CodeNet

This dataset is the MO-RELISH CodeNet code-generation quality-estimation collection derived from official Project CodeNet artifacts. It pairs a programming problem, sample I/O, and submitted source code with execution and source-size metadata. It is hosted as one Hugging Face dataset with one config per programming language plus an all config across curated languages.

Configs And Splits

Language	Config	Group	Train	Validation	ID test	OOD test
C++	cpp	high_resource	100,000	1,000	1,000	1,000
Python	python	high_resource	100,000	1,000	1,000	1,000
Java	java	high_resource	100,000	1,000	1,000	1,000
C	c	high_resource	100,000	1,000	1,000	1,000
Ruby	ruby	high_resource	100,000	1,000	1,000	1,000
C#	csharp	high_resource	100,000	1,000	1,000	1,000
Rust	rust	high_resource	100,000	1,000	1,000	1,000
Go	go	medium_resource	10,000	1,000	1,000	1,000
JavaScript	javascript	medium_resource	10,000	1,000	1,000	1,000
Haskell	haskell	medium_resource	10,000	1,000	1,000	1,000
Kotlin	kotlin	medium_resource	10,000	1,000	1,000	1,000
PHP	php	medium_resource	10,000	1,000	1,000	1,000
Scala	scala	medium_resource	10,000	1,000	1,000	1,000
Perl	perl	low_resource	1,000	1,000	1,000	1,000
Fortran	fortran	low_resource	1,000	1,000	1,000	1,000
Julia	julia	low_resource	1,000	1,000	1,000	1,000
OCaml	ocaml	low_resource	1,000	1,000	1,000	1,000
Lisp	lisp	low_resource	1,000	1,000	1,000	1,000
Lua	lua	low_resource	1,000	1,000	1,000	1,000
Pascal	pascal	low_resource	1,000	1,000	1,000	1,000

test_in_distribution shares problem IDs with training data. Every in-distribution test problem ID has at least one same-problem training row. test_out_of_distribution uses a global problem-ID pool that is disjoint from train, validation, and in-distribution test across all curated CodeNet languages.

Columns

Input columns:

• source_text: plain-text problem statement.
• input_text: submitted source code to evaluate.
• reference_outputs: list containing the sample output when available.
• prompt_components.problem_context: same problem statement as source_text.
• prompt_components.sample_input: official sample input.
• prompt_components.gold_output: official sample output.
• prompt_components.input_to_evaluate: same source code as input_text.

Prediction targets:

• targets.memory_kb
• targets.cpu_time_ms
• targets.code_size_bytes

Output dimensions:

• targets.memory_kb: Peak memory usage reported by official Project CodeNet metadata, in kilobytes.
• targets.cpu_time_ms: Execution CPU time reported by official Project CodeNet metadata, in milliseconds.
• targets.code_size_bytes: Submitted source-code file size, in bytes.

The retained metadata.status field records the original CodeNet submission status but is not a prediction target. Prediction targets: memory_kb, cpu_time_ms, code_size_bytes.

Loading

from datasets import load_dataset

python_ds = load_dataset("Samsoup/CodeNet", "python")
all_ds = load_dataset("Samsoup/CodeNet", "all")