LOG_FILE} 2>&1
-diff ${DATA_DIR}/data_lm/gen-sampling-beams-sol.txt $TMP_OUT_DIR/gen_sampling
+diff ${DATA_DIR}/data_lm/gen-sampling-beams-sol$(python -c "import torch; print(torch.__version__[0])").txt $TMP_OUT_DIR/gen_sampling
 [ "$?" -eq 0 ] || error_exit
 echo "Succeeded" | tee -a ${LOG_FILE}
 rm $TMP_OUT_DIR/gen_sampling

setup.py

@@ -11,7 +11,7 @@
     description="A python implementation of OpenNMT",
     long_description=long_description,
     long_description_content_type="text/markdown",
-    version="3.2.0",
+    version="3.3",
     packages=find_packages(),
     project_urls={
         "Documentation": "http://opennmt.net/OpenNMT-py/",
@@ -21,7 +21,7 @@
     },
     python_requires=">=3.8",
     install_requires=[
-        "torch>=1.13,<2",
+        "torch>=1.13,<2.1",
         "configargparse",
         "ctranslate2>=3.2,<4",
         "tensorboard>=2.3",