ti_2k

Screen-grounding training cell from the gui-agent-research pipeline (datagen v2 text negatives + image-neg synthetic screenshots), packed with every screenshot it references so it trains without the AgentNet image archives.

rows	steps	trajectories	text rows	image rows	ref rows	images	size
8850	6453	1742	4425	4425	0	18572	11.9 GB

Systems: {'Windows': 1907, 'Darwin': 1776, 'Ubuntu': 5167}. Negative types: {'N-stale': 2188, 'I-D': 4003, 'N-linger': 2166, 'N-anchor': 65, 'I-D0': 422, 'N-act': 6}. State-line format: list.

Which cell to train on (read this first)

This dataset holds two renderings of the same 8,850 rows (same keys, order, steps, negatives and images). Each belongs to exactly one base model. Never mix them, and never reuse a reference (ref) file across models.

base model	train on	reference log-probs	NOT this
Qwen/Qwen3.5-9B	cells/ti_2k.jsonl	not shipped: compute with Qwen3.5-9B on ti_2k.jsonl	ti_2k_venus.jsonl
inclusionAI/UI-Venus-2-9B	cells/ti_2k_venus.jsonl	not shipped: compute with UI-Venus-2-9B on ti_2k_venus.jsonl	ti_2k.jsonl

• Feeding ti_2k.jsonl to UI-Venus-2, or ti_2k_venus.jsonl to Qwen3.5, trains the model in a format it is not evaluated in. Don't.
• A ref file computed under one model is silently wrong for the other. For Venus compute the ref under UI-Venus-2-9B (training-handoff: scripts/ref.sh --model venus; research repo: datagen/scripts/ref.py --backend venus).
• ti_2k_venus.jsonl was added on 2026-11-07 (builder commit 0e1da0e, the same build as ti_1k_venus.jsonl / ti_5k_venus.jsonl); cells/ti_2k_venus.stats.json has its sha256, builder commit and format pins. ti_2k.jsonl and the images are unchanged.

Files

• cells/ti_2k.jsonl -- Qwen3.5-9B only. One row per (step, negative): key, tid, t, system, instruction, images (the step's screenshot window, paths relative to this dataset), messages (chat history up to the step), pos / neg (the two assistant chains; a text negative differs in the <current_state> line, an image negative has the same chain and images_neg = the window with its last screenshot replaced by the synthetic o^-), type, pivot, fmt.
• cells/ti_2k.ref.jsonl (only when ref rows > 0; otherwise run ref.py first, see below) -- the same rows with ref_pos / ref_neg / n_pos / n_neg / prompt_tok: sequence log-probs of the stock policy Qwen/Qwen3.5-9B (its chat template, the qwen_vendor message format and image preprocessing) used as the DPO / contrastive reference. These numbers are valid only when that exact model is the reference. Training another base model: keep cells/ti_2k.jsonl and the images, recompute the ref with datagen/scripts/ref.py --chains cells/ti_2k.jsonl --out <ref> under that model, then train. (For UI-Venus-2 do not use ti_2k.jsonl at all: train on cells/ti_2k_venus.jsonl.)
• cells/ti_2k_venus.jsonl -- UI-Venus-2-9B only. The same rows in the official UI-Venus-2 Computer format (github inclusionAI/UI-Venus @ 6d5f8ba, models/computer/computer_example.py, n_img 2, reasoning on): messages = system prompt with the task, each prior step as user ("" or "History Screenshot:" + image) + assistant <think>{state block + action prose}</think>\n<action>{Venus action}</action>, then "Current Screenshot:" + image; pos / neg chain = the continuation after the generation prompt <think>\n, i.e. {thought}\n</think>\n\n<action>...</action>. Actions map 1:1 from the AgentNet ground-truth code (Click(box=(x, y)), Type(content='...\n'), Drag, Swipe, Hotkey, ...). The thought carries the same state claims as ti_2k.jsonl, so the contrast (which claim is wrong) is identical. Same images / images_neg paths.
• cells/ti_2k_venus.stats.json -- counts, sha256, model / vendor / builder pins of the Venus cell.
• images/ -- AgentNet frames (<uuid>.png) and synthetic negatives (<tid>_<t>_<...>.png); packed as images/part-NNNNN.tar shards (fetch unpacks them in place).
• stats.json -- counts and the sha256 of the source cell.

Train

python research/screen-grounding/datagen/scripts/hf_pack.py fetch --repo <user>/ti_2k --out <dir>
bash research/screen-grounding/contrastive-dpo/scripts/train_cell.sh --exp <dir> --cell ti_2k --name <run> --loss pairsum

This command is the Qwen3.5-9B path. For UI-Venus-2-9B use the training-handoff repo's venus branch: prepare_data.py --repo yanghs1018/ti_2k, then ref.sh/train.sh --model venus (reads cells/ti_2k_venus.jsonl, computes the Venus ref itself). fetch rewrites the image paths to absolute ones under <dir>/images; the trainer reads cells/ti_2k.ref.jsonl. To get the plain state-line format: refmt.py --cell cells/ti_2k.jsonl --out <x> --fmt plain (list -> plain needs nothing else; plain -> list needs the annotation steps), then recompute the ref with datagen/scripts/ref.py.

Source: AgentNet (xlangai/AgentNet) trajectories, annotated with claude-sonnet-5-5 (v7 whole x diff); image negatives synthesised by image-neg (OmniParser-v2 + EasyOCR elements, retrieval bank, render_neg). Packed 2026-11-06T17:31:07.