biscet/DFlash-DeepSeek-TraceMix

DFlash DeepSeek TraceMix

DeepSeek teacher traces. Contexts for the next draft.

50,000 task contexts  ·  200 agent trajectories  ·  294 next-action steps  ·  RU / EN

Explore the data · Manifest · Privacy report

---

Built for DFlash distillation experiments. A bilingual mix of task prompts and DeepSeek-generated agent traces for preparing draft-model training and studying tool use.

The main corpus contains 50,000 prompts and contexts, not DeepSeek completions. The separate agent supplements contain actual reasoning, historical tool calls, real observations and final responses distilled from the DeepSeek teacher models listed below through Qwen Code.

Teacher models

We distilled the agent traces from DeepSeek Flash v4.1 (1017) and DeepSeek Pro V4 (0813).

Teacher model	Recorded API alias	Trajectories
DeepSeek Flash v4.1 (1017)	deepseek-flash	161
DeepSeek Pro V4 (0813)	deepseek-v4-pro	39

Trace records retain the original API aliases in their model and provenance fields.

Repository branches

Branch	Contents
main	Latest complete release: task contexts, DeepSeek agent data, capture exports, manifests and reports.
v1.0.0	Archived v1.0.0 release snapshot.

The complete dataset is available from the default main branch and in the Dataset Viewer. Version branches preserve release snapshots; pass revision="v1.0.0" for the archived release. Agent records are train-only supplements, separate from the main held-out evaluation splits.

The mix

Configuration	Focus	Train	Validation	Test
tasks	Coding, text processing, knowledge and reasoning contexts	45,000	2,500	2,500
agent_steps	One next-action step with tool history and observations	294	—	—
agent_trajectories	Complete DeepSeek sessions with reasoning and tool use	200	—	—

The 294 steps are derived from the 200 trajectories. They are two views of the same agent sessions, not independent extra sessions.

• Main languages: 24,982 Russian and 25,018 English tasks; no machine translations.
• Task mix: 60.4% text/reasoning tasks and 39.6% software development.
• Source mix: 29,598 literary-text tasks, 19,082 anonymized project-code tasks, 1,262 upstream prompt tasks and 58 self-created tasks.
• Splits: 11,927 semantic families, with each family confined to one split.
• Agent traces: 604 reasoning blocks and 611 tool calls across 200 teacher sessions, kept separate from the 50K task contexts.

Files

Path	Contents
data/train.jsonl, data/validation.jsonl, data/test.jsonl	Task IDs, messages, source/provenance, rendered prompts and exact token lengths.
agents/agent_trajectories.jsonl	Sanitized complete DeepSeek trajectories.
agents/agent_steps.jsonl	Sanitized next-action records compatible with the task schema.
capture/compact.jsonl	Whittle capture inputs, at most 1,536 prompt tokens.
capture/extended.jsonl	Longer Whittle capture inputs, bounded by 8,192 tokens.
quality_report.json	Main-corpus schema, split, duplicate, token-length and decontamination checks.
privacy_report.json	Publication cleanup scope and automated privacy checks.
manifest.json	Release counts, file sizes and SHA-256 checksums.
LICENSE, NOTICE	Apache-2.0 release license and retained source attribution.

Load the release

from datasets import load_dataset

tasks = load_dataset(
    "biscet/DFlash-DeepSeek-TraceMix", "tasks"
)
agent_steps = load_dataset(
    "biscet/DFlash-DeepSeek-TraceMix", "agent_steps"
)

Before your training run

• Target format: task prompts use the chat template for Whittle-Qwen-3.8-35B-A3B-Q5_K_M, with enable_thinking=false. They contain no assistant answers for ordinary tasks. Agent records can contain historical assistant tool calls and actual tool observations.
• DFlash capture: this release does not contain target-model hidden features, logits or captured teacher token IDs. Generate the target model's responses and features through your capture pipeline before draft-model training.
• Teacher scope: DeepSeek provenance applies to the agent configurations. The 50K task-context configuration is not a corpus of 50K DeepSeek completions.
• Lengths: main-task prompts range from 52 to 5,030 tokens; agent steps range from 1,128 to 8,176 tokens. Run compact and extended capture separately.
• Quality: the saved main-corpus checks pass, with no detected exact or near duplicates across families and no remaining cross-split 13-gram contamination after boilerplate filtering. Agent outputs are model-generated material, not verified reference answers or a held-out benchmark.

Privacy and publication processing

Personal paths, configured private identifiers, internal project names, email addresses, private IPs, phone numbers, authorization values and known token/key patterns are redacted across JSON keys and values, including reasoning and tool arguments. Session identifiers are removed from normalized agent records.

Changed task prompts and token lengths are recalculated. The publication audit checked 100,494 JSONL records, including the capture copies, and found no remaining matches to the configured sensitive-data patterns. This is an automated pattern audit, not a guarantee that every possible identifying detail was removed.

Raw logs, local configuration, checkpoints, reference solutions and rejected candidates are excluded. See the privacy report.

Attribution & terms

Release license: Apache-2.0. Original contributions and release packaging are provided under this license. Third-party source terms remain applicable; the license does not replace source notices recorded in each task's source and provenance fields.

Sources include anonymized project-code excerpts, public-domain literary passages from Project Gutenberg and Wikisource, self-created tasks and prompts from biscet/DFlash-TraceMix-Lite v1.0.0. Tulu-related sources retain ODC-BY-1.0 and subset-specific terms. Literary source records retain book titles, authors, URLs and license notices. Agent records retain the DeepSeek model and Qwen Code generator provenance.

See NOTICE for attribution and source terms.