gnitoahc/ceed-b1

CEED B1 — gemma-4-e4b-it, supervised fine-tuning, no teacher

A LoRA fine-tune of google/gemma-4-e4b-it trained with cross-entropy on gold answers only, with no teacher -- the control that says how much of a distilled Group's gain is distillation rather than fine-tuning.

The adapter has been folded into the base weights, so this is a standalone checkpoint: load it exactly like the base model, with no PEFT and no CEED code.

This is Group B1 of the CEED study (Causal Expert–Evidence Distillation), a research artifact published for reproducibility. It is not a product.

Usage

from transformers import AutoModelForImageTextToText, AutoProcessor

model = AutoModelForImageTextToText.from_pretrained("gnitoahc/ceed-b1", dtype="float16")
processor = AutoProcessor.from_pretrained("gnitoahc/ceed-b1")

The model was trained and scored with a short-answer instruction in the prompt. Without it an instruction-tuned model answers "The total written in the image is **28**." against gold "28" and scores zero on every metric here.

Training

Corpus	chartqa 2,500, docvqa 5,349, gqa 10,000 (17,849 examples, 80/10/10 split by example id)
Passes over the training split	2.69
Adapter	LoRA rank 4
Final cross-entropy	0.5048
Final KD term	0.0000
Seed	0
Run identity	f973d6eeb743e5b5b9b4d7512da2b37665a519c980afa2fb88f3abc54c5a1706

Evaluation

Dataset	Metric	Score	n
docvqa	ANLS	0.8798	565
gqa	exact match	0.6959	1016
chartqa	relaxed accuracy	0.7871	249

Scored by CEED's own harness (harness_version: ceed-direct-1) with greedy decoding, on CEED's own 10% validation split.

These numbers are not comparable to published DocVQA / GQA / ChartQA leaderboard results. Different splits, different prompt, different decoding. They are meaningful only against the other CEED Groups, which were scored identically.

Limitations

• This is a LoRA result. Merging folds the adapter into the weights; it does not turn a rank-4 adapter into a full fine-tune. CEED's own ADR-0005 bars LoRA numbers from the study's headline table, because a null result under a small adapter cannot be attributed between "the signal does not transfer" and "the adapter lacked the capacity to hold it". Read any comparison involving this checkpoint with that in mind.
• This Group has no teacher. B1 is the study's control: it isolates how much of a distilled Group's gain is distillation rather than plain fine-tuning. It is not itself a distillation result.
• Trained on document, natural-image and chart VQA in English only. Behaviour outside that is untested.
• Inherits the base model's limitations and the Gemma licence.

ceed_provenance.json beside the weights carries the source run's identity, parameter-efficiency mode, and metrics.