serendipity-zk commited on
Commit
527707a
·
verified ·
1 Parent(s): a2c7fc4

Publish TraceLab dataset v0.0.2

Browse files

Mirror the byte-identical v0.0.2 GitHub release JSONL and DuckDB assets with a Hugging Face Dataset Card, license, and checksums.

.gitattributes CHANGED
@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ artifacts/v0.0.2/syfi_coding_trace.duckdb filter=lfs diff=lfs merge=lfs -text
LICENSE-DATASET.md ADDED
@@ -0,0 +1,42 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Dataset License — Creative Commons Attribution 4.0 International (CC BY 4.0)
2
+
3
+ The **TraceLab public coding-agent trace dataset** — the sanitized, normalized JSONL
4
+ trace and the prebuilt DuckDB database distributed as GitHub Release assets (e.g.
5
+ `syfi_coding_trace.jsonl.gz` and `syfi_coding_trace.duckdb`) — is licensed under the
6
+ **Creative Commons Attribution 4.0 International License (CC BY 4.0)**.
7
+
8
+ > Note: This dataset license applies to the **released trace data only**. All source
9
+ > code in this repository (collection, sanitization, analysis, validation, and the web
10
+ > app) is licensed separately under the Apache License 2.0 — see [`LICENSE`](LICENSE).
11
+
12
+ ## You are free to
13
+
14
+ - **Share** — copy and redistribute the material in any medium or format.
15
+ - **Adapt** — remix, transform, and build upon the material for any purpose, even
16
+ commercially.
17
+
18
+ The licensor cannot revoke these freedoms as long as you follow the license terms.
19
+
20
+ ## Under the following terms
21
+
22
+ - **Attribution** — You must give appropriate credit, provide a link to the license,
23
+ and indicate if changes were made. You may do so in any reasonable manner, but not in
24
+ any way that suggests the licensor endorses you or your use.
25
+ - **No additional restrictions** — You may not apply legal terms or technological
26
+ measures that legally restrict others from doing anything the license permits.
27
+
28
+ Full legal text: https://creativecommons.org/licenses/by/4.0/legalcode
29
+
30
+ ## How to attribute
31
+
32
+ When you use the dataset, please credit **TraceLab (SyFI Lab, University of Washington)**
33
+ and link back to <https://tracelab.cs.washington.edu>. A formal citation entry will be
34
+ added later.
35
+
36
+ ## Privacy & responsible use
37
+
38
+ The dataset is built from real Claude Code / Codex sessions that have been **sanitized**:
39
+ session, round, turn, tool-call, project, and user identifiers are replaced with stable
40
+ pseudonyms, and local context (paths, `cwd`, tool inputs, etc.) is stripped before
41
+ release. Even so, please use the data responsibly and do not attempt to re-identify
42
+ contributors.
NOTICE ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ TraceLab
2
+ Copyright 2026 SyFI Lab, University of Washington
3
+
4
+ This product includes software developed at the SyFI Lab, University of Washington
5
+ (https://tracelab.cs.washington.edu).
6
+
7
+ The source code in this repository is licensed under the Apache License, Version 2.0
8
+ (see LICENSE).
9
+
10
+ The public coding-agent trace dataset distributed via GitHub Releases (the normalized
11
+ JSONL trace and the prebuilt DuckDB database) is licensed separately under the
12
+ Creative Commons Attribution 4.0 International License (CC BY 4.0). See LICENSE-DATASET.md.
README.md ADDED
@@ -0,0 +1,145 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: "TraceLab Coding Agent Workload Traces"
3
+ license: cc-by-4.0
4
+ tags:
5
+ - agents
6
+ - coding
7
+ - traces
8
+ - llm-serving
9
+ - tabular
10
+ - datasets
11
+ size_categories:
12
+ - 100K<n<1M
13
+ source_datasets:
14
+ - original
15
+ configs:
16
+ - config_name: default
17
+ default: true
18
+ data_files:
19
+ - split: train
20
+ path: "data/v0.0.2/syfi_coding_trace.jsonl.gz"
21
+ ---
22
+
23
+ # TraceLab Coding Agent Workload Traces
24
+
25
+ TraceLab is a sanitized, real-world workload trace of Claude Code and Codex. Each JSONL row
26
+ represents one LLM invocation and retains the serving-relevant structure needed to study sessions,
27
+ token accounting, prompt-cache behavior, tool calls, and timing.
28
+
29
+ - **665,453** agent steps
30
+ - **8,058** sessions
31
+ - **743,819** normalized tool calls
32
+ - **52** pseudonymous users
33
+ - Providers: Claude Code and Codex
34
+ - Release: **v0.0.2**
35
+
36
+ Project links:
37
+
38
+ - [Paper: TraceLab: Characterizing Coding Agent Workloads for LLM Serving](https://arxiv.org/abs/2606.30560)
39
+ - [GitHub repository](https://github.com/uw-syfi/TraceLab)
40
+ - [Interactive website](https://tracelab.cs.washington.edu/)
41
+ - [Pinned GitHub release v0.0.2](https://github.com/uw-syfi/TraceLab/releases/tag/v0.0.2)
42
+
43
+ ## Load the dataset
44
+
45
+ ```python
46
+ from datasets import load_dataset
47
+
48
+ dataset = load_dataset("UW-SyFI/TraceLab", split="train")
49
+ print(dataset.num_rows)
50
+ print(dataset.features)
51
+ ```
52
+
53
+ Pin the published snapshot for reproducible studies:
54
+
55
+ ```python
56
+ dataset = load_dataset("UW-SyFI/TraceLab", revision="v0.0.2", split="train")
57
+ ```
58
+
59
+ The repository also includes the query-ready DuckDB database as an auxiliary artifact:
60
+
61
+ ```python
62
+ from huggingface_hub import hf_hub_download
63
+ import duckdb
64
+
65
+ database_path = hf_hub_download(
66
+ repo_id="UW-SyFI/TraceLab",
67
+ repo_type="dataset",
68
+ filename="artifacts/v0.0.2/syfi_coding_trace.duckdb",
69
+ revision="v0.0.2",
70
+ )
71
+ connection = duckdb.connect(database_path, read_only=True)
72
+ print(connection.execute("SELECT provider, count(*) FROM rounds GROUP BY provider").fetchall())
73
+ ```
74
+
75
+ ## Data format
76
+
77
+ Top-level fields include pseudonymous provider/session/round identifiers, model, token counts,
78
+ cache-prefix accounting, source store, an ordered `timing_events` list, a nested `tools` list, and a
79
+ stable `trace_key`.
80
+
81
+ Important token fields include:
82
+
83
+ - `input_tokens_total`: total input tokens reported for the invocation.
84
+ - `prefix_tokens`: provider-reported cached input tokens.
85
+ - `newly_append_tokens`: non-cached input tokens appended for the invocation.
86
+ - `output_tokens`: generated output tokens.
87
+
88
+ `tools[]` contains privacy-safe structural metadata such as tool name, pseudonymous tool-call id,
89
+ timestamps, input/result character counts, latency, error status, executable labels, and command
90
+ skeletons. Full tool inputs and outputs are not published.
91
+
92
+ ## Sanitization and responsible use
93
+
94
+ The public trace is generated by TraceLab's sanitization pipeline. User, project, session, round,
95
+ turn, tool-call, and process-session identifiers are replaced with stable pseudonyms. Local paths,
96
+ working directories, source session paths, tool inputs, prompts, responses, and full tool outputs are
97
+ removed. Custom executable and integration names are replaced with stable generic labels unless they
98
+ belong to an explicitly reviewed public allowlist.
99
+
100
+ This is a trace of real developer activity. Do not attempt to re-identify contributors or infer their
101
+ projects from the remaining workload structure.
102
+
103
+ ## Known limitations
104
+
105
+ - Consecutive `round_index` values are consecutive normalized records, but should not always be
106
+ interpreted as a single reconstructable linear conversation lineage.
107
+ - A small fraction of adjacent transitions report `prefix_tokens` larger than the immediately
108
+ reconstructable previous context. For synthetic session-local KV-cache replay, conservatively cap
109
+ the next reusable prefix at `previous.input_tokens_total + previous.output_tokens`. See
110
+ [TraceLab issue #22](https://github.com/uw-syfi/TraceLab/issues/22).
111
+ - Provider tokenization, message framing, cache boundaries, retries, resumes, branches, compaction,
112
+ and historical extractor behavior can affect adjacent-round accounting.
113
+ - The sanitized analytical rows intentionally do not contain the message text required for a
114
+ conversational Agent Trace Viewer.
115
+
116
+ ## Files and integrity
117
+
118
+ | File | SHA-256 |
119
+ |---|---|
120
+ | `data/v0.0.2/syfi_coding_trace.jsonl.gz` | `11ce51ec0a25e3d1d95b025bca2f7d1647e47571eb7cc968acd5fc64d4b4fb65` |
121
+ | `artifacts/v0.0.2/syfi_coding_trace.duckdb` | `a7bab286bc640844560850965ccf47975cf66407154132abaab90f27ec9be744` |
122
+
123
+ The JSONL and DuckDB files are byte-identical to the assets published in the TraceLab v0.0.2 GitHub
124
+ release.
125
+
126
+ ## License
127
+
128
+ The released dataset is licensed under
129
+ [Creative Commons Attribution 4.0 International](LICENSE-DATASET.md). Please credit **TraceLab
130
+ (SyFI Lab, University of Washington)** and link to <https://tracelab.cs.washington.edu/>. TraceLab
131
+ source code is separately licensed under Apache-2.0 in the GitHub repository.
132
+
133
+ ## Citation
134
+
135
+ ```bibtex
136
+ @misc{zhu2026tracelabcharacterizingcodingagent,
137
+ title = {TraceLab: Characterizing Coding Agent Workloads for LLM Serving},
138
+ author = {Kan Zhu and Mathew Jacob and Chenxi Ma and Yi Pan and Stephanie Wang and Arvind Krishnamurthy and Baris Kasikci},
139
+ year = {2026},
140
+ eprint = {2606.30560},
141
+ archivePrefix = {arXiv},
142
+ primaryClass = {cs.LG},
143
+ url = {https://arxiv.org/abs/2606.30560}
144
+ }
145
+ ```
SHA256SUMS ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ 11ce51ec0a25e3d1d95b025bca2f7d1647e47571eb7cc968acd5fc64d4b4fb65 data/v0.0.2/syfi_coding_trace.jsonl.gz
2
+ a7bab286bc640844560850965ccf47975cf66407154132abaab90f27ec9be744 artifacts/v0.0.2/syfi_coding_trace.duckdb
artifacts/v0.0.2/syfi_coding_trace.duckdb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a7bab286bc640844560850965ccf47975cf66407154132abaab90f27ec9be744
3
+ size 160182272
data/v0.0.2/syfi_coding_trace.jsonl.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:11ce51ec0a25e3d1d95b025bca2f7d1647e47571eb7cc968acd5fc64d4b4fb65
3
+ size 100939722