CausalLens (Qwen3-VL-8B)
Community Qwen3-VL-8B research checkpoint pairing a language-model LoRA with a gated residual Q-Former on image tokens and an MM-Vet ablation runner.
What happened lately
CausalLens (Qwen3-VL-8B) timeline
CausalLens is a community research release from litcoderr, based on Qwen3-VL-8B-Instruct. The model card describes a rank-16 language-model LoRA and a causal residual Q-Former that modifies visual tokens using a fixed confounder dictionary. The download includes a supplied fine-tuned base checkpoint, adapters and processor files.
The companion repository provides Python 3.11 environment setup and MM-Vet inference modes for the base, LoRA-only and complete model. Its hardware and timing statements are author-reported. The reviewed repository does not provide a benchmark score table or an explicit project license. Publicly downloadable code and weights should not be treated as an established permissive license grant. AITrending has not run the model or established that it improves causal reasoning.