𝑖i (Imaginary Interpreter)

Install with Pen
Pen.el on GitHub	https://github.com/semiosis/pen.el/
Tutorial	https://mullikine.github.io/posts/pen-el-tutorial/
Latest demos	AI Terminals // Bodacious Blog

Summary

This is an introduction to the imaginary interpreter built into Pen.el.

The reason I am building REPLs and emacs modes based around imaginary interpreters (which are not at all deterministic or logically sound) is that I believe they will be very powerful and useful in the future.

It’s important to develop a language-agnostic harness for working with arbitrarily many imagined languages.

Solving the problems of utilising current LMs will make it easier down the line to utilise more advanced ones.

Steps to run

On your host machine, or from within the docker container’s shell
- Run ii python, or any desired language as the first argument

To see what interpreters are available:

1

penl | grep -P -- "imagine.*-interpreter" | head -n 3

pf-imagine-a-smalltalk-interpreter/1
pf-imagine-a-scheme-interpreter/1
pf-imagine-a-scala-interpreter/1

Architecture

The interpreter prompt

An interpreter prompt is selected depending on if a specialised prompt exists for the chosen language.
- The generic 𝑖i prompts is comprised of:
  - kickstarter: A subprompt declaring that the following are examples of the specified language.
  - Consecutive Input/Output pairs.
- Language-specific 𝑖i prompts are comprised of:
  - kickstarter: A subprompt which is typically a banner.
  - postpostprocessor: A script that takes the interpreter history and generates a new “user prompt”.
An interpreter prompt accepts as a first argument the history of the interaction with the interpreter. This way, the interpreter has an imaginary state as you work with it.

`ii` script

Accepts a language as an argument
Docker bash-host interop: 𝑖i communicates with arbitrary LMs via the bash (docker-host) interop of Pen.el.
rlwrap is used to keep a history of the inputs specific to each interpreter.
A REPL that keeps builds a transcript and passes that as input to the interpreter prompt.

Code

`𝑖i` (Imaginary Interpreter for bash)

full code: http://github.com/semiosis/pen.el/blob/master/scripts/ii

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30


#!/bin/bash

# Imaginary interpreter
# Version 2.0, bash

#...

memory=

prompt_the_lm() {
    user_input="$1"

    if test -n "$fun"; then
        penf -u "$fun" "$memory" "$user_input"
    else
        penf -u pf-generic-interpreter/3 "$memory" "$user_input" "$lang"
    fi
}

p "${kickstarter} "
memory="${kickstarter} "

while read -e -r user_input; do
    gen="$(prompt_the_lm "$user_input" | chomp)"
    memory+="$user_input\n"
    p "$gen"
    memory+="$gen"
done

#...

The imaginary interpreter may be implemented for different languages, by using Pen.el’s prompt functions via the bash-docker interop.

An imaginary python interpreter

When a supported language such as Python is chosen, a prompt specifically designed for it may be used.

Prompt: http://github.com/semiosis/prompts/blob/master/prompts/imagine-a-python-interpreter-2.prompt

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31


include: Generic Interpreter/3
task: Imagine a <language> interpreter
language: python
subprompts:
- kickstarter: |+
    Python 3.8.5 (default, Jan 27 2021, 15:41:15)
    Type 'copyright', 'credits' or 'license' for more information
    IPython 7.21.0 -- An enhanced Interactive Python. Type '?' for help.

    In [1]:    
prompt: |+
    <history><expression>
    <:pp>Out    
user-prompt: "^In \\[[0-9]*\\]: "
# Unfortunately, we can't generate the next In
# prompt because we need to match on it with stop-sequences.
# So the user prompt must be reconstructed manually.
stop-sequences:
- "In ["
# Create a user prompt with the number incremented.
# This is like postprocessor but happens even later.
# It is used in special circumstances when the prompt history is also required.
postpostprocessor: pen-s python-gen-next-user-prompt
vars:
- history
- expression
var-defaults:
- kickstarter
examples:
- "In [1]: "
- "5 + 5"

Using the bash interop

-p ensures that the entire prompt along with the generated output is returned.

-u ensures that the cache is updated and a new generation is returned.

Withholding the first argument

By supplying an empty string as the first argument, the history is not passed to the interpreter prompt. Instead, the prompt function will use the kickstarter subprompt, as it has been supplied as the default value for the first variable.

1

penf -p -u imagine-a-python-interpreter/2 "" "5 + 5"

Python 3.8.5 (default, Jan 27 2021, 15:41:15)
Type 'copyright', 'credits' or 'license' for more information
IPython 7.21.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: 5 + 5
Out[1]: 10

1

penf -p -u imagine-a-python-interpreter/2 "In [3]: " "5 + 5"

In [3]: 5 + 5
Out[3]: 10

Demo

On the right, an imaginary interpreter.

An imaginary interpreter with any language

prompt: http://github.com/semiosis/prompts/blob/master/prompts/generic-interpreter-3.prompt

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35


task: "Generic Interpreter"
prompt-version: 1
subprompts:
- kickstarter: |+
    <language> interpreter.

    Input:    
prompt: |+
    <history><expression>
    <:pp>Output:    
lm-command: "openai-complete.sh"
engine: "OpenAI Davinci"
temperature: 0.4
max-tokens: 60
top-p: 1.0
cache: on
stop-sequences:
- "Input: "
vars:
- history
- expression
- language
var-defaults:
- kickstarter
examples:
- "Input: "
- "5 + 5"
- Python
filter: no
completion: off
insertion: off
interpreter: on
conversation: on
external-related:
- "https://semiosis.github.io/posts/imaginary-programming-with-gpt-3/"

Demo of ruby using the generic interpreter

This will use a generic prompt which does any language, but the name of the language has been specified as Rubylang.

When using 𝑖i, if a specialised interpreter prompt for a given language can be found then that will be used.

However, since no interpreter can be found for ‘Rubylang’, the generic interpreter is used.

However, the generic interpreter can still imagine the ‘Ruby’ language to a degree, but is far less accurate.

The beauty of this, of course, is that we don’t need to prime the interpreter with a banner or terminal history, and we don’t need to know what the prompt for Ruby looks like.

And a demo of a more catered imaginary interpreter

prompt: http://github.com/semiosis/prompts/blob/master/prompts/imagine-a-ruby-interpreter-2.prompt

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34


include: Generic Interpreter/3
task: Imagine a <language> interpreter
language: ruby
subprompts:
- kickstarter: |+
    $ ruby -v
    ruby 2.7.0p0 (2019-12-25 revision 647ee6f091) [x86_64-linux-gnu]
    $ irb
    2.7.0 :001 > RUBY_VERSION
    => "2.7.0"
    2.7.0 :002 >    
# Because the prompt generates from at the end of the user expression
# The generation will contain a starting newline.
# Want to trim that with =𝑖i=, but not in the prompt function because a better harness could be made.
prompt: |+
        <history><expression>
user-prompt: "^2.7.0 :[0-9]* > "
# Unfortunately, we can't generate the next In
# prompt because we need to match on it with stop-sequences.
# So the user prompt must be reconstructed manually.
stop-sequences:
# This isn't ideal but ruby-gen-next-user-prompt needs to see a full prompt
# So it can generate the next user prompt.
- " > "
postpostprocessor: pen-str ruby-gen-next-user-prompt
vars:
- history
- expression
var-defaults:
- kickstarter
n-completions: 10
examples:
- "2.7.0 :002 > "
- "puts \"Hi\""

Support scripts

These are used as post-processors in the prompt and in 𝑖i.

String utilities (`pen-s`)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32


remove-trailing-whitespace) {
    sed -z -e "s/\s*\$//" -e "s/\n*\$//"
}
;;

generic-interpreter-tidy-result) {
    pen-s remove-trailing-whitespace | { awk 1; echo -n "Input: "; }
}
;;

ruby-gen-next-user-prompt) {
    IFS= read -rd '' in < <(cat);typeset -p in &>/dev/null

    prompt_number="$(printf -- "%s" "$in" | awk 1 | tac | sed -n "/^2.7.0 :/{s/^2.7.0 :\\([0-9]*\\) .*/\\1/p;q}")"
    echo "$prompt_number"
    new_prompt_number="$(printf "%03d" $(( prompt_number + 1 )))"

    printf -- "%s" "$in" | awk 1
    echo -n "2.7.0 :$new_prompt_number > "
}
;;

python-gen-next-user-prompt) {
    IFS= read -rd '' in < <(cat);typeset -p in &>/dev/null

    prompt_number="$(printf -- "%s" "$in" | awk 1 | tac | sed -n "/^Out\\[/{s/^Out\\[\\([0-9]*\\)\\]:.*/\\1/p;q}")"
    new_prompt_number="$(( prompt_number + 1 ))"

    printf -- "%s" "$in" | awk 1
    echo -n "In [$new_prompt_number]: "
}
;;