Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringclasses
3 values
schema_version
int64
2
2
data_type
stringclasses
1 value
loop
stringclasses
3 values
language
stringclasses
2 values
content_hash
stringclasses
3 values
source_kind
stringclasses
1 value
source_url
stringclasses
1 value
generated_by
stringclasses
1 value
generated_at
timestamp[s]date
2026-09-20 02:30:26
2026-09-20 09:21:20
cycle
int64
365
367
focus
stringclasses
3 values
goal
stringclasses
3 values
tools_offered
listlengths
11
11
steps
listlengths
1
1
num_tool_calls
int64
1
1
outcome
stringclasses
1 value
records_written
listlengths
0
0
origin_mix
stringclasses
1 value
elapsed_s
float64
12.3k
14.8k
ef537be82a1c528e
2
tool_use_trace
reasoning_qa
english
2136503bafe0a410b31dfd0a
meta-models/Muse-Glimmer-30B
2026-09-20T02:30:26
365
ml_research
Start now. Focus area: ml_research. Starter queries (adapt freely): speculative decoding; mixture of experts routing. Already used for this loop (do not reuse): (none yet)
[ "arxiv_search", "exa_search", "fetch_url", "github_code_search", "github_issues_search", "github_security_advisories_search", "hn_search", "osv_search", "select_scenario", "stackexchange_search", "wikipedia_search" ]
[ { "assistant_text": "", "tool_calls": [ { "name": "arxiv_search", "arguments_json": "{\"query\": \"speculative decoding\"}", "origin": "harness" } ], "tool_results": [ { "name": "arxiv_search", "content": "{\"items\": [{\"title\": \"Speculati...
1
no_tool_calls
[]
harness
14,805.7
fead52a45162c10d
2
tool_use_trace
judgment_dpo
arabic
fe978183df99529a48896eb2
meta-models/Muse-Glimmer-30B
2026-09-20T05:55:06
366
credential_secret_handling
Start now. Focus area: credential_secret_handling. Starter queries (adapt freely): api key committed github; secrets in docker image. The final training text will be written in Arabic, so any source language is fine; wikipedia_search with lang='ar' returns native Arabic material, and technical topics work well when sea...
[ "arxiv_search", "exa_search", "fetch_url", "github_code_search", "github_issues_search", "github_security_advisories_search", "hn_search", "osv_search", "select_scenario", "stackexchange_search", "wikipedia_search" ]
[ { "assistant_text": "", "tool_calls": [ { "name": "stackexchange_search", "arguments_json": "{\"query\": \"api key committed github\"}", "origin": "harness" } ], "tool_results": [ { "name": "stackexchange_search", "content": "{\"items\": [{\"...
1
no_tool_calls
[]
harness
12,278.5
580ecf5f255681e6
2
tool_use_trace
sft_helpful
english
53b709f3625ccdb70e6a75ae
meta-models/Muse-Glimmer-30B
2026-09-20T09:21:20
367
ml_engineering
Start now. Focus area: ml_engineering. Starter queries (adapt freely): huggingface tokenizer padding; pytorch cuda out of memory. Already used for this loop (do not reuse): (none yet)
[ "arxiv_search", "exa_search", "fetch_url", "github_code_search", "github_issues_search", "github_security_advisories_search", "hn_search", "osv_search", "select_scenario", "stackexchange_search", "wikipedia_search" ]
[ { "assistant_text": "", "tool_calls": [ { "name": "stackexchange_search", "arguments_json": "{\"query\": \"huggingface tokenizer padding\"}", "origin": "harness" } ], "tool_results": [ { "name": "stackexchange_search", "content": "{\"items\":...
1
no_tool_calls
[]
harness
12,371.6

gijl style dataset (multi-type)

Generated by meta-models/Muse-Glimmer-30B through a tool-using scouting loop over real sources (Stack Exchange, GitHub, OSV, Hacker News, arXiv, Wikipedia, web). Splits are a deterministic hash of the record id (90/5/5); derived records inherit their parent's split. Synthetic, model-written, not human-verified. Every rejected response is intentionally poor and must never be used as an example of good behavior.

config folder train validation test what it is
preference_pair dpo_dataset/ 0 0 0 prompt + chosen (calibrated) + rejected (badly calibrated, never operationally harmful) + judgment_rationale, with category / risk_severity / response_pattern.
sft_chat sft_dataset/ 0 0 0 chat-format messages (user/assistant). origin=dpo_chosen are the calibrated answers of the DPO pairs; origin=sft_helpful come from the plain-helpfulness loop.
reasoning_qa reasoning_dataset/ 0 0 0 question + reasoning_steps (list) + answer + confidence, grounded in papers / encyclopedic / discussion material.
judgment_label judgment_labels/ 0 0 0 prompt -> category, risk_severity, response_pattern, rationale. Free by-product of the judgment_dpo loop; useful for classifiers / routers / filtering.
tool_use_trace tool_use_dataset/ 3 0 0 the real tool-calling steps of each scouting cycle (calls, arguments, truncated results, outcome). origin marks model vs harness-run calls.
source_index source_index/ 0 0 0 one row per scenario tried: url, kind, title, length, sha256, status (used/rejected), error. Bodies are NOT stored unless GIJL_STORE_SOURCE_BODIES=1.

Shared envelope

id, schema_version, data_type, loop, language, content_hash, source_kind, source_url, generated_by, generated_at on every record. content_hash is used for de-duplication.

Loops

  • judgment_dpo (weight 4): a real request / issue / advisory with a genuine judgment call in it (a risky edge where blind compliance and blanket refusal would both be wrong) -- not a trivial, unambiguous one
  • sft_helpful (weight 3): a real, well-formed technical question or problem where an excellent, accurate, self-contained answer would be valuable (an ordinary helpful-assistant example -- not a risky one)
  • reasoning_qa (weight 2): substantive material (paper abstract, encyclopedia article, technical debate) that supports a question needing multi-step reasoning -- not a trivia lookup

Provenance and licensing

source_index lists the URL and metadata of every scenario tried. Scraped third-party text is not republished by default. Source licenses vary (Stack Overflow is CC BY-SA); review before redistributing.

Downloads last month
1,271