A file verification and remediation tool.

It applies the following validations:
- prints the number of examples, and warns if it's lower than 100
- ensures prompt and completion columns are present
- optionally removes any additional columns
- ensures all completions are non-empty
- infers which type of fine-tuning the data is most likely in (classification, conditional generation and open-ended generation)
- optionally removes duplicate rows
- infers the existence of a common suffix, and if there is none, suggests one for classification and conditional generation
- optionally prepends a space to each completion, to make tokenization better
- optionally splits into training and validation set for the classification use case
- optionally ensures there's an ending string for all completions
- optionally lowercases completions or prompts if more than a 1/3 of alphanumeric characters are upper case

It interactively asks the user to accept or reject recommendations. If the user is happy, then it saves the modified output file as a jsonl, which is ready for being used in fine-tuning with the printed command.

Author: BorisPower

bin/openai

@@ -1,13 +1,11 @@
 #!/usr/bin/env python
 import argparse
-import json
 import logging
-import os
 import sys
 
 import openai
 from openai.cli import display_error
-from openai.cli import register as api_register
+from openai.cli import api_register, tools_register
 
 logger = logging.getLogger()
 formatter = logging.Formatter("[%(asctime)s] %(message)s")
@@ -40,9 +38,11 @@ def main():
     parser.set_defaults(func=help)
 
     subparsers = parser.add_subparsers()
-    sub = subparsers.add_parser("api", help="Direct API calls")
+    sub_api = subparsers.add_parser("api", help="Direct API calls")
+    sub_tools = subparsers.add_parser("tools", help="Client side tools for convenience")
 
-    api_register(sub)
+    api_register(sub_api)
+    tools_register(sub_tools)
 
     args = parser.parse_args()
     if args.verbosity == 1:

openai/cli.py

@@ -3,6 +3,13 @@
 import signal
 import sys
 import warnings
+from openai.validators import (
+    write_out_file,
+    apply_necessary_remediation,
+    apply_optional_remediation,
+    read_any_format,
+    get_validators,
+)
 
 import openai
 
@@ -393,8 +400,69 @@ def cancel(cls, args):
         resp = openai.FineTune.cancel(id=args.id)
         print(resp)
 
+    @classmethod
+    def prepare_data(cls, args):
+
+        sys.stdout.write("Analyzing...\n")
+        fname = args.file
+        df, remediation = read_any_format(fname)
+        apply_necessary_remediation(None, remediation)
+
+        validators = get_validators()
+
+        optional_remediations = []
+        if remediation is not None:
+            optional_remediations.append(remediation)
+        for validator in validators:
+            remediation = validator(df)
+            if remediation is not None:
+                optional_remediations.append(remediation)
+                df = apply_necessary_remediation(df, remediation)
+
+        any_optional_or_necessary_remediations = any(
+            [
+                remediation
+                for remediation in optional_remediations
+                if remediation.optional_msg is not None
+                or remediation.necessary_msg is not None
+            ]
+        )
+
+        if any_optional_or_necessary_remediations:
+            sys.stdout.write(
+                "\n\nBased on the analysis we will perform the following actions:\n"
+            )
+
+            for remediation in optional_remediations:
+                df = apply_optional_remediation(df, remediation)
+        else:
+            sys.stdout.write("\n\nNo remediations found.\n")
+
+        write_out_file(df, fname, any_optional_or_necessary_remediations)
+
+
+def tools_register(parser):
+    subparsers = parser.add_subparsers(
+        title="Tools", help="Convenience client side tools"
+    )
+
+    def help(args):
+        parser.print_help()
+
+    parser.set_defaults(func=help)
+
+    sub = subparsers.add_parser("fine_tunes.prepare_data")
+    sub.add_argument(
+        "-f",
+        "--file",
+        required=True,
+        help="JSONL, JSON, CSV, TSV, TXT or XLSX file containing prompt-completion examples to be analyzed."
+        "This should be the local file path.",
+    )
+    sub.set_defaults(func=FineTune.prepare_data)
+
 
-def register(parser):
+def api_register(parser):
     # Engine management
     subparsers = parser.add_subparsers(help="All API subcommands")