Datasets:

nebius
/

Longtail-SWE

instance_id stringlengths 12 43	repo stringlengths 8 38	base_commit stringlengths 40 40	created_at stringdate 2026-01-02 19:04:12 2026-08-16 13:49:31	language stringclasses 13 values	license stringclasses 4 values	problem_statement stringlengths 155 15.7k	problem_statement_base stringlengths 30 15.2k	contract_addendum stringlengths 0 12.7k	patch stringlengths 1.28k 237k	test_patch stringlengths 984 406k	FAIL_TO_PASS listlengths 1 617	PASS_TO_PASS listlengths 0 1.64k	pr_url stringlengths 35 66	install_config dict
actorforth__actorforth-35	ActorForth/ActorForth	3e2370c65482191634ff049f0019b7481e657fe4	2026-04-16T20:20:43Z	Erlang	BSD-3-Clause	Core Erlang compilation target ## Goal Add a Core Erlang compilation target alongside the existing Erlang abstract-forms path (`af_word_compiler.erl`, `af_ring1.erl` native mode), selectable per-compile. ## Motivation - Core Erlang is the compiler's intermediate representation. Generating it directly gives finer con...	Core Erlang compilation target ## Goal Add a Core Erlang compilation target alongside the existing Erlang abstract-forms path (`af_word_compiler.erl`, `af_ring1.erl` native mode), selectable per-compile. ## Motivation - Core Erlang is the compiler's intermediate representation. Generating it directly gives finer con...	# Core Erlang target: module `af_core_target` (public contract) This delivery is the first stage of the Core Erlang compilation target. It introduces a new public module, `src/af_core_target.erl`, that turns word definitions (lists of `#operation{}` records from `include/operation.hrl`) into Core Erlang and compiles t...	diff --git a/src/af_core_target.erl b/src/af_core_target.erl new file mode 100644 index 0000000..5e1db05 --- /dev/null +++ b/src/af_core_target.erl @@ -0,0 +1,212 @@ +%% af_core_target.erl -- Core Erlang compilation target +%% +%% Experimental alternative to af_word_compiler's Erlang abstract-forms path. +%% Emits Core...	diff --git a/test/af_core_target_tests.erl b/test/af_core_target_tests.erl new file mode 100644 index 0000000..d64cd6b --- /dev/null +++ b/test/af_core_target_tests.erl @@ -0,0 +1,237 @@ +-module(af_core_target_tests). + +-include_lib("eunit/include/eunit.hrl"). +-include("operation.hrl"). + +%% Build a minimal #operat...	[ "af_core_target_tests:22: -supported_primitives_test_/0-fun-13-", "af_core_target_tests:23: -supported_primitives_test_/0-fun-11-", "af_core_target_tests:24: -supported_primitives_test_/0-fun-9-", "af_core_target_tests:25: -supported_primitives_test_/0-fun-7-", "af_core_target_tests:26: -supported_primitive...	[]	https://github.com/ActorForth/ActorForth/pull/35	{ "base_image_name": "erlang_base:v2026-08-18-erlang-otp27-python3dev-rebench", "install": [ "rebar3 compile" ], "log_parser": "parse_log_erlang", "test_cmd": "rebar3 eunit -v -m af_core_target_tests" }
agentjido__jido-153	agentjido/jido	b099b27c7629f5c419ae1fcacb79133f30bd5867	2026-02-14T15:46:22Z	Elixir	Apache-2.0	refactor(errors): normalize error contracts across Util/AgentServer/Storage ## Context Split from #128 audit umbrella. ## Problem Error contracts remain inconsistent across `Jido.Util`, `AgentServer`, and storage layers. ## Scope - Normalize return shapes for validation and runtime errors. - Standardize not-found sem...	refactor(errors): normalize error contracts across Util/AgentServer/Storage ## Context Split from #128 audit umbrella. ## Problem Error contracts remain inconsistent across `Jido.Util`, `AgentServer`, and storage layers. ## Scope - Normalize return shapes for validation and runtime errors. - Standardize not-found sem...		diff --git a/lib/jido.ex b/lib/jido.ex index 3e5dc62d..1f1e4060 100644 --- a/lib/jido.ex +++ b/lib/jido.ex @@ -187,7 +187,7 @@ defmodule Jido do end @doc "Thaw an agent from storage." - @spec thaw(module(), term()) :: {:ok, Jido.Agent.t()} \| :not_found \| {:error, term()} + @spec thaw(module(),...	diff --git a/test/examples/persistence/persistence_storage_test.exs b/test/examples/persistence/persistence_storage_test.exs index 60c284f..070a5d2 100644 --- a/test/examples/persistence/persistence_storage_test.exs +++ b/test/examples/persistence/persistence_storage_test.exs @@ -67,7 +67,7 @@ defmodule JidoExampleTest...	[ "multiple agents: hibernate/thaw multiple agents with different IDs thawing nonexistent agent returns :not_found", "thaw/3 returns {:error, :not_found} for missing agent", "validate_actions/2 with opts rejects invalid action module with opts", "validate_actions/2 with opts rejects invalid action list with opt...	[ "hibernate and thaw round-trip basic round-trip preserves agent state", "hibernate and thaw round-trip basic round-trip preserves agent state (excluded)", "hibernate and thaw round-trip thaw returns :not_found for non-existent agent", "hibernate and thaw round-trip thaw returns :not_found for non-existent age...	https://github.com/agentjido/jido/pull/153	{ "base_image_name": "elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench", "install": [ "env -u LD_LIBRARY_PATH mix deps.get" ], "log_parser": "parse_log_elixir", "test_cmd": "( env -u LD_LIBRARY_PATH MIX_ENV=test mix compile ) && env -u LD_LIBRARY_PATH MIX_ENV=test mix test --seed 0 --no-color test/exam...
agentjido__jido-286	agentjido/jido	cb43e3f9afc691d75538ce1953ed7dcedf6ae196	2026-05-19T22:19:05Z	Elixir	Apache-2.0	Attention control of agent ``` Right now sensors are wired to agents at startup and stay static. The agent receives whatever the sensor delivers. But what if the agent's current goal or strategy should change what it's paying attention to? Example: an agent in idle mode polls a metric every 60s. It detects an anomaly ...	Attention control of agent ``` Right now sensors are wired to agents at startup and stay static. The agent receives whatever the sensor delivers. But what if the agent's current goal or strategy should change what it's paying attention to? Example: an agent in idle mode polls a metric every 60s. It detects an anomaly ...	Jido supports attentional control: while an agent runs, it can dynamically change what it pays attention to — starting a sensor it was not watching, replacing or reconfiguring a sensor it already watches, and stopping a sensor entirely. Plugins can also declare sensor subscriptions, including several independently tagg...	diff --git a/README.md b/README.md index 30747eea..652b2480 100644 --- a/README.md +++ b/README.md @@ -94,7 +94,7 @@ Jido isn't "better GenServer" - it's a formalized agent pattern built on GenSe ### Directive-Based Effects - Actions transform state and may perform required work - Directives describe runtime-owned...	diff --git a/test/jido/agent/directive_test.exs b/test/jido/agent/directive_test.exs index 3cc79c9..fbc3553 100644 --- a/test/jido/agent/directive_test.exs +++ b/test/jido/agent/directive_test.exs @@ -111,6 +111,52 @@ defmodule JidoTest.Agent.DirectiveTest do end end + describe "start_sensor/3" do + test ...	[ "start_sensor/3 creates StartSensor directive with defaults", "start_sensor/3 creates StartSensor directive with lifecycle options", "stop_sensor/2 creates StopSensor directive with default reason", "stop_sensor/2 creates StopSensor directive with custom reason", "start/1 starts the runtime unlinked to the ...	[ "emit/2 creates Emit directive without dispatch", "emit/2 creates Emit directive with dispatch config", "error/2 creates Error directive without context", "error/2 creates Error directive with context", "spawn/2 creates Spawn directive without tag", "spawn/2 creates Spawn directive with tag", "spawn_age...	https://github.com/agentjido/jido/pull/286	{ "base_image_name": "elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench", "install": [ "unset LD_LIBRARY_PATH", "mix local.hex --force", "mix local.rebar --force", "mix deps.get" ], "log_parser": "parse_log_elixir", "test_cmd": "( unset LD_LIBRARY_PATH ) && ( mix compile ) && unset LD_LIBRAR...
agentjido__jido_ai-299	agentjido/jido_ai	4833c863cf80cd8b7b0db209e390eb117e63c195	2026-05-21T12:54:51Z	Elixir	Apache-2.0	"move AI runtime error contract ownership into Jido.AI.Error\n## Summary\n\n`jido_ai` has an AI-spec(...TRUNCATED)	"move AI runtime error contract ownership into Jido.AI.Error\n## Summary\n\n`jido_ai` has an AI-spec(...TRUNCATED)	"Additional observable requirements for the AI runtime error contract:\n\nCanonical envelope\n\n- Th(...TRUNCATED)	"diff --git a/guides/developer/error_model_and_recovery.md b/guides/developer/error_model_and_recove(...TRUNCATED)	"diff --git a/test/jido_ai/error/model_test.exs b/test/jido_ai/error/model_test.exs\nindex 59c0ae2e.(...TRUNCATED)	["LLM runtime paths generation failures produce a canonical :llm_error envelope","directive runtime (...TRUNCATED)	["LLM runtime paths task start failures emit a canonical retryable :supervisor envelope","api errors(...TRUNCATED)	https://github.com/agentjido/jido_ai/pull/299	{"base_image_name":"elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench","install":["unset LD_LIBRAR(...TRUNCATED)
agentjido__jido_ai-325	agentjido/jido_ai	0f8f7e0dcf98fc1453176effb200a4f25b0bce46	2026-07-15T01:26:22Z	Elixir	Apache-2.0	"[Open agent skills] Specifications drift\nWe are currently implementing Skills in ZAQ and wanted to(...TRUNCATED)	"[Open agent skills] Specifications drift\nWe are currently implementing Skills in ZAQ and wanted to(...TRUNCATED)	"This task requires turning the Open agent skills support from opt-in pieces into an\nautomatically (...TRUNCATED)	"diff --git a/examples/lib/skills/calculator.ex b/examples/lib/skills/calculator.ex\nindex f53413b4b(...TRUNCATED)	"diff --git a/test/jido_ai/agent_test.exs b/test/jido_ai/agent_test.exs\nindex 77487ce9..3c90ac70 10(...TRUNCATED)	["Agent macro agent_skills resolves at agent initialization instead of module compilation","Agent ma(...TRUNCATED)	["Agent macro agent has correct description","Agent macro agent has correct name","Agent macro compi(...TRUNCATED)	https://github.com/agentjido/jido_ai/pull/325	{"base_image_name":"elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench","install":["unset LD_LIBRAR(...TRUNCATED)
agentjido__req_llm-543	agentjido/req_llm	fa2791930623be6c50224c713fccf1eb2ddb3fa5	2026-03-24T12:24:22Z	Elixir	Apache-2.0	"[Feature]: Handle slightly broken JSON in structured outputs\n### Feature Type\n\nProvider feature (...TRUNCATED)	"[Feature]: Handle slightly broken JSON in structured outputs\n### Feature Type\n\nProvider feature (...TRUNCATED)	"Structured output from model providers may contain lightly malformed JSON\n(for example a trailing (...TRUNCATED)	"diff --git a/lib/req_llm/json.ex b/lib/req_llm/json.ex\nnew file mode 100644\nindex 000000000..cc00(...TRUNCATED)	"diff --git a/test/req_llm/generation_test.exs b/test/req_llm/generation_test.exs\nindex 24ba199..51(...TRUNCATED)	["JSON repair applies to ordinary tool names args_map repairs malformed arguments and honors strict (...TRUNCATED)	["Inspect implementation provides readable inspection format","Inspect implementation shows empty ar(...TRUNCATED)	https://github.com/agentjido/req_llm/pull/543	{"base_image_name":"elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench","install":["env -u LD_LIBRA(...TRUNCATED)
agentjido__req_llm-670	agentjido/req_llm	44331b8422c7df21bf8f68ae263bcb8a6502dc02	2026-05-05T01:04:31Z	Elixir	Apache-2.0	"Default generated token limit should use model output metadata\nFeedback from @zachdaniel: ReqLLM c(...TRUNCATED)	"Default generated token limit should use model output metadata\nFeedback from @zachdaniel: ReqLLM c(...TRUNCATED)		"diff --git a/lib/req_llm/provider/defaults.ex b/lib/req_llm/provider/defaults.ex\nindex 56a75064b..(...TRUNCATED)	"diff --git a/test/providers/anthropic_test.exs b/test/providers/anthropic_test.exs\nindex b13376d8.(...TRUNCATED)	["Amazon Bedrock streaming uses the model output metadata","Azure object preparation defaults from t(...TRUNCATED)	["Alibaba shared object preparation uses model output metadata","OpenRouter-specific features app at(...TRUNCATED)	https://github.com/agentjido/req_llm/pull/670	{"base_image_name":"elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench","install":["mix local.hex -(...TRUNCATED)
agentjido__req_llm-819	agentjido/req_llm	e2090172143ad2e97f91bdfddc56fb26f7413c08	2026-07-14T19:43:34Z	Elixir	Apache-2.0	"OpenAI provider ignores caller max_retries when attaching retry step\n## Summary\n\nIn ReqLLM 1.17.(...TRUNCATED)	"OpenAI provider ignores caller max_retries when attaching retry step\n## Summary\n\nIn ReqLLM 1.17.(...TRUNCATED)	"Contract clarifications for caller-supplied retry bounds:\n\n- Attaching the shared non-streaming r(...TRUNCATED)	"diff --git a/lib/req_llm/provider/defaults.ex b/lib/req_llm/provider/defaults.ex\nindex cd2d188c2..(...TRUNCATED)	"diff --git a/test/providers/openai_test.exs b/test/providers/openai_test.exs\nindex 0fef405a..7c110(...TRUNCATED)	["attach/1 preserves max_retries already configured on the request","Google Vertex caller retry limi(...TRUNCATED)	["attach/1 configures retry options on request","attach/2 honors max_retries in the real Req pipelin(...TRUNCATED)	https://github.com/agentjido/req_llm/pull/819	{"base_image_name":"elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench","install":["env -u LD_LIBRA(...TRUNCATED)
agentjido__req_llm-887	agentjido/req_llm	5d4fe8c5e0413113b54374ce4c29763c1e7e79e7	2026-07-17T02:51:14Z	Elixir	Apache-2.0	"V1: Add namespaced provider options alongside legacy options\nParent tracker: #829\nRoadmap: [ReqLL(...TRUNCATED)	"V1: Add namespaced provider options alongside legacy options\nParent tracker: #829\nRoadmap: [ReqLL(...TRUNCATED)	"# Provider-Keyed Option Namespace Contract\n\nThis addendum defines the observable requirements for(...TRUNCATED)	"diff --git a/guides/configuration.md b/guides/configuration.md\nindex e452b945..2c4ee61a 100644\n--(...TRUNCATED)	"diff --git a/test/req_llm/generation_test.exs b/test/req_llm/generation_test.exs\nindex 3ea15b9e..4(...TRUNCATED)	["namespace normalization accepts namespaced and legacy options for every V1 request operation","nam(...TRUNCATED)	["OCR tuple defaults change only the corresponding request key","Options.process/4 - core functional(...TRUNCATED)	https://github.com/agentjido/req_llm/pull/887	{"base_image_name":"elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench","install":["mix deps.get"],(...TRUNCATED)
agentjido__req_llm-898	agentjido/req_llm	d1f6c0c9ee01d4ef4d497e20e05773221ec5a6ab	2026-07-17T10:18:48Z	Elixir	Apache-2.0	"V1: Harden canonical tool exchange contracts\nParent tracker: #829\nRoadmap: [ReqLLM 1.x roadmap](h(...TRUNCATED)	"V1: Harden canonical tool exchange contracts\nParent tracker: #829\nRoadmap: [ReqLLM 1.x roadmap](h(...TRUNCATED)	"## Public interface additions and refinements\n\nThe hidden tests exercise the following observable(...TRUNCATED)	"diff --git a/lib/req_llm/context.ex b/lib/req_llm/context.ex\nindex 29552731f..e2b9e09eb 100644\n--(...TRUNCATED)	"diff --git a/test/provider/openai/responses_api_unit_test.exs b/test/provider/openai/responses_api_(...TRUNCATED)	["body encoding & context translation encode_body prefers explicit model-facing content over applica(...TRUNCATED)	["Collectable protocol into/1 allows collecting messages","Collectable protocol into/1 with empty co(...TRUNCATED)	https://github.com/agentjido/req_llm/pull/898	{"base_image_name":"elixir_base:v2026-07-02-elixir-1.18.4-otp27-rebench","install":["env -u LD_LIBRA(...TRUNCATED)

End of preview. Expand in Data Studio

Dataset Summary

Longtail-SWE is a benchmark of real-world software engineering tasks in 13 programming languages. Each task is based on a GitHub issue and its associated pull request. Given the issue and a repository at its base commit, an agent is expected to produce a patch that passes the evaluation tests.

The dataset contains 471 tasks across Clojure, Dart, Elixir, Erlang, F#, Haskell, Julia, Kotlin, Lua, OCaml, R, Scala, and Swift:

Split	Tasks	Repositories	Description
`hard`	202	129	The Longtail-SWE benchmark.
`easy`	269	168	An easier companion set.

Agent runs on the hard split are available in nebius/Longtail-SWE-trajectories. The saved run results use historical task versions and have not been regraded for every subsequent test change. Pin dataset revisions when comparing results; see source_release_manifest.json for changes to the task release.

Resources

How to use

from datasets import load_dataset

hard = load_dataset("nebius/Longtail-SWE", split="hard")
easy = load_dataset("nebius/Longtail-SWE", split="easy")

print(hard[0]["instance_id"], hard[0]["problem_statement"])

Dataset Structure

Both splits have the same fields:

Field name	Type	Description
`instance_id`	`str`	Unique task identifier.
`repo`	`str`	GitHub repository owner and name.
`base_commit`	`str`	Repository commit before the solution.
`created_at`	`str`	Task timestamp.
`language`	`str`	Programming language.
`license`	`str`	Source repository license.
`problem_statement`	`str`	Issue and any added requirements provided to the agent.
`problem_statement_base`	`str`	Base issue text.
`contract_addendum`	`str`	Additional task requirements, when present.
`patch`	`str`	Reference solution in unified diff format.
`test_patch`	`str`	Evaluation test patch in unified diff format.
`FAIL_TO_PASS`	`list[str]`	Tests expected to fail before the fix and pass after it.
`PASS_TO_PASS`	`list[str]`	Tests expected to pass before and after the fix.
`pr_url`	`str`	Source pull request URL.
`install_config`	`dict`	Base image, installation commands, test command, and log parser.

The reference patch and test patch are intended for evaluation, not as input to the agent. Task records include shell commands and should be executed only in isolated environments.

License

The dataset compilation is licensed under CC BY 4.0. Source repositories retain their own licenses; see the license field and LEGAL.md.

Changes in v1.1

hard: grading fixes for 5 tasks (test names, two test commands, a dead download URL, and a one-line hidden-test edit in scottarbeit__grace-136, whose FAIL_TO_PASS tests passed without any fix). With them the reference patch passes and an empty patch fails on all 202 tasks.
easy: 99 tasks got repaired hidden tests or test lists; 27 tasks were excluded (296 → 269). All exclusions and their reasons are in excluded_tasks.json; per-task details are in CHANGES_v1.1.md on GitHub.

Citation

Paper: Longtail-SWE: High-Quality SWE Tasks for Underrepresented Languages

@misc{dobrovolskii2026longtailswe,
  title         = {Longtail-SWE: High-Quality SWE Tasks for Underrepresented Languages},
  author        = {Dobrovolskii, Vladimir and Badertdinov, Ibragim and Golubev, Alexander},
  year          = {2026},
  eprint        = {2611.10304},
  archivePrefix = {arXiv},
  primaryClass  = {cs.SE},
  url           = {https://arxiv.org/abs/2611.10304}
}

Downloads last month: 6

Paper for nebius/Longtail-SWE

Longtail-SWE: High-Quality SWE Tasks for Underrepresented Languages

Paper • 2611.10304 • Published 4 days ago • 13