Datasets:
Tasks:
Text Classification
Modalities:
Text
Formats:
json
Languages:
English
Size:
10K - 100K
License:
state stringlengths 246 19.2k | questions unknown | _meta dict |
|---|---|---|
<task>
How to match exception properties?
I have a webapp with helper methods that raises `bottle.HTTPError` and I want to ensure that a raised exception has a `status_code` property, but it seems the calling/raises pair cannot directly assert this? E.g. something like
```
assert_that(
calling(helper).with_args(... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-swe-agent/train-00000-of-00012/5542@30",
"group_id": "hamcrest__PyHamcrest-129",
"traj_id": "nebius-swe-agent/train-00000-of-00012/5542",
"task_id": "hamcrest__PyHamcrest-129",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"scaffold": "swe-agent",
"policy... |
<task>
Many datastores and helpers assume created and modified fields exist in objects
As an example, `CompositeDataSource`s attempt to access the `modified` key directly as well as making use of the `deduplicate()` helper which attempts to access `modified` and then falls back to `created` if that's missing. The follo... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-openhands/chatcmpl-887400d4ee9c87d595bccde678d2f367@20",
"group_id": "oasis-open__cti-python-stix2-444",
"traj_id": "nebius-openhands/chatcmpl-887400d4ee9c87d595bccde678d2f367",
"task_id": "oasis-open__cti-python-stix2-444",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split": "te... |
<task>
incorrect column widths
```
import pyRestTable
table = pyRestTable.Table()
table.labels = ['h', 'k', 'l', 'mu', 'omega', 'chi', 'phi', 'gamma', 'delta']
table.addRow([0, 1, 1, 0.0, 3.4824458166048444, 22.712897698011936, 0.0, 0.0, 13.799774663132288])
table.addRow([0.0, 8.0, 0.0, 0.0, 22.31594087562736, 89.... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-openhands/chatcmpl-1998a6015fe1f5fcb264f419b29b2c60@22",
"group_id": "prjemian__pyRestTable-32",
"traj_id": "nebius-openhands/chatcmpl-1998a6015fe1f5fcb264f419b29b2c60",
"task_id": "prjemian__pyRestTable-32",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split": "test",
"domain":... |
<task>
Unable to creat 'Bundle' when using custom fields with SDO
Hi,
When attempting to generate a bundle, a failure message is created when passing an SDO with custom objects even with `allow_custom=True` set on the SDO object.
example:
`v = factory.create(
Vulnerability,
name="Test Vulnerability",
custom... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-swe-agent/train-00000-of-00012/816@12",
"group_id": "oasis-open__cti-python-stix2-74",
"traj_id": "nebius-swe-agent/train-00000-of-00012/816",
"task_id": "oasis-open__cti-python-stix2-74",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"scaffold": "swe-agent... |
<task>
Enable pattern matching by default
Now that the speed-up PR landed, it should be safe for us to enable pattern matching by default as the last grammar case to try. This would still make us initially test the 3.7+ and 3.0-3.6 grammars first before we try to run with the backtracking.
</task>
<policy>Qwen3-Coder-4... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-openhands/chatcmpl-044335d34a405d51157bf615261c8ae6@25",
"group_id": "psf__black-2758",
"traj_id": "nebius-openhands/chatcmpl-044335d34a405d51157bf615261c8ae6",
"task_id": "psf__black-2758",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split": "test",
"domain": "swe",
"scaffol... |
<task>
MetricFrame and multiple/named sensitive features
I was writing an example notebook, and I encountered something which I found strange. Basically, this works:
``` python
multi_metric = MetricFrame(
{
"average precision": skm.average_precision_score,
"roc-auc": skm.roc_auc_score,
... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-swe-agent/train-00008-of-00012/6303@4",
"group_id": "fairlearn__fairlearn-762",
"traj_id": "nebius-swe-agent/train-00008-of-00012/6303",
"task_id": "fairlearn__fairlearn-762",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"scaffold": "swe-agent",
"policy_... |
<task>
Render @topic wildcard mentions correctly when displayed in a message feed.
[CZO discussion](https://chat.zulip.org/#narrow/stream/378-api-design/topic/.40topic-mention.20data-user-id.20.2326195/near/1605762)
For stream wildcard mentions (all, everyone, stream), the rendered content is:
```
<span class="use... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-openhands/chatcmpl-0f9ce9768edff6af23a5f3e5767dbc9e@21",
"group_id": "zulip__zulip-terminal-1449",
"traj_id": "nebius-openhands/chatcmpl-0f9ce9768edff6af23a5f3e5767dbc9e",
"task_id": "zulip__zulip-terminal-1449",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split": "test",
"doma... |
<task>
AttributeError: 'NoneType' object has no attribute 'lower'
The following tag `<meta content="en-US" data-rh="true" itemprop="inLanguage"/>` causes an exception because it does not have `name` or `proprety` attribute.
```python
Traceback (most recent call last):
File "/opt/pycharm-professional/helpers/pyde... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-swe-agent/train-00008-of-00012/4404@8",
"group_id": "scottwernervt__favicon-17",
"traj_id": "nebius-swe-agent/train-00008-of-00012/4404",
"task_id": "scottwernervt__favicon-17",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"scaffold": "swe-agent",
"polic... |
<task>
Increase flexibility of setting seed
I was running some code that has randomness during evaluation. However I found that my, supposedly, random preprocessing was always done in exactly the same way.
This is related to setting the random seed by default to 12: https://github.com/pytorch/ignite/blob/master/igni... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-openhands/chatcmpl-0d6383691b59671653cb619faa395478@22",
"group_id": "pytorch__ignite-799",
"traj_id": "nebius-openhands/chatcmpl-0d6383691b59671653cb619faa395478",
"task_id": "pytorch__ignite-799",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split": "test",
"domain": "swe",
... |
<task>
ModelGenerator's output raises NameError for decimal fields
**Describe the bug**
Code generated by the `ModelGenerator` for schemas containing a decimal logical type field cannot run due to a missing import.
**To Reproduce**
```py
import dataclasses_avroschema
schema = {
"type": "record",
"nam... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-swe-agent/train-00002-of-00012/3876@4",
"group_id": "marcosschroh__dataclasses-avroschema-236",
"traj_id": "nebius-swe-agent/train-00002-of-00012/3876",
"task_id": "marcosschroh__dataclasses-avroschema-236",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"sc... |
<task>
Wrong type in JWE_EC.dec_setup
Hello cryptojwt team,
Thanks for your work, it usually works like a charm.
I found an issue in `cryptojwt.jwe.jwe_ec.JWE_EC.dec_setup`.
https://github.com/IdentityPython/JWTConnect-Python-CryptoJWT/blob/727bb9bff122a67e4e270330d990a7c1a1199664/src/cryptojwt/jwe/jwe_ec.py#L147
... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-openhands/chatcmpl-e3431ac1aa01f4048656d060f3dab6dc@30",
"group_id": "IdentityPython__JWTConnect-Python-CryptoJWT-170",
"traj_id": "nebius-openhands/chatcmpl-e3431ac1aa01f4048656d060f3dab6dc",
"task_id": "IdentityPython__JWTConnect-Python-CryptoJWT-170",
"source": "nebius/SWE-rebench-openhands... |
<task>
Create footer for report
When I use the `report` command
I want to get a footer with totals.
</task>
<policy>unknown</policy>
<recent>
## step 1
$ search_dir "report"
Found 21 matches for "report" in /accloudtant:
/accloudtant/CHANGELOG.md (1 matches)
/accloudtant/README.md (7 matches)
/accloudtant/accloudtant.e... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-swe-agent/train-00006-of-00012/1727@2",
"group_id": "ifosch__accloudtant-104",
"traj_id": "nebius-swe-agent/train-00006-of-00012/1727",
"task_id": "ifosch__accloudtant-104",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"scaffold": "swe-agent",
"policy_mo... |
<task>
Unnamed nodes should make up useful names for themselves
When you print a node, you get something like this:
```
<Node called "char_range" matching "c-a">
<Node called "class_char" matching "c">
<RegexNode matching "c">
<Node matching "-">
<Node called "class_char" matching "a">
<Reg... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-openhands/chatcmpl-c96a8293bcc36b0206345f9d81c560d4@12",
"group_id": "erikrose__parsimonious-113",
"traj_id": "nebius-openhands/chatcmpl-c96a8293bcc36b0206345f9d81c560d4",
"task_id": "erikrose__parsimonious-113",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split": "test",
"doma... |
<task>
ISO8601 DateTimes ending with Z considered not valid in 2.19.4
Probably related to #1247 and #1234 - in marshmallow `2.19.4`, with `python-dateutil` _not_ installed, it seems that loading a datetime in ISO8601 that ends in `Z` (UTC time) results in an error:
```python
class Foo(Schema):
date = DateTime(... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-openhands/chatcmpl-1cfc843fc6bae6cc97abff660b72482d@70",
"group_id": "marshmallow-code__marshmallow-1252",
"traj_id": "nebius-openhands/chatcmpl-1cfc843fc6bae6cc97abff660b72482d",
"task_id": "marshmallow-code__marshmallow-1252",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split":... |
<task>
ZT Crashes when viewing a message in browser in MacOS.
<img width="1031" alt="Screenshot 2024-02-15 at 9 44 55 PM" src="https://github.com/zulip/zulip-terminal/assets/71403193/deeadd45-decb-42f5-880c-30f709268a29">
ZT crashes when viewing a message in browser on MacOS.
My OS version = Ventura 13.6
Python ... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-swe-agent/train-00008-of-00012/3142@27",
"group_id": "zulip__zulip-terminal-1472",
"traj_id": "nebius-swe-agent/train-00008-of-00012/3142",
"task_id": "zulip__zulip-terminal-1472",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"scaffold": "swe-agent",
"po... |
<task>
jwt.decode returns Signature has expired
I expect the decoded result of JWT, but it returns an error `Signature has expired`
## Reproduction Steps
```python
import jwt
try:
payload: Dict = jwt.decode(token_obj[1], SECRET_KEY, algorithms=['HS256'], verify=True)
except Exception as err:
... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-openhands/chatcmpl-4a0021456afd5980bedd739a79148ca1@52",
"group_id": "jpadilla__pyjwt-657",
"traj_id": "nebius-openhands/chatcmpl-4a0021456afd5980bedd739a79148ca1",
"task_id": "jpadilla__pyjwt-657",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split": "test",
"domain": "swe",
... |
<task>
AttributeError: 'NoneType' object has no attribute 'fake'
**Describe the bug**
In my pydantic data model I have the following attribute `X: List[Optional[types.Float32]]`. When I call fake() of this class I get the error 'AttributeError: 'NoneType' object has no attribute 'fake''
**To Reproduce**
```py
f... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-swe-agent/train-00003-of-00012/5906@22",
"group_id": "marcosschroh__dataclasses-avroschema-244",
"traj_id": "nebius-swe-agent/train-00003-of-00012/5906",
"task_id": "marcosschroh__dataclasses-avroschema-244",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"s... |
<task>
Race condition between spinner thread and write()
When I use `write()` a lot on a spinner it often happens to me that the written text is displayed after the spinner message, like this:
```
[...]
noot noot
⠋ spinning...noot noot
noot noot
[...]
```
I used the script below to recreate the problem. It ta... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": true
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits ... | {
"id": "nebius-openhands/chatcmpl-9df30c5172dfd6b95b7270d355727ce3@41",
"group_id": "pavdmyt__yaspin-35",
"traj_id": "nebius-openhands/chatcmpl-9df30c5172dfd6b95b7270d355727ce3",
"task_id": "pavdmyt__yaspin-35",
"source": "nebius/SWE-rebench-openhands-trajectories",
"split": "test",
"domain": "swe",
"s... |
<task>
ds.connect() only connects to single project and runs overwritten [fusemachines]
Fuse machines run ds.connect() from a local machine but it only connects to a single project.
Also runs are getting overwritten when running on the same machine.
This text is just a placeholder: it was decided on the Fuse mach... | {
"p_success": {
"type": "noul",
"instructions": "Will this agent run end in success, meaning its final patch passes the task's hidden tests?",
"label": false
},
"stuck": {
"type": "noul",
"instructions": "Is the agent stuck: repeating the same actions, hitting the same error, or cycling edits... | {
"id": "nebius-swe-agent/train-00004-of-00012/5068@45",
"group_id": "dotmesh-io__dotscience-python-37",
"traj_id": "nebius-swe-agent/train-00004-of-00012/5068",
"task_id": "dotmesh-io__dotscience-python-37",
"source": "nebius/SWE-agent-trajectories",
"split": "test",
"domain": "swe",
"scaffold": "swe-a... |
End of preview. Expand in Data Studio
AgentCompass-Bench
A benchmark for value models of AI agents: models that look at an agent mid-task and predict, without running the agent further, whether it will succeed, whether it is stuck, which next action is best, whether to escalate, and how far it is from done.
Companion to agent-compass. Every state here comes from a repository or task that no agent-compass model saw in training.
Files
| file | domain | records | runs | questions |
|---|---|---|---|---|
test.jsonl |
software engineering, SWE-agent and OpenHands scaffolds, held-out repositories incl. SWE-bench Verified | 64,124 | 13,486 | p_success, stuck, progress, steps_left (successes), best_next (where candidates exist), escalate |
test_tau2.jsonl |
tool-use customer-service agents, tau2-bench, held-out tasks, 30 policy models | 3,832 | 840 | p_success, stuck, steps_left, best_next, escalate |
Record format is the agent-compass record (state text, questions map in System One shape, _meta with trajectory, task, policy, prefix position and outcome). See azlanmalikai/agent-compass-data for the full field description.
Metrics
Reported by scripts/eval_rows.py in the repository, with task-grouped bootstrap confidence intervals:
p_success: AUROC overall and by prefix quarter, Brier, ECE (15 bins), confident-error rate at p >= 0.9, abort recall at 5 / 10 / 25% false-positive rate.stuck,escalate: AUROC, accuracy, ECE.progress,steps_left: accuracy, within-one accuracy, MAE, ranked probability score.best_next: top-1 accuracy against the chance rate 1/K.- Offline agent simulations (
scripts/offline_best_of_n.py,scripts/offline_early_abort.py): within-policy best-of-N resolve-rate lift, and steps saved versus resolve-rate loss under a false-abort budget.
Recommended fixed subsets for cost: the first 20,000 records of test.jsonl (the files are shuffled) for every question, all of test_tau2.jsonl.
Reference results
| model | p_success AUROC (SWE) | p_success AUROC (tau2) | stuck | best_next top-1 |
|---|---|---|---|---|
| step-count heuristic | 0.614 (SWE-agent) | |||
| hand-feature GBT | 0.667 / 0.619 by scaffold | |||
| zero-shot Qwen3.5-2B instruct | 0.679 (5k) | |||
| agent-compass-2b v0 | 0.775 [0.746, 0.800] | |||
| agent-compass-2b v1 | 0.761 [0.735, 0.791] | 0.780 [0.726, 0.832] | 0.973 | 0.483 (chance 0.255); tau2 0.633 |
Full tables in docs/m3_report.md and docs/m4_report.md of the repository.
Rules
- Do not train on these repositories or tasks. The split table is
agent_compass/data/resources/repo_split.json. - Report AUROC with confidence intervals, and calibration (ECE, confident-error) alongside discrimination.
- For best-of-N claims, group by policy model; selecting the stronger LLM is not a value-model skill.
License
Apache-2.0 for the processing and labels. Trajectories keep their source licenses: the two Nebius datasets are CC-BY-4.0; AgentSuite/tau2-bench-trajectories states no license on its card (tau2-bench itself is MIT).
- Downloads last month
- 18