"Construction of a trivial graph to add two float tensors and then repeated invocation of the single graph using placeholders", from https://github.com/tensorflow/tensorflow/issues/6781#issuecomment-271888411 in Clojure based on the original Java

tensor-with-placeholders.clj

;; make sure your ns has `(:import [org.tensorflow DataType Graph Output Session Tensor])`
;; or if in REPL, evaluate `(import [org.tensorflow DataType Graph Output Session Tensor])` first

;; Construct a graph to add two float Tensors, using placeholders.
(let [g (Graph.)
      s (Session. g)
      x (-> (.opBuilder g "Placeholder" "x")
            (.setAttr "dtype" DataType/FLOAT)
            (.build)
            (.output 0))
      ;; (`y` is identical except for name)
      y (-> (.opBuilder g "Placeholder" "y")
            (.setAttr "dtype" DataType/FLOAT)
            (.build)
            (.output 0))
    
      z (-> (.opBuilder g "Add" "z")
            (.addInput x)
            (.addInput y)
            (.build)
            (.output 0))
      big-x (mapv float [1 2 3])
      big-y (mapv float [4 5 6])]
  
  ;; Execute the graph multiple times, each time with a different value of x and y
  (for [i (range (count big-x))]
    (let [tx (Tensor/create (nth big-x i))
          ty (Tensor/create (nth big-y i))
          tz (-> (.runner s)
                 (.feed "x" tx)
                 (.feed "y" ty)
                 (.fetch "z")
                 (.run)
                 (.get 0))]
      (println (nth big-x i) " + " (nth big-y i) " = " (.floatValue tz)))))

;; result:

1.0  +  4.0  =  5.0
2.0  +  5.0  =  7.0
3.0  +  6.0  =  9.0
(nil nil nil)