Releases·Model Release
SenseNova-Vision-7B-MoT fine-tune of ByteDance Bagel does detection, OCR, depth and masks in one model
SenseNova-Vision-7B-MoT is a fine-tune of ByteDance's Bagel that unifies object detection, OCR, depth estimation, and segmentation in one 14B-total, 7B-active model. Boxes and OCR output as text with coordinates; depth maps and masks are generated as images. Benchmarks are strong but not uniformly best-in-class, and local deployment is currently impractical.
5 engagement1 sourceSat, Sep 26, 2026, 04:55 PM
AI Overview
- SenseNova-Vision-7B-MoT is a fine-tune of ByteDance's Bagel with 14B total parameters, 7B active, emitting boxes and OCR as text coordinates and depth/masks as generated images.
- Its paper ranks it first on object detection and OCR (one tie) among compared models and beats Depth Anything V2 on every depth set, though it trails MoGe-2 on most.
- Dedicated segmentation models still win most mask tests.
- The README's only tested setup is an 80GB A800; no GGUF exists, llama.cpp cannot load Bagel, and weights are non-commercial despite Bagel's Apache 2.0 license.
Related
Releases·Model Release
SenseNova-Vision: 7B open model unifies vision tasks without task-specific heads
SenseNova-Vision is a 7B mixture-of-transformers model released under Apache 2.0 that treats all computer vision tasks as a single generation problem, handling segmentation, depth, detection, OCR, and 3D reconstruction without task-specific heads. It accepts natural language instructions and visual hints, outputting text or images, simplifying vision pipelines for practitioners.
SenseNova-Vision is a 7B mixture-of-transformers model released under Apache 2.0 that treats all computer vision tasks as a single generation problem, handling segmentation, depth, detection, OCR, and 3D reconstruction without task-specific heads. It accepts natural language instructions and visual hints, outputting text or images, simplifying vision pipelines for practitioners.
15 engagement1 sourceredditThu, Aug 13, 2026, 03:07 PM
Social·Community
Redditor runs 2x Radeon AI Pro R9700 in a GMKtec EVO-X2 via NVMe-to-PCIe adapters, doubling llama.cpp decode speed
A user modded a GMKtec EVO-X2 (Strix Halo) mini PC to run two Radeon AI Pro R9700 GPUs through its NVMe slots using PCIe adapters, connected via USB4 and limited to PCIe 4.0 x4 each. llama.cpp benchmarks with a Qwen 27B model show a 32% token-generation speedup over a single card, after fixing tensor parallelism issues in the Vulkan build and enabling RCCL in the ROCm build.
A user modded a GMKtec EVO-X2 (Strix Halo) mini PC to run two Radeon AI Pro R9700 GPUs through its NVMe slots using PCIe adapters, connected via USB4 and limited to PCIe 4.0 x4 each. llama.cpp benchmarks with a Qwen 27B model show a 32% token-generation speedup over a single card, after fixing tensor parallelism issues in the Vulkan build and enabling RCCL in the ROCm build.
23 engagement1 sourceredditMon, Sep 21, 2026, 12:57 AM
Releases·Tool Release
VisionBridge and LocalEyes: tools give text-only LLMs vision via separate models
Two open-source tools, VisionBridge and LocalEyes, enable text-only reasoning models (DeepSeek, Qwen, GLM, CodeLlama) to process images by querying a separate vision model. VisionBridge is an OpenAI-compatible proxy that uses tool calls (look, OCR, scan, crop, compare) and works with LM Studio, Ollama, vLLM, and others. LocalEyes integrates with Claude Code and uses a local Ollama vision model for screen capture, clipboard reading, and image analysis, all without cloud uploads. Both require no training or weight modifications.
Two open-source tools, VisionBridge and LocalEyes, enable text-only reasoning models (DeepSeek, Qwen, GLM, CodeLlama) to process images by querying a separate vision model. VisionBridge is an OpenAI-compatible proxy that uses tool calls (look, OCR, scan, crop, compare) and works with LM Studio, Ollama, vLLM, and others. LocalEyes integrates with Claude Code and uses a local Ollama vision model for screen capture, clipboard reading, and image analysis, all without cloud uploads. Both require no training or weight modifications.
Updated Tue, Jul 7, 2026, 07:13 PM — LocalEyes released on 2026-07-06, VisionBridge on 2026-07-07, both providing vision for text-only LLMs via separate visi
58 engagement2 sourcesgithubMon, Jul 6, 2026, 09:30 PM
Developments·Paper
Smol-VL-BLV: 500M on-device VLM for blind users via distillation and GRPO post-training
A research paper introduces Smol-VL-BLV, a compact 500M vision-language model built for blind and low-vision users. It combines teacher-student distillation with GRPO post-training targeting directional language, metric distances, and hazard detection, and runs fully offline on a mid-range Android phone at roughly 450 MB. The model, dataset, and code are publicly released, with large gains in spatial grounding and OCR over the baseline.
A research paper introduces Smol-VL-BLV, a compact 500M vision-language model built for blind and low-vision users. It combines teacher-student distillation with GRPO post-training targeting directional language, metric distances, and hazard detection, and runs fully offline on a mid-range Android phone at roughly 450 MB. The model, dataset, and code are publicly released, with large gains in spatial grounding and OCR over the baseline.
0 engagement1 sourcearxivWed, Sep 23, 2026, 08:10 PM