Steered text and image generations from UniAR (ShareLab-SII/UniAR-RL), one of the unified vision-language models in the unified-vlm-steering project. Steering adds alpha vhat (the per-layer unit difference-of-means vector) to the residual stream at every layer of a layer config. Qwen3-VL backbone, 36 decoder layers (0-indexed). Images are BSQ tokens rendered by an SD3 decoder (16 decoding steps, 544 px). Every quadrant covers 7 concepts x 7 layer configs x 11 alphas x 20 prompts. Layer configs (0-indexed): L9 9, L18 18, L27 27, early 7-10, mid 16-19, late 25-28, all 0-35. - Quadrant names are 2: txt2txt and img2txt steer text generation, img2img and txt2img steer image generation. - alpha =…
Independent publisher
Alain Zeng
saintsauce
Models in Library0
Datasets in Library3
Models on Hugging Face28
Followers—
Compare the published evidence
This profile connects the 3 model and dataset records shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.
Datasets
Steered text and image generations from Emu3.5 (BAAI), one of the unified vision-language models in the unified-vlm-steering project. Steering adds alpha vhat (the per-layer unit difference-of-means vector) to the residual stream at every layer of a layer config. 34B dense autoregressive model, 64 decoder layers (0-indexed). Images are 32x32 IBQ tokens (512 px), generated on BAAI's patched vLLM engine with request-id-keyed CFG. The mid window in both sweeps is mid8 = layers 28-35 (it replaces the 4-layer mid4 window). vectors/steertablevllm.pt holds the per-layer unit vectors and layer masks the image engine used. Every quadrant covers 7 concepts x 7 layer configs x 11 alphas x 20 prompts.…
Steered text and image generations from Liquid (FoundationVision LiquidV17B), one of the unified vision-language models in the unified-vlm-steering project. Steering adds alpha vhat (the per-layer unit difference-of-means vector) to the residual stream at every layer of a layer config. Gemma-7B backbone, 28 decoder layers (0-indexed). Images are VQGAN codes (512 px), CFG 7.0, top-k 4096, top-p 0.96, temperature 0.99. Every quadrant covers 7 concepts x 7 layer configs x 11 alphas x 20 prompts. Layer configs (0-indexed): L7 7, L14 14, L21 21, early 5-8, mid 12-15, late 19-22, all 0-27. - Quadrant names are 2: txt2txt and img2txt steer text generation, img2img and txt2img steer image…