You can deploy this model and start earning money today!

multimodalart/
qwen3-vl-8b-instruct-lm-head

Name: qwen3-vl-8b-instruct-lm-head
Availability: InStock

Views

No views yet

Qwen3-VL-8B-Instruct — LM head

The standalone lm_head.weight ([151936, 4096], bf16) extracted from Qwen/Qwen3-VL-8B-Instruct.

Some diffusion pipelines ship the Qwen3-VL text encoder as the base Qwen3VLModel (no LM head, tie_word_embeddings=False), so it can only produce hidden states. Graft this single tensor onto that encoder to make it generative (e.g. for prompt upsampling) without re-downloading the full model.

python

from safetensors.torch import load_file
head = load_file("lm_head.safetensors")["lm_head.weight"]
# attach to a Qwen3VLForConditionalGeneration shell whose .model is the loaded encoder body

Derived from Qwen3-VL-8B-Instruct (Apache-2.0).

multimodalart/qwen3-vl-8b-instruct-lm-head

Qwen3-VL-8B-Instruct — LM head

multimodalart/
qwen3-vl-8b-instruct-lm-head