remove nightly

CHANGES.md

@@ -11,18 +11,11 @@ Github. Parentheses after an item show the name or github id of the contributor
 of that change.
 
 
-
-
-
-
 ## 1.0.35.dev0 (Work In Progress)
 
-### New:
-
 ### Changed:
 
-### Fixed:
-
+- Update deps to release version of pytorch v1
 
 
 ## 1.0.34 (2018-12-06)
@@ -32,7 +25,6 @@ of that change.
 - pypi wheel `dataclasses` dependency for py3.6 is there again
 
 
-
 ## 1.0.33 (2018-12-05)
 
 ### New:
@@ -52,8 +44,6 @@ of that change.
 
 ## 1.0.32 (2018-12-02)
 
-### New:
-
 ### Changed:
 
 - `TabularDatBunch.from_df` accepts a `test_df` argument

README.md

@@ -42,20 +42,18 @@ More advanced installation issues, such as installing only partial dependencies
 * GPU
 
    ```bash
-   conda install -c pytorch pytorch-nightly cuda92
-   conda install -c fastai torchvision-nightly
+   conda install -c pytorch pytorch torchvision
    conda install -c fastai fastai
    ```
 
 * CPU
 
    ```bash
-   conda install -c pytorch pytorch-nightly-cpu
-   conda install -c fastai torchvision-nightly-cpu
+   conda install -c pytorch pytorch-cpu torchvision-cpu
    conda install -c fastai fastai
    ```
 
-   On MacOS for non-GPU build install `pytorch-nightly` and `torchvision-nightly` instead of `pytorch-nightly-cpu` and `torchvision-nightly-cpu`.
+   On MacOS for non-GPU build install `pytorch` and `torchvision` instead of `pytorch-cpu` and `torchvision-cpu`.
 
 Note that JPEG decoding can be a bottleneck, particularly if you have a fast CPU. You can optionally install an optimized JPEG decoder as follows (Linux):
 
@@ -71,20 +69,17 @@ CC="cc -mavx2" pip install --no-cache-dir -U --force-reinstall pillow-simd
 * GPU
 
    ```bash
-   pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cu92/torch_nightly.html
+   pip install torch torchvision
    pip install fastai
    ```
 
 * CPU
 
    ```bash
-   pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
+   pip install http://download.pytorch.org/whl/cpu/torch-1.0.0-cp36-cp36m-linux_x86_64.whl
    pip install fastai
    ```
 
-NB: this set will also fetch `torchvision-nightly`, which supports `torch-1.x`.
-
-
 
 ### Developer Install
 
@@ -166,28 +161,25 @@ conda install conda
 
    Since fastai-1.0 relies on pytorch-1.0, you need to be able to install pytorch-1.0 first.
 
-   As of this moment pytorch.org's pre-1.0.0 version (`torch-nightly`) supports:
+   As of this moment pytorch.org's 1.0 version supports:
 
     | Platform | GPU    | CPU    |
     |----------|--------|--------|
     | linux    | binary | binary |
     | mac      | source | binary |
-    | windows  | source | source |
+    | windows  | binary | binary |
 
    Legend: `binary` = can be installed directly, `source` = needs to be built from source.
 
-   This will change once `pytorch` 1.0.0 is released and installable packages made available for your system, which could take some time after the official release is made. Please watch for updates [here](https://pytorch.org/get-started/locally/).
-
    If there is no `pytorch` preview conda or pip package available for your system, you may still be able to [build it from source](https://pytorch.org/get-started/locally/).
 
-   Alternatively, please consider installing and using the very solid "0.7.x" version of `fastai`. Please see the [instructions](https://github.com/fastai/fastai/tree/master/old).
-
 4. How do you know which pytorch cuda version build to choose?
 
-   It depends on the version of the installed NVIDIA driver. Here are the requirements for CUDA versions supported by pre-built `pytorch-nightly` releases:
+   It depends on the version of the installed NVIDIA driver. Here are the requirements for CUDA versions supported by pre-built `pytorch` releases:
 
     | CUDA Toolkit | NVIDIA (Linux x86_64) |
     |--------------|-----------------------|
+    | CUDA 10.0    | >= 410.00             |
     | CUDA 9.2     | >= 396.26             |
     | CUDA 9.0     | >= 384.81             |
     | CUDA 8.0     | >= 367.48             |
@@ -196,7 +188,6 @@ conda install conda
    You can find a complete table with all variations [here](https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html).
 
 
-
 ## Updates
 
 In order to update your environment, simply install `fastai` in exactly the same way you did the initial installation.

azure-pipelines.yml

@@ -53,8 +53,8 @@ jobs:
     # Install dependencies and fastai git HEAD or PR
   - script: |
       python -m pip install --upgrade pip && pip3 install --upgrade setuptools
-      pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
-      # test that we can install fastai package (and torchvision)
+      pip install torch torchvision
+      # test that we can install fastai package
       pip install fastai
       pip uninstall -y fastai
       # install editable fastai, which tests the HEAD or PR
@@ -145,8 +145,7 @@ jobs:
       conda install -y conda
       conda install -y pip setuptools
       # pytorch + torchvision
-      conda install -y -c pytorch pytorch-nightly-cpu
-      conda install -y -c fastai torchvision-nightly-cpu
+      conda install -y -c pytorch pytorch torchvision
       # test that we can install fastai package
       conda install -y -c fastai fastai
       conda uninstall -y fastai
@@ -318,9 +317,7 @@ jobs:
       conda install -y conda
       conda install -y pip setuptools
       # pytorch + torchvision
-      # pytorch on osx doesn't ship GPU builds, so pytorch-nightly is pytorch-nightly-cpu. while the latter package exists, it's not maintained since 20181014
-      conda install -y -c pytorch pytorch-nightly
-      conda install -y -c fastai torchvision-nightly
+      conda install -y -c pytorch pytorch torchvision
       # test that we can install fastai package
       conda install -y -c fastai fastai
       conda uninstall -y fastai

conda/meta.yaml

@@ -45,9 +45,6 @@ requirements:
 
 # skipped requirements:
 # jupyter_contrib_nbextensions - only needed for internal tools - it's not on conda and requires too many of its own dependencies which aren't on conda either
-# XXX: change pytorch-nightly to pytorch >=1.0.0 once it's released on conda
-# XXX: change torchvision-nightly to torchvision >=0.2.2 once 0.2.2 (against pytorch-1.0) is released on conda
-# XXX: temporarily pinning spacy and its dependencies (regex, thinc, and cymem) to have a stable environment during the course duration.
 #
 # note: "module # [py36]" tells conda-build that this requirement is only for python3.6, it's not a comment.
   run:
@@ -62,14 +59,18 @@ requirements:
     - python
     - requests
     - scipy
-    - spacy ==2.0.16
-    - regex ==2018.8.29
-    - thinc ==6.12.0
-    - cymem ==2.0.2
+    - spacy
+    - regex
+    - thinc
+    - cymem
     - typing
     - pyyaml
+    - pytorch >=1.0
+    - torchvision
 
-# testing removal of stuff only used for doc() and dev-time
+# Need to handle CPU vs GPU still
+
+# Stuff only used for doc() and dev-time:
 #- ipython
 #- jupyter
 #- notebook>=5.7.0
@@ -78,19 +79,10 @@ requirements:
 #- traitlets
 
 # removed these and instead are asking users to install those directly until a proper pytorch-1.0.0 and the corresponding torchvision are released on conda. otherwise can't support these two sets:
-#    - pytorch-nightly
-#    - torchvision-nightly
-#
-#    - pytorch-nightly-cpu
-#    - torchvision-nightly-cpu
-
-# removing until pytorch-1.0.0 release - conda can't handle these dependencies and tries to pull in pytorch-0.4.1 due to torchvision depending on it. yet, pytorch doesn't release torchvision nightly build... no way out
-#    - pytorch >=1.0.0
 
 test:
   imports:
-# removing the loading test for now, since we temporarily made fastai conda package not depend on torch/torchvision due to dependencies hell - see above.
-#    - fastai
+    - fastai
 
   requires:
     - python
@@ -121,3 +113,4 @@ extra:
     - setuptools >=0.39
     - twine >=1.12
   recipe-maintainers: ''
+

docs/dev/release.md

@@ -937,8 +937,7 @@ conda create -y  python=3.6 --name fastai-py3.6
 conda activate fastai-py3.6
 conda install -y conda
 conda install -y pip setuptools
-conda install -y -c pytorch pytorch-nightly cuda92
-conda install -y -c fastai torchvision-nightly
+conda install -y -c pytorch pytorch cuda92 torchvision
 conda install -c fastai fastai
 conda uninstall -y fastai
 pip install -e .
@@ -1066,38 +1065,38 @@ AND                     |"numpy>=1.8,<2"      |1.8, 1.9, not 2.0
   `conda search` outputs results as following:
 
     ```
-    conda search -c pytorch "pytorch-nightly"
+    conda search -c pytorch "pytorch"
     Loading channels: done
     # Name                  Version           Build                   Channel
-    pytorch-nightly 0.5.0.dev20180914 py3.5_cpu_0                     pytorch
-    pytorch-nightly 0.5.0.dev20180914 py3.5_cuda8.0.61_cudnn7.1.2_0   pytorch
-    pytorch-nightly 0.5.0.dev20180914 py3.5_cuda9.0.176_cudnn7.1.2_0  pytorch
-    pytorch-nightly 0.5.0.dev20180914 py3.5_cuda9.2.148_cudnn7.1.4_0  pytorch
+    pytorch 0.5.0.dev20180914 py3.5_cpu_0                     pytorch
+    pytorch 0.5.0.dev20180914 py3.5_cuda8.0.61_cudnn7.1.2_0   pytorch
+    pytorch 0.5.0.dev20180914 py3.5_cuda9.0.176_cudnn7.1.2_0  pytorch
+    pytorch 0.5.0.dev20180914 py3.5_cuda9.2.148_cudnn7.1.4_0  pytorch
     [...]
     ```
 
     To narrow the results, e.g. show only python3 cpu builds:
 
     ```
-    conda search -c pytorch "pytorch-nightly[build=py3*_cpu_0]"
+    conda search -c pytorch "pytorch[build=py3*_cpu_0]"
     ```
 
-    and then feed it to `conda install` with specific `==version=build` after the package name, e.g. `pytorch-nightly==1.0.0.dev20180916=py3.6_cpu_0`
+    and then feed it to `conda install` with specific `==version=build` after the package name, e.g. `pytorch==1.0.0.dev20180916=py3.6_cpu_0`
 
 
     To search for packages for a given system (by default, packages for your current
 platform are shown):
 
     ```
-    conda search -c pytorch "pytorch-nightly[subdir=osx-64]"
+    conda search -c pytorch "pytorch[subdir=osx-64]"
     ```
 
     Some of the possible platforms include `linux-32`, `linux-64`, `win-64`, `osx-64`.
 
     And these can be combined:
 
     ```
-    conda search -c pytorch "pytorch-nightly[subdir=osx-64, build=py3.7*]"
+    conda search -c pytorch "pytorch[subdir=osx-64, build=py3.7*]"
     ```
 
     To search all packages released by user `fastai`:

docs/install.md

@@ -15,33 +15,34 @@ First, install `fastai` without its dependencies, and then install the dependenc
 
 ```
 pip install --no-deps fastai
-pip install "fastprogress>=0.1.15" "matplotlib" "numpy>=1.12" ...
+pip install "fastprogress>=0.1.18" "matplotlib" "numpy>=1.12" ...
 ```
 this will work with conda too:
 
 ```
 conda install --no-deps -c fastai fastai
-conda install -c fastai "fastprogress>=0.1.15" "matplotlib" "numpy>=1.12" ...
+conda install -c fastai "fastprogress>=0.1.18" "matplotlib" "numpy>=1.12" ...
 ```
 
-Don't forget to add `-c fastai` for the conda installs, e.g. it's needed for `torchvision-nightly`.
+Don't forget to add `-c fastai` for the conda installs, e.g. it's needed for `torchvision`.
 
 Below you will find the groups of dependencies for you to choose from. `fastai.base` is mandatory, the rest are optional:
 
 ```
 fastai.base:
 
-  "fastprogress>=0.1.15" "matplotlib" "numpy>=1.12" "pandas" "bottleneck" "numexpr" "Pillow" "requests" "scipy" "typing" "pyyaml"
+  "fastprogress>=0.1.18" "matplotlib" "numpy>=1.12" "pandas" "bottleneck" "numexpr" "Pillow" "requests" "scipy" "typing" "pyyaml"
 
 fastai.text:
 
-  "spacy==2.0.16" "regex" "thinc==6.12.0" "cymem==2.0.2"
+  "spacy" "regex" "thinc" "cymem"
 
 fastai.text.qrnn:
 
   "cupy"
 
 fastai.vision:
 
-  "torchvision-nightly"
+  "torchvision"
+
 ```

docs/troubleshoot.md

@@ -20,7 +20,7 @@ despite having `nvidia-smi` working just fine. Which means that `pytorch` can't
 
 note: `pytorch` installs itself as `torch`. So we refer to the project and its packages as `pytorch`, but inside python we use it as `torch`.
 
-First, starting with `pytorch-1.0.x` it doesn't matter which CUDA version you have installed on your system, always try first to install the latest `pytorch-nightly` with `cuda92` - it has all the required libraries built into the package. However, note, that you most likely will **need 396.xx+ driver for `pytorch` built with `cuda92`**. For older drivers you will probably need to install `pytorch` with `cuda90` or ever earlier.
+First, starting with `pytorch-1.0.x` it doesn't matter which CUDA version you have installed on your system, always try first to install the latest `pytorch` - it has all the required libraries built into the package. However, note, that you most likely will **need 396.xx+ driver for `pytorch` built with `cuda92`**. For older drivers you will probably need to install `pytorch` with `cuda90` or ever earlier.
 
 The only thing you to need to ensure is that you have a correctly configured NVIDIA driver, which usually you can test by running: `nvidia-smi` in your console.
 

docs_src/index.ipynb

@@ -100,7 +100,7 @@
     "To install or update fastai, we recommend `conda`:\n",
     "\n",
     "```\n",
-    "conda install -c pytorch -c fastai fastai pytorch-nightly cuda92\n",
+    "conda install -c pytorch -c fastai fastai pytorch\n",
     "```\n",
     "For troubleshooting, and alternative installations (including pip and CPU-only options) see the [fastai readme](https://github.com/fastai/fastai/blob/master/README.md)."
    ]

fastai/vision/learner.py

@@ -27,8 +27,10 @@ def cnn_config(arch):
     torch.backends.cudnn.benchmark = True
     return model_meta.get(arch, _default_meta)
 
-def create_body(model:nn.Module, cut:Optional[int]=None, body_fn:Callable[[nn.Module],nn.Module]=None):
+def create_body(arch:Callable, pretrained:bool=True, cut:Optional[int]=None, body_fn:Callable[[nn.Module],nn.Module]=None):
     "Cut off the body of a typically pretrained `model` at `cut` or as specified by `body_fn`."
+    model = arch(pretrained)
+    if not cut and not body_fn: cut = cnn_config(arch)['cut']
     return (nn.Sequential(*list(model.children())[:cut]) if cut
             else body_fn(model) if body_fn else model)
 
@@ -50,7 +52,7 @@ def create_cnn(data:DataBunch, arch:Callable, cut:Union[int,Callable]=None, pret
                 bn_final:bool=False, **kwargs:Any)->Learner:
     "Build convnet style learners."
     meta = cnn_config(arch)
-    body = create_body(arch(pretrained), ifnone(cut,meta['cut']))
+    body = create_body(arch, pretrained, cut)
     nf = num_features_model(body) * 2
     head = custom_head or create_head(nf, data.c, lin_ftrs, ps=ps, bn_final=bn_final)
     model = nn.Sequential(body, head)
@@ -65,7 +67,7 @@ def unet_learner(data:DataBunch, arch:Callable, pretrained:bool=True, blur_final
                  self_attention:bool=False, **kwargs:Any)->None:
     "Build Unet learners. `kwargs` are passed down to `conv_layer`."
     meta = cnn_config(arch)
-    body = create_body(arch(pretrained), meta['cut'])
+    body = create_body(arch, pretrained)
     model = to_device(models.unet.DynamicUnet(body, n_classes=data.c, blur=blur, blur_final=blur_final,
                                               self_attention=self_attention, norm_type=norm_type), data.device)
     learn = Learner(data, model, **kwargs)

setup.py

@@ -21,16 +21,11 @@ def to_list(buffer): return list(filter(None, map(str.strip, buffer.splitlines()
 #   pip install fastai
 #   pip install -e .
 #
-# XXX: to fix later in time:
-# - require torch>=1.0.0 once it's released, for now get the user to install it explicitly
-# - using a workaround for torchvision, once torch-1.0.0 is out and a new torchvision depending on it is released switch to torchvision>=0.2.2
-# - temporarily pinning spacy and its dependencies (regex, thinc, and cymem) to have a stable environment during the course duration.
-#
 # notes:
 # - bottleneck and numexpr are performance-improvement extras for numpy
 #
 # dependencies to skip for now:
-# - cupy - is only required for QRNNs - sgguger thinks later he will get rid of this dep.
+# - cupy - is only required for QRNNs - sgugger thinks later he will get rid of this dep.
 
 requirements = to_list("""
     bottleneck
@@ -46,9 +41,11 @@ def to_list(buffer): return list(filter(None, map(str.strip, buffer.splitlines()
     regex
     requests
     scipy
-    spacy==2.0.16
-    thinc==6.12.0
-    torchvision-nightly
+    spacy
+    regex
+    thinc
+    cymem
+    torchvision
     typing
 """)