item_id stringlengths 21 26 | family stringclasses 383
values | twin_of stringlengths 0 21 | system stringlengths 1.51k 1.51k | user stringlengths 18.6k 288k | input_tokens int64 6.38k 123k | budget_tokens int64 1k 3.07k | cut int64 19 151 | steps int64 32 166 | resolved bool 1
class | image stringlengths 25 33 | source stringclasses 1
value | facts stringlengths 585 1.33k | probes stringlengths 1.29k 1.44k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
ctx_compress-s1-00000 | snippetdesk | You compact a coding agent's session so the agent can continue from your summary alone. The session is inside <session> tags: [user] messages, [assistant] text, [tool_use] calls with their kind and file, and [tool_result] outputs. Write the summary in plain markdown with exactly these nine headings, in this order, each... | <session>
[user]
wrap_slug_window produces doubled dashes for consecutive separators. It lives in src/snippetdesk/limits.py; `python -m pytest tests/test_limits.py -q` shows the failures. Fix it without changing the public signature and add a regression test.
[assistant]
I will start by mapping the snippetdesk layout.... | 31,317 | 1,565 | 56 | 67 | false | synth:5995469358269906430@1 | synth | [{"aliases": ["src/snippetdesk/limits.py", "limits.py"], "kind": "file", "value": "src/snippetdesk/limits.py"}, {"aliases": ["tests/test_limits.py", "test_limits.py"], "kind": "file", "value": "tests/test_limits.py"}, {"aliases": ["data/chunk_1.csv", "chunk_1.csv"], "kind": "file", "value": "data/chunk_1.csv"}, {"alias... | [{"answers": ["src/snippetdesk/limits.py", "limits.py"], "question": "Which file did the agent edit most recently?", "type": "exact"}, {"answers": ["yes"], "question": "Has the agent changed any existing file in the repository yet? Answer yes or no.", "type": "word"}, {"answers": ["pass"], "question": "Did the agent's ... | |
ctx_compress-s1-00000-twin | snippetdesk | ctx_compress-s1-00000 | You compact a coding agent's session so the agent can continue from your summary alone. The session is inside <session> tags: [user] messages, [assistant] text, [tool_use] calls with their kind and file, and [tool_result] outputs. Write the summary in plain markdown with exactly these nine headings, in this order, each... | <session>
[user]
wrap_slug_window produces doubled dashes for consecutive separators. It lives in src/snippetdesk/limits.py; `python -m pytest tests/test_limits.py -q` shows the failures. Fix it without changing the public signature and add a regression test.
[assistant]
I will start by mapping the snippetdesk layout.... | 18,066 | 1,000 | 40 | 67 | false | synth:5995469358269906430@1 | synth | [{"aliases": ["src/snippetdesk/limits.py", "limits.py"], "kind": "file", "value": "src/snippetdesk/limits.py"}, {"aliases": ["tests/test_limits.py", "test_limits.py"], "kind": "file", "value": "tests/test_limits.py"}, {"aliases": ["data/chunk_1.csv", "chunk_1.csv"], "kind": "file", "value": "data/chunk_1.csv"}, {"alias... | [{"answers": ["src/snippetdesk/limits.py", "limits.py"], "question": "Which file did the agent edit most recently?", "type": "exact"}, {"answers": ["yes"], "question": "Has the agent changed any existing file in the repository yet? Answer yes or no.", "type": "word"}, {"answers": ["pass"], "question": "Did the agent's ... |
ctx_compress-s1-00001 | draftledger | "You compact a coding agent's session so the agent can continue from your summary alone. The session(...TRUNCATED) | "<session>\n[user]\nfold_clause_window rejects a range where lo equals hi, which is a valid single p(...TRUNCATED) | 66,549 | 3,072 | 84 | 103 | false | synth:7735715960199495141@2.32595 | synth | "[{\"aliases\": [\"src/draftledger/core.py\", \"core.py\"], \"kind\": \"file\", \"value\": \"src/dra(...TRUNCATED) | "[{\"answers\": [\"no\"], \"question\": \"Has the agent changed any existing file in the repository (...TRUNCATED) | |
ctx_compress-s1-00001-twin | draftledger | ctx_compress-s1-00001 | "You compact a coding agent's session so the agent can continue from your summary alone. The session(...TRUNCATED) | "<session>\n[user]\nfold_clause_window rejects a range where lo equals hi, which is a valid single p(...TRUNCATED) | 40,704 | 2,035 | 55 | 103 | false | synth:7735715960199495141@2.32595 | synth | "[{\"aliases\": [\"src/draftledger/core.py\", \"core.py\"], \"kind\": \"file\", \"value\": \"src/dra(...TRUNCATED) | "[{\"answers\": [\"no\"], \"question\": \"Has the agent changed any existing file in the repository (...TRUNCATED) |
ctx_compress-s1-00002 | spanledger | "You compact a coding agent's session so the agent can continue from your summary alone. The session(...TRUNCATED) | "<session>\n[user]\nsmooth_counter_all raises ValueError when a value itself contains an equals sign(...TRUNCATED) | 65,234 | 3,072 | 99 | 124 | false | synth:3719260088301377541@2.30043 | synth | "[{\"aliases\": [\"src/spanledger/io.py\", \"io.py\"], \"kind\": \"file\", \"value\": \"src/spanledg(...TRUNCATED) | "[{\"answers\": [\"src/spanledger/io.py\", \"io.py\"], \"question\": \"Which file did the agent edit(...TRUNCATED) | |
ctx_compress-s1-00002-twin | spanledger | ctx_compress-s1-00002 | "You compact a coding agent's session so the agent can continue from your summary alone. The session(...TRUNCATED) | "<session>\n[user]\nsmooth_counter_all raises ValueError when a value itself contains an equals sign(...TRUNCATED) | 50,824 | 2,541 | 71 | 124 | false | synth:3719260088301377541@2.30043 | synth | "[{\"aliases\": [\"src/spanledger/io.py\", \"io.py\"], \"kind\": \"file\", \"value\": \"src/spanledg(...TRUNCATED) | "[{\"answers\": [\"src/spanledger/io.py\", \"io.py\"], \"question\": \"Which file did the agent edit(...TRUNCATED) |
ctx_compress-s1-00003 | sessionledger | "You compact a coding agent's session so the agent can continue from your summary alone. The session(...TRUNCATED) | "<session>\n[user]\nextend_grant lets defaults overwrite explicit overrides. It lives in src/session(...TRUNCATED) | 41,326 | 2,066 | 87 | 94 | false | synth:4826020194796927585@1 | synth | "[{\"aliases\": [\"src/sessionledger/audit.py\", \"audit.py\"], \"kind\": \"file\", \"value\": \"src(...TRUNCATED) | "[{\"answers\": [\"src/sessionledger/audit.py\", \"audit.py\"], \"question\": \"Which file did the a(...TRUNCATED) | |
ctx_compress-s1-00003-twin | sessionledger | ctx_compress-s1-00003 | "You compact a coding agent's session so the agent can continue from your summary alone. The session(...TRUNCATED) | "<session>\n[user]\nextend_grant lets defaults overwrite explicit overrides. It lives in src/session(...TRUNCATED) | 26,547 | 1,327 | 49 | 94 | false | synth:4826020194796927585@1 | synth | "[{\"aliases\": [\"src/sessionledger/audit.py\", \"audit.py\"], \"kind\": \"file\", \"value\": \"src(...TRUNCATED) | "[{\"answers\": [\"no\"], \"question\": \"Has the agent changed any existing file in the repository (...TRUNCATED) |
ctx_compress-s1-00004 | sessiondesk | "You compact a coding agent's session so the agent can continue from your summary alone. The session(...TRUNCATED) | "<session>\n[user]\nextend_grant raises KeyError for a missing key instead of returning the default.(...TRUNCATED) | 56,850 | 2,842 | 96 | 103 | false | synth:2636109939138424196@1.22435 | synth | "[{\"aliases\": [\"src/sessiondesk/audit.py\", \"audit.py\"], \"kind\": \"file\", \"value\": \"src/s(...TRUNCATED) | "[{\"answers\": [\"src/sessiondesk/audit.py\", \"audit.py\"], \"question\": \"Which file did the age(...TRUNCATED) | |
ctx_compress-s1-00004-twin | sessiondesk | ctx_compress-s1-00004 | "You compact a coding agent's session so the agent can continue from your summary alone. The session(...TRUNCATED) | "<session>\n[user]\nextend_grant raises KeyError for a missing key instead of returning the default.(...TRUNCATED) | 31,893 | 1,594 | 57 | 103 | false | synth:2636109939138424196@1.22435 | synth | "[{\"aliases\": [\"src/sessiondesk/audit.py\", \"audit.py\"], \"kind\": \"file\", \"value\": \"src/s(...TRUNCATED) | "[{\"answers\": [\"no\"], \"question\": \"Has the agent changed any existing file in the repository (...TRUNCATED) |
End of preview. Expand in Data Studio
Microtensor agent sessions
Coding agent sessions for the ctx_compress arena on Microtensor (Bittensor testnet netuid 576). Each row is one compaction request in the arena's exact format: the compaction instruction as the system message and the session, rendered canonically inside <session> tags, as the user message. The label holds the facts and probe questions the grader checks.
Every session is generated by Microtensor. A seed coins a Python project with one injected bug, and a scripted agent works the task: it lists the tree, searches, reads modules and tests, runs pytest, often tries a wrong fix and reverts it, applies the right fix and adds a regression test. Every tool result is the real output of the real command. Nothing here comes from a public dataset.
| Sessions | 2,000 |
| Rows | 3,981 (each session cut near its end, and a twin cut near its middle: 1,981 twins) |
| Input tokens | 6,381 to 122,704, median 45,439, Granite 4.0 H Micro tokenizer |
| Arena input cap | 122,880 tokens |
| Output | nine heading markdown summary within budget_tokens |
Hidden evaluation items come from the same generator under a secret seed and are never published.
Fields
| Field | Meaning |
|---|---|
item_id |
row id; a twin's id ends in -twin |
twin_of |
the full item a twin was cut from, empty otherwise |
family |
the generated project name |
system, user |
the request exactly as a miner's model receives it |
input_tokens, budget_tokens |
Granite token count of the request and the output budget it states |
cut, steps |
the step the session was cut at and its full length |
facts |
JSON list of files, errors and symbols the summary must keep |
probes |
JSON list of continuation questions with accepted answers |
image |
synth:<seed>@<scale>; the arenas package rebuilds the buggy project from it |
Use
from datasets import load_dataset
rows = load_dataset("microtensor-io/agent-sessions", split="dev")
There are no reference summaries. Write targets with a teacher model and keep those the arena grader passes; the miner guide shows the loop. Make more sessions with your own seed:
pip install "microtensor-arenas[compress] @ git+https://github.com/microtensor-io/microtensor-arenas"
mta dev --arena ctx_compress --count 2000 --seed 7 --out dev/
Licence
Microtensor dataset terms, see https://www.microtensor.cloud/terms.
- Downloads last month
- -