使用Reticulate加载TensorFlow时R崩溃：非法操作数错误

现象 #

在RStudio中加载TensorFlow时，会话直接异常终止，报错：

ERROR The previous R session terminated abnormally; LOGGED FROM: rstudio::core::Error {anonymous}::rInit(const rstudio::r::session::RInitInfo&) src/cpp/session/SessionMain.cpp:728

绕过RStudio，用repl_python()直接调用Python环境导入TensorFlow时，得到更详细的错误：

> repl_python()
Python 3.10.12 (/home/jon/.virtualenvs/test-1-env/bin/python)
Reticulate 1.34.0 REPL -- A Python interpreter in R.
Enter 'exit' or 'quit' to exit the REPL and return to R.
>>> import tensorflow

 *** caught illegal operation ***
address 0x7fb5c8bb2ecb, cause 'illegal operand'

Traceback:
 1: py_call_impl(callable, call_args$unnamed, call_args$named)
 2: builtins$eval(compiled, globals, locals)
 3: py_compile_eval(code, capture = FALSE)
 4: doTryCatch(return(expr), name, parentenv, handler)
 5: tryCatchOne(expr, names, parentenv, handlers[[1L]])
 6: tryCatchList(expr, names[-nh], parentenv, handlers[-nh])
 7: doTryCatch(return(expr), name, parentenv, handler)
 8: tryCatchOne(tryCatchList(expr, names[-nh], parentenv, handlers[-nh]),     names[nh], parentenv, handlers[[nh]])
 9: tryCatchList(expr, classes, parentenv, handlers)
10: tryCatch(py_compile_eval(code, capture = FALSE), error = handle_error,     interrupt = handle_interrupt)
11: repl()
12: doTryCatch(return(expr), name, parentenv, handler)
13: tryCatchOne(expr, names, parentenv, handlers[[1L]])
14: tryCatchList(expr, classes, parentenv, handlers)
15: tryCatch(repl(), interrupt = identity)
16: repl_python()

安装方式 #

通过reticulate在虚拟环境中安装TensorFlow：

virtualenv_install("test-1-env", "tensorflow")

安装过程无异常，日志显示：

Using virtual environment 'test-1-env' ...
+ /home/jon/.virtualenvs/test-1-env/bin/python -m pip install --upgrade --no-user tensorflow
Requirement already satisfied: tensorflow in /home/jon/.virtualenvs/test-1-env/lib/python3.10/site-packages (2.15.0)
...
Requirement already satisfied: oauthlib>=3.0.0 in /home/jon/.virtualenvs/test-1-env/lib/python3.10/site-packages (from requests-oauthlib>=0.7.0->google-auth-oauthlib<2,>=0.5->tensorboard<2.16,>=2.15->tensorflow) (3.2.2)

环境配置 #

> sessionInfo()
R version 4.3.2 (2023-10-31)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 22.04.3 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3 
LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.20.so;  LAPACK version 3.10.0

locale:
 [1] LC_CTYPE=C.UTF-8       LC_NUMERIC=C           LC_TIME=C.UTF-8        LC_COLLATE=C.UTF-8    
 [5] LC_MONETARY=C.UTF-8    LC_MESSAGES=C.UTF-8    LC_PAPER=C.UTF-8       LC_NAME=C             
 [9] LC_ADDRESS=C           LC_TELEPHONE=C         LC_MEASUREMENT=C.UTF-8 LC_IDENTIFICATION=C   

time zone: America/New_York
tzcode source: system (glibc)

attached base packages:
[1] stats     graphics  grDevices datasets  utils     methods   base      

other attached packages:
 [1] arrow_14.0.0.2    dplyr_1.1.4       reticulate_1.34.0 plotly_4.10.3     ggplot2_3.4.4    
 [6] nat_1.10.4        rgl_1.2.8         scales_1.3.0      cgalMeshes_2.2.0  knitr_1.45       

解决思路 #

1. 检查CPU指令集兼容性 #

illegal operand报错大多是因为TensorFlow预编译版本用到了CPU不支持的指令集（如AVX2、FMA）。Ubuntu默认安装的TensorFlow针对新CPU优化，若你的CPU不支持这些指令就会触发错误。

用以下命令检查CPU支持的指令集：

cat /proc/cpuinfo | grep flags

如果输出里没有avx2、fma字段，说明需要安装兼容旧CPU的TensorFlow版本。

2. 安装兼容旧CPU的TensorFlow #

进入虚拟环境：

source /home/jon/.virtualenvs/test-1-env/bin/activate

安装兼容版本（以2.15.0为例）：

pip install tensorflow==2.15.0 --no-binary :all:

也可以通过conda-forge安装预编译的兼容版：

conda install -c conda-forge tensorflow=2.15.0

3. 验证Reticulate环境配置 #

确保Reticulate正确指向目标虚拟环境：

library(reticulate)
use_virtualenv("test-1-env", required = TRUE)
# 检查Python环境配置
py_config()
# 尝试导入TensorFlow验证
tf <- import("tensorflow")
tf$`__version__`

4. 禁用TensorFlow的CPU优化 #

若安装兼容版后仍有问题，可尝试禁用CPU优化：

Sys.setenv(TF_CPP_MIN_LOG_LEVEL = "2")
Sys.setenv(TF_ENABLE_ONEDNN_OPTS = "0")
library(reticulate)
use_virtualenv("test-1-env")
tf <- import("tensorflow")

5. 排查RStudio兼容性 #

• 更新RStudio到最新版本
• 禁用RStudio中所有无关插件后重试
• 用终端启动R，再加载TensorFlow，排查是否是RStudio自身的问题

内容的提问来源于stack exchange，提问作者myfatson