---
license: apache-2.0
license_link: https://ai.google.dev/gemma/docs/gemma_4_license
base_model: google/diffusiongemma-26B-A4B-it
pipeline_tag: image-text-to-text
library_name: transformers
tags:
  - diffusion-language-model
  - structured-decision-making
  - jev
  - djev
datasets:
  - LocalLLaMA/typed-decisions
---

# DiffusionGemma 26B DJev v10 — step 160

This is the selected one-step DJev checkpoint from the v10 continuation of
`google/diffusiongemma-26B-A4B-it`. It is tuned for typed probabilistic
decisions through the Jev/System One request schema. The selected operating
point uses **no generated reasoning block** (`think=0`), four samples, and a
64-token answer canvas.

This is a research release. It returns calibrated decision distributions but
should not be used as the sole decision-maker in high-impact settings.

## Intended serving path

The evaluated path uses the structured DiffusionGemma runtime and the
`/v1/systemone` endpoint in the accompanying
[Turbo-dLLM repository](https://github.com/tarsur909/Turbo-dLLM/tree/feature/djev-turnkey).
The checkpoint is a full Hugging Face export, not an adapter.

Recommended DJev settings:

```json
{
  "think": 0,
  "samples": 4,
  "canvas_size": 64
}
```

The upstream `generation_config.json` remains in the export for general
DiffusionGemma generation. The DJev scores below were measured with the
structured one-step serving path, not generic `generate()` defaults.

## Evaluation

### JevBench public v1.3-format set

All 231 public requests completed without error. This public set was reported,
not used to select step 160; selection used held-out gates.

| Overall accuracy | Easy | Standard | Judge | Hard | Hard ECE | C13 | I13 estimate |
|---:|---:|---:|---:|---:|---:|---:|---:|
| 0.8831 | 1.000 | 0.986 | 0.882 | 0.745 | 0.044 | 82.9 | 83.2 |

In a paired run, this checkpoint exceeded `Mapika/decider-4b` v2 on the public
set (0.8831 versus 0.8355) and on all five internal held-out sets. See the
[comparison report](https://github.com/tarsur909/Turbo-dLLM/blob/feature/djev-turnkey/runs/decider-v2-comparison/README.md)
for the pinned revisions, confidence intervals, latency methodology, and raw
artifact hashes.

### LocalLLaMA/typed-decisions test

This result is **in-task/specialist**, not zero-shot: the model's training
lineage includes the dataset's disjoint train/development rows. The 400-case
test split was evaluation-only.

| Accuracy | Soft accuracy | KL | TV | Brier | ECE | Score MAE | Within 1 |
|---:|---:|---:|---:|---:|---:|---:|---:|
| 0.687 | 0.580 | 0.421 | 0.247 | 0.166 | 0.111 | 0.436 | 0.925 |

The exact dataset revision, test hash, scorer, per-type results, and prediction
artifact hashes are in the
[evaluation report](https://github.com/tarsur909/Turbo-dLLM/blob/feature/djev-turnkey/runs/typed-decisions-eval/README.md).

## Training and selection

- Base: `google/diffusiongemma-26B-A4B-it` at revision
  `f7f5b7f5fa82ffc52addd066915886d497f5517b`.
- Warm start: DJev v9 step 343.
- Update: BF16 all-text continuation with FSDP2 across eight H100s, a 4,096
  token training maximum, preservation distillation, and mixed teacher/gold
  targets.
- Optimizer: AdamW, peak learning rate `1e-6`, cosine decay, batch size one per
  rank with four gradient-accumulation steps.
- Candidate checkpoints: steps 160, 240, and 320. Step 160 was selected on
  held-out hard/generalization gates rather than the public score.
- No JevBench public evaluation row or expected answer was used as training
  data. The public-development and synthetic/replay construction is recorded
  in the repository's recipes and run scripts.

Reproduction entry points:

- `dllm_parallel/recipes/runs/diffusiongemma-26b-djev-distill-v10.yaml`
- `runs/data-v10/run_v10.sh`
- `runs/decider-v2-comparison/run_comparison.sh`
- `runs/typed-decisions-eval/evaluate_endpoint.py`

## Limitations

- The checkpoint is specialized for structured decisions; the fine-tune was
  not evaluated as a general chat or multimodal assistant.
- Typed Decisions Choice and ordinal Score remain weaker than the published
  TypeSafe Jev 1.13.0 reference, while Noul accuracy is higher.
- H100 latency measurements are serving-stack and hardware specific.
- The model inherits the capabilities, failure modes, and license obligations
  of the DiffusionGemma base model.