AIcell/guava-v13b-qwen3.5-4b-no-counterfactual

guava-v13b-qwen3.5-4b-no-counterfactual

A Qwen3.5-VL 4B model fine-tuned on the Guava v13b robot-manipulation dataset, trained without counterfactual (branch) episodes — the ablation counterpart to AIcell/guava-v13b-qwen3.5-4b. This is the full-weight inference export of checkpoint 147.

Training data: AIcell/guava-v13b (private). The full dataset holds 2,268 episodes — 1,554 main and 714 branch — across 16 tabletop tasks.

Status — read before using

Published as exported, with no evaluation run here. Specifically:

• —No evaluation results. No success rates, no held-out scores, and no measured comparison against the counterfactual-trained model. Nothing in this repo establishes what the ablation actually costs or gains.
• —The "no counterfactual" label is the uploader's designation. The export itself carried no run metadata identifying its training split — it is not independently verifiable from these files.
• —One mid-training checkpoint. Step 147 — not selected against a validation metric, because the dataset ships no held-out split. It is also a different step count than the 213 of the comparison model, so the two are not matched-step.
• —Inherited data caveat. The training data carries an unresolved physical-clearance issue on 05-24__push_cereal__trial_0308.
• —Untested load path. Requires transformers 5.8.1 (model_type: qwen3_5).

Model

Config, tokenizer, and chat template are byte-identical to the comparison model; only the two weight shards differ.

Usage

from transformers import AutoModelForImageTextToText, AutoProcessor

model = AutoModelForImageTextToText.from_pretrained(
    "AIcell/guava-v13b-qwen3.5-4b-no-counterfactual", dtype="bfloat16", device_map="auto"
)
processor = AutoProcessor.from_pretrained("AIcell/guava-v13b-qwen3.5-4b-no-counterfactual")

Episodes contain up to 30 images, and coordinates are already table-aligned (tabletop is z = 0) — do not apply table-height normalization a second time.

Tasks

apple_juice_order, bread_near_lemon, can_in_bin, close_drawer, cube_stack, cube_under_cup, hotdog_near_donut, milk_near_cup, open_drawer, pick_up_orange, push_basket, push_cereal, red_objects_in_basket, remove_cube_from_tray, shell_game, tomato_in_bowl