webbrain-one/DeepSeek-V4-Flash-0731-Vision-BF16

DeepSeek V4 Flash 0731 Vision (BF16 source overlay)

DeepSeek V4 Flash 0731 with sight. This development checkpoint connects DeepSeek's reasoning and agentic backbone to the MoonViT vision encoder from Kimi-K2.6 through WebBrain's trained, routing-aware PatchMerger projector.

This is a complete copy of DeepSeek's DeepSeek-V4-Flash-0731 text checkpoint with WebBrain's frozen MoonViT tower, trained PatchMerger projector, routing bridge, and custom SGLang integration added in the same repository.

The BF16 suffix describes the added vision tower and projector, not every text tensor. The text checkpoint is the official upstream release, whose config declares torch_dtype: bfloat16 and FP8 quantization and whose stored tensors use mixed formats. No text tensor was converted or relabeled.

Why vision at WebBrain

At WebBrain, we build browser agents that need to understand the visual state of the web—not just extracted text. Screenshots, charts, dashboards, rich editors, and the location and appearance of controls are part of real browser work, so vision is a practical product requirement.

In our American–Chinese open-model frontier benchmark, DeepSeek V4 Flash stood out as a strong and cost-efficient model, while the upstream checkpoint remains text-only. This project attaches a MoonViT vision bridge while keeping the language backbone and vision tower frozen.

Contents

Component	Provenance	Status
Text checkpoint	deepseek-ai/DeepSeek-V4-Flash-0731 at 9e165c30e2704aec5d9d593cce3eebd58bbef1cb	48 shards; copied server-side without tensor changes
Vision tower	MoonViT-3d extracted from moonshotai/Kimi-K2.6 at 7eb5002f6aadc958aed6a9177b7ed26bb94011bb	329 BF16 tensors; frozen
Projector	WebBrain PatchMerger, trained with the text backbone and tower frozen	6 BF16 tensors; 40,119,040 parameters
Serving	External DeepSeek/MoonViT SGLang model and processor	Experimental; source patch required

The added components are byte-identical to the published WebBrain vision artifacts:

File	Size	SHA-256
vision_tower.safetensors	833,765,768 bytes	1382c41f1a4afc91791ade630e2b1e1cef68cc5a1e09668a45970a5d5e1b8f15
mm_projector.safetensors	80,238,568 bytes	7024d9d5c9714c7abbc09abda015f083b7d7b107745eb78879f019bf4721577a

The initial mirror commit, e181152a5285f5bcaad16f032267b192fd7cf99f, matched all 74 upstream paths, sizes, Git blobs, and LFS SHA-256 values (166,898,660,330 bytes). The final package intentionally replaces the model card and augments config.json and model.safetensors.index.json so the vision components can be discovered. The 48 text shards remain unchanged.

See VISION_ADAPTER_MANIFEST.json for machine-readable provenance.

Build status

• Pin and exactly mirror the official 0731 text checkpoint.
• Preserve all 48 text shards and the 0731 DSpark/MTP configuration.
• Add and fingerprint-verify the frozen BF16 MoonViT tower.
• Add and fingerprint-verify the trained BF16 PatchMerger projector.
• Add the routing palette, external processor/model package, source patch, launch wrapper, and regression tests.
• Augment the source config and index without changing upstream text tensor mappings or quantization metadata.
• Pass a full-model GPU loader/startup gate and live image-generation smoke test with the established deployment path.
• Run fresh text-parity and live image-generation smoke tests on the target hardware and SGLang image.

Integration details

The projector performs:

LayerNorm -> 2x2 patch merge -> Linear(4608, 4608) -> GELU -> Linear(4608, 4096)

Text token routing IDs are preserved. Image positions receive deterministic routing IDs from the checked-in 64-ID palette. The integration currently supports one literal <image> marker and targets at most 512 merged image tokens.

This is not a stock Transformers or stock SGLang multimodal architecture. Use the checked-in external model/processor package and the SGLang revision pinned in docs/SGLANG_DEPLOYMENT.md.

Validation status

• Exact upstream mirror metadata was verified before the vision overlay.
• Both added safetensors files were verified by LFS SHA-256 after publication.
• The source config's architecture, 0731 DSpark fields, and FP8 quantization metadata are preserved; only vision metadata was added.
• The index still references the same 48 text shards and additionally maps the 335 vision/projector tensor names.
• A full GPU loader/startup and image-generation smoke test has not yet been rerun for this assembled 0731 checkpoint. Earlier validation of the same vision adapter with another text package is not claimed as validation here.

Usage

Download the complete repository, then point the included wrapper at that local snapshot:

export DEEPSEEK_VISION_MODEL_PATH=/models/DeepSeek-V4-Flash-0731-Vision-BF16
export DEEPSEEK_VISION_PYTHONPATH="$DEEPSEEK_VISION_MODEL_PATH/sglang_ext"
export DEEPSEEK_VISION_TP=5
scripts/launch_sglang_moonvit.sh

Begin with SGLang's native /generate endpoint as documented in the deployment guide. OpenAI chat image parts, production concurrency, CUDA graphs, and broad quality behavior remain unvalidated.

Licenses

The DeepSeek text files retain their upstream terms in LICENSE and LICENSE_DEEPSEEK_V4_FLASH. The MoonViT-derived tower is covered by LICENSE_KIMI_K2.6. The WebBrain projector and integration source are provided under the terms included with this repository. Downstream users are responsible for satisfying all applicable licenses.

Method credit

The overall construction and model-card approach was inspired by Baseten's GLM-5.2-Vision-NVFP4: keep the text backbone and MoonViT tower frozen, train a compact PatchMerger projector between them, and publish provenance and hardware constraints explicitly. Credit to the Baseten team for demonstrating this practical recipe.

This project adds a DeepSeek-specific routing bridge so mixed text/image embeddings preserve hash-routed text behavior. It does not reuse Baseten model weights, benchmark results, or deployment artifacts.

Acknowledgements

Built on DeepSeek AI's DeepSeek V4 Flash 0731 and Moonshot AI's Kimi-K2.6, with the vision-attachment method inspired by Baseten's GLM-5.2-Vision-NVFP4. These teams were not involved in this experimental development checkpoint; please do not direct issues with this repository to them.

Want this model on your inference provider?

Ask your inference provider—such as OpenRouter or another managed inference service—to deploy this exact repository with its multimodal processor and serving plugin. Deploying only the upstream text model will not enable image input. Providers should complete the fresh 0731 GPU validation gates above before offering it as a production endpoint.

Experimental status, roadmap, and get involved

Experimental vision checkpoint. Repository assembly, provenance, hashes, and source preservation have been verified, but a fresh full GPU loader and image-generation smoke test has not been run for this 0731 package. This is not a production-readiness or quality claim. Fine-grained OCR, small-object and control identification, GUI grounding, hallucination calibration, concurrency, and broader hardware coverage still need evaluation. Do not use this model as the sole decision source for safety-critical automation. The roadmap includes larger and more diverse datasets, higher-resolution OCR/UI examples, and broader parameter-efficient tuning. Interested in contributing evaluation or training data, sponsoring compute, or working with us as a design partner? Tell us here.