---
license: gemma
base_model: google/gemma-4-e4b-it
library_name: transformers
pipeline_tag: image-text-to-text
tags:
  - vision-language
  - visual-question-answering
  - knowledge-distillation
  - lora
  - merged
  - research
datasets:
  - lmms-lab/DocVQA
  - lmms-lab/GQA
  - lmms-lab/ChartQA
language:
  - en
---

# CEED B4 — gemma-4-e4b-it, distilled, with visual-advantage reweighting

A LoRA fine-tune of [`google/gemma-4-e4b-it`](https://huggingface.co/google/gemma-4-e4b-it)
trained with B2's objective plus **VA-OPD's visual-advantage reweighting** (arXiv:2605.21924).
The teacher is the sparse mixture-of-experts [`google/gemma-4-26b-a4b-it`](https://huggingface.co/google/gemma-4-26b-a4b-it).

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 B4** of the CEED study (Causal Expert–Evidence Distillation),
a research artifact published for reproducibility. It is not a product.

## Usage

```python
from transformers import AutoModelForImageTextToText, AutoProcessor

model = AutoModelForImageTextToText.from_pretrained("gnitoahc/ceed-b4", dtype="float16")
processor = AutoProcessor.from_pretrained("gnitoahc/ceed-b4")
```

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 | 1.0979 |
| Final KD term | 2.1309 |
| Seed | 0 |
| Run identity | `6a4544ff1af675be8b364825a3f3f5c7bd825801471267ebd5f4956ff3febaa8` |

## Evaluation

| Dataset | Metric | Score | n |
| --- | --- | --- | --- |
| docvqa | ANLS | 0.8538 | 565 |
| gqa | exact match | 0.6102 | 1016 |
| chartqa | relaxed accuracy | 0.6185 | 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.
- **The distillation gain is not established.** The no-teacher control (CEED B1), trained identically but with `kd_weight: 0`, scored **above this checkpoint on every dataset** (docvqa 0.8538 vs 0.8798; gqa 0.6102 vs 0.6959; chartqa 0.6185 vs 0.7871). Whatever this checkpoint's objective contributes, it is not visible as an advantage over supervised fine-tuning here.
- 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.