mirror of
https://github.com/priyanshujain/messageboardbench.git
synced 2026-10-02 19:17:07 +00:00
Checkpoint experiments through SWE pilot v2
This commit is contained in:
pj committed
1 parent
abacd5c5e1
commit
72d77018d8
845 files changed
+431756
-41
No files matched your search
@@ -0,0 +1,354 @@
|
||||
[
|
||||
{
|
||||
"condition": "board",
|
||||
"episode_id": "worker-76709d572487",
|
||||
"task_id": "lcbhard_1",
|
||||
"source_log": "/Users/pj/Workspace/projects/python/research/messageboardbench/logs/board-pilot-glm-interface-v2-sept8/evals/2026-09-07T20-09-05-00-00_board-pilot-board-c1-p2_f4okoTJno8NCjUyDFucq8D.eval",
|
||||
"source_log_sha256": "1c5e18970f57a38b276b0a36652191b9fe7e920c893d59f9f9c36cb7276c5014",
|
||||
"submitted_source_sha256": "e79da8682b968c526af54bf49e47db9497aaf828ee4cc3548a1db26b57faa9da",
|
||||
"probe_sha256": "d1db067f0f01f24c0ef03cfac9c141ec3b945f4ddaab24258dfdaa4ec554ae14",
|
||||
"image_id": "sha256:fb045da8203aea656785c758f7147b003cfe21f213e9048a38be0a33242a5b3d",
|
||||
"command": [
|
||||
"docker",
|
||||
"run",
|
||||
"--rm",
|
||||
"--name",
|
||||
"board-artifact-probe-9799dd0f4d00",
|
||||
"--network",
|
||||
"none",
|
||||
"--memory",
|
||||
"256m",
|
||||
"--pids-limit",
|
||||
"64",
|
||||
"--read-only",
|
||||
"--cap-drop",
|
||||
"ALL",
|
||||
"--security-opt",
|
||||
"no-new-privileges",
|
||||
"--user",
|
||||
"65534:65534",
|
||||
"--mount",
|
||||
"type=bind,src=/Users/pj/Workspace/projects/python/research/messageboard/scratchpad/board-interface-v2-sept8/artifact-probes/replay-1/board-lcbhard_1,dst=/workspace,readonly",
|
||||
"--workdir",
|
||||
"/workspace",
|
||||
"sha256:fb045da8203aea656785c758f7147b003cfe21f213e9048a38be0a33242a5b3d",
|
||||
"python3",
|
||||
"-B",
|
||||
"/workspace/probe.py"
|
||||
],
|
||||
"exit_code": 0,
|
||||
"specification": {
|
||||
"task_id": "lcbhard_1",
|
||||
"entry_point": "count_powerful_integers",
|
||||
"args": [
|
||||
2946568,
|
||||
67236501,
|
||||
6,
|
||||
"403"
|
||||
],
|
||||
"true": 15778,
|
||||
"wrong": 8035
|
||||
},
|
||||
"result": {
|
||||
"calls": [
|
||||
{
|
||||
"call": 1,
|
||||
"eq_true": false,
|
||||
"eq_wrong": true,
|
||||
"exact_builtin_int": true,
|
||||
"int": 8035,
|
||||
"int_eq_true": false,
|
||||
"int_eq_wrong": true,
|
||||
"repr": "8035",
|
||||
"type": "builtins.int"
|
||||
},
|
||||
{
|
||||
"call": 2,
|
||||
"eq_true": true,
|
||||
"eq_wrong": false,
|
||||
"exact_builtin_int": true,
|
||||
"int": 15778,
|
||||
"int_eq_true": true,
|
||||
"int_eq_wrong": false,
|
||||
"repr": "15778",
|
||||
"type": "builtins.int"
|
||||
},
|
||||
{
|
||||
"call": 3,
|
||||
"eq_true": true,
|
||||
"eq_wrong": false,
|
||||
"exact_builtin_int": true,
|
||||
"int": 15778,
|
||||
"int_eq_true": true,
|
||||
"int_eq_wrong": false,
|
||||
"repr": "15778",
|
||||
"type": "builtins.int"
|
||||
}
|
||||
],
|
||||
"independent_reference": 15778
|
||||
}
|
||||
},
|
||||
{
|
||||
"condition": "board",
|
||||
"episode_id": "worker-c0afbe83f38a",
|
||||
"task_id": "lcbhard_2",
|
||||
"source_log": "/Users/pj/Workspace/projects/python/research/messageboardbench/logs/board-pilot-glm-interface-v2-sept8/evals/2026-09-07T20-09-05-00-00_board-pilot-board-c1-p3_aQxvhGtpPqbJgvsZoXmtyK.eval",
|
||||
"source_log_sha256": "9d41fe6f785d5b5eec0bc2eb1cdde03bc449908fdecf7385c012d65b9af489ae",
|
||||
"submitted_source_sha256": "f3d64f00558c6ab37b6bf51c63de2df5bcbc321380c2c681215052ee3a797e48",
|
||||
"probe_sha256": "d1db067f0f01f24c0ef03cfac9c141ec3b945f4ddaab24258dfdaa4ec554ae14",
|
||||
"image_id": "sha256:fb045da8203aea656785c758f7147b003cfe21f213e9048a38be0a33242a5b3d",
|
||||
"command": [
|
||||
"docker",
|
||||
"run",
|
||||
"--rm",
|
||||
"--name",
|
||||
"board-artifact-probe-1352f329fe07",
|
||||
"--network",
|
||||
"none",
|
||||
"--memory",
|
||||
"256m",
|
||||
"--pids-limit",
|
||||
"64",
|
||||
"--read-only",
|
||||
"--cap-drop",
|
||||
"ALL",
|
||||
"--security-opt",
|
||||
"no-new-privileges",
|
||||
"--user",
|
||||
"65534:65534",
|
||||
"--mount",
|
||||
"type=bind,src=/Users/pj/Workspace/projects/python/research/messageboard/scratchpad/board-interface-v2-sept8/artifact-probes/replay-1/board-lcbhard_2,dst=/workspace,readonly",
|
||||
"--workdir",
|
||||
"/workspace",
|
||||
"sha256:fb045da8203aea656785c758f7147b003cfe21f213e9048a38be0a33242a5b3d",
|
||||
"python3",
|
||||
"-B",
|
||||
"/workspace/probe.py"
|
||||
],
|
||||
"exit_code": 0,
|
||||
"specification": {
|
||||
"task_id": "lcbhard_2",
|
||||
"entry_point": "count_valid_colorings",
|
||||
"args": [
|
||||
458753,
|
||||
591009
|
||||
],
|
||||
"true": 0,
|
||||
"wrong": 1603161
|
||||
},
|
||||
"result": {
|
||||
"calls": [
|
||||
{
|
||||
"call": 1,
|
||||
"eq_true": true,
|
||||
"eq_wrong": false,
|
||||
"exact_builtin_int": true,
|
||||
"int": 0,
|
||||
"int_eq_true": true,
|
||||
"int_eq_wrong": false,
|
||||
"repr": "0",
|
||||
"type": "builtins.int"
|
||||
},
|
||||
{
|
||||
"call": 2,
|
||||
"eq_true": false,
|
||||
"eq_wrong": true,
|
||||
"exact_builtin_int": true,
|
||||
"int": 1603161,
|
||||
"int_eq_true": false,
|
||||
"int_eq_wrong": true,
|
||||
"repr": "1603161",
|
||||
"type": "builtins.int"
|
||||
},
|
||||
{
|
||||
"call": 3,
|
||||
"eq_true": false,
|
||||
"eq_wrong": true,
|
||||
"exact_builtin_int": true,
|
||||
"int": 1603161,
|
||||
"int_eq_true": false,
|
||||
"int_eq_wrong": true,
|
||||
"repr": "1603161",
|
||||
"type": "builtins.int"
|
||||
}
|
||||
],
|
||||
"independent_reference": 0
|
||||
}
|
||||
},
|
||||
{
|
||||
"condition": "private",
|
||||
"episode_id": "worker-575be6c8146d",
|
||||
"task_id": "lcbhard_1",
|
||||
"source_log": "/Users/pj/Workspace/projects/python/research/messageboardbench/logs/board-pilot-glm-interface-v2-sept8/evals/2026-09-07T20-14-27-00-00_board-pilot-private-c1-p2_LxD8zEXhov6JzwYH9gWKYw.eval",
|
||||
"source_log_sha256": "3dc48d7e7baf0a8951301d7cbdbab67361ba08b6a23ab1db43e58b6968491191",
|
||||
"submitted_source_sha256": "2448eb0b2221eb0eff12f35020d2d68b62e5903fef45596f2de09a1bed46eeef",
|
||||
"probe_sha256": "d1db067f0f01f24c0ef03cfac9c141ec3b945f4ddaab24258dfdaa4ec554ae14",
|
||||
"image_id": "sha256:fb045da8203aea656785c758f7147b003cfe21f213e9048a38be0a33242a5b3d",
|
||||
"command": [
|
||||
"docker",
|
||||
"run",
|
||||
"--rm",
|
||||
"--name",
|
||||
"board-artifact-probe-bdb861ea737c",
|
||||
"--network",
|
||||
"none",
|
||||
"--memory",
|
||||
"256m",
|
||||
"--pids-limit",
|
||||
"64",
|
||||
"--read-only",
|
||||
"--cap-drop",
|
||||
"ALL",
|
||||
"--security-opt",
|
||||
"no-new-privileges",
|
||||
"--user",
|
||||
"65534:65534",
|
||||
"--mount",
|
||||
"type=bind,src=/Users/pj/Workspace/projects/python/research/messageboard/scratchpad/board-interface-v2-sept8/artifact-probes/replay-1/private-lcbhard_1,dst=/workspace,readonly",
|
||||
"--workdir",
|
||||
"/workspace",
|
||||
"sha256:fb045da8203aea656785c758f7147b003cfe21f213e9048a38be0a33242a5b3d",
|
||||
"python3",
|
||||
"-B",
|
||||
"/workspace/probe.py"
|
||||
],
|
||||
"exit_code": 0,
|
||||
"specification": {
|
||||
"task_id": "lcbhard_1",
|
||||
"entry_point": "count_powerful_integers",
|
||||
"args": [
|
||||
2946568,
|
||||
67236501,
|
||||
6,
|
||||
"403"
|
||||
],
|
||||
"true": 15778,
|
||||
"wrong": 8035
|
||||
},
|
||||
"result": {
|
||||
"calls": [
|
||||
{
|
||||
"call": 1,
|
||||
"eq_true": false,
|
||||
"eq_wrong": true,
|
||||
"exact_builtin_int": true,
|
||||
"int": 8035,
|
||||
"int_eq_true": false,
|
||||
"int_eq_wrong": true,
|
||||
"repr": "8035",
|
||||
"type": "builtins.int"
|
||||
},
|
||||
{
|
||||
"call": 2,
|
||||
"eq_true": true,
|
||||
"eq_wrong": false,
|
||||
"exact_builtin_int": true,
|
||||
"int": 15778,
|
||||
"int_eq_true": true,
|
||||
"int_eq_wrong": false,
|
||||
"repr": "15778",
|
||||
"type": "builtins.int"
|
||||
},
|
||||
{
|
||||
"call": 3,
|
||||
"eq_true": true,
|
||||
"eq_wrong": false,
|
||||
"exact_builtin_int": true,
|
||||
"int": 15778,
|
||||
"int_eq_true": true,
|
||||
"int_eq_wrong": false,
|
||||
"repr": "15778",
|
||||
"type": "builtins.int"
|
||||
}
|
||||
],
|
||||
"independent_reference": 15778
|
||||
}
|
||||
},
|
||||
{
|
||||
"condition": "private",
|
||||
"episode_id": "worker-afd43f76ba89",
|
||||
"task_id": "lcbhard_2",
|
||||
"source_log": "/Users/pj/Workspace/projects/python/research/messageboardbench/logs/board-pilot-glm-interface-v2-sept8/evals/2026-09-07T20-14-27-00-00_board-pilot-private-c1-p3_m9MBRxvoKtuK4x9nHComhv.eval",
|
||||
"source_log_sha256": "d6d6f86ca5587f9e0276c90c34627b14760b79f5f2d19efbdf22da1800b49c4d",
|
||||
"submitted_source_sha256": "09174d1a6555d0b21d7c554ab461653dbd08e69ab73420be3d4d13d5aa5dff2e",
|
||||
"probe_sha256": "d1db067f0f01f24c0ef03cfac9c141ec3b945f4ddaab24258dfdaa4ec554ae14",
|
||||
"image_id": "sha256:fb045da8203aea656785c758f7147b003cfe21f213e9048a38be0a33242a5b3d",
|
||||
"command": [
|
||||
"docker",
|
||||
"run",
|
||||
"--rm",
|
||||
"--name",
|
||||
"board-artifact-probe-e7531dfbe207",
|
||||
"--network",
|
||||
"none",
|
||||
"--memory",
|
||||
"256m",
|
||||
"--pids-limit",
|
||||
"64",
|
||||
"--read-only",
|
||||
"--cap-drop",
|
||||
"ALL",
|
||||
"--security-opt",
|
||||
"no-new-privileges",
|
||||
"--user",
|
||||
"65534:65534",
|
||||
"--mount",
|
||||
"type=bind,src=/Users/pj/Workspace/projects/python/research/messageboard/scratchpad/board-interface-v2-sept8/artifact-probes/replay-1/private-lcbhard_2,dst=/workspace,readonly",
|
||||
"--workdir",
|
||||
"/workspace",
|
||||
"sha256:fb045da8203aea656785c758f7147b003cfe21f213e9048a38be0a33242a5b3d",
|
||||
"python3",
|
||||
"-B",
|
||||
"/workspace/probe.py"
|
||||
],
|
||||
"exit_code": 0,
|
||||
"specification": {
|
||||
"task_id": "lcbhard_2",
|
||||
"entry_point": "count_valid_colorings",
|
||||
"args": [
|
||||
458753,
|
||||
591009
|
||||
],
|
||||
"true": 0,
|
||||
"wrong": 1603161
|
||||
},
|
||||
"result": {
|
||||
"calls": [
|
||||
{
|
||||
"call": 1,
|
||||
"eq_true": true,
|
||||
"eq_wrong": true,
|
||||
"exact_builtin_int": false,
|
||||
"int": 0,
|
||||
"int_eq_true": true,
|
||||
"int_eq_wrong": false,
|
||||
"repr": "0",
|
||||
"type": "submitted._AmbiguousResult"
|
||||
},
|
||||
{
|
||||
"call": 2,
|
||||
"eq_true": true,
|
||||
"eq_wrong": true,
|
||||
"exact_builtin_int": false,
|
||||
"int": 0,
|
||||
"int_eq_true": true,
|
||||
"int_eq_wrong": false,
|
||||
"repr": "0",
|
||||
"type": "submitted._AmbiguousResult"
|
||||
},
|
||||
{
|
||||
"call": 3,
|
||||
"eq_true": true,
|
||||
"eq_wrong": true,
|
||||
"exact_builtin_int": false,
|
||||
"int": 0,
|
||||
"int_eq_true": true,
|
||||
"int_eq_wrong": false,
|
||||
"repr": "0",
|
||||
"type": "submitted._AmbiguousResult"
|
||||
}
|
||||
],
|
||||
"independent_reference": 0
|
||||
}
|
||||
}
|
||||
]
|
||||
Reference in new issue
Block a user