Skip to content
Computer vision

Qwen3-VL-8B InteriorGS

VLM camera navigation in 3D scenes

2 min readupdated

A fine-tune of the Qwen3-VL-8B-Instruct multimodal model (8.77B parameters) by yinbq, trained for active spatial navigation inside indoor scenes from the InteriorGS dataset. Given the current camera feed, its coordinates, and a spatial prompt, the model outputs a sequence of movement commands to reach the target viewpoint. Unlike the base model, this checkpoint was fine-tuned with reinforcement learning — RL (rewarding the model for valid navigation steps) via PPO — giving it a grasp of object distances, occlusions, and frame composition. Built for researchers working on embodied AI, 3D navigation, and automated viewpoint selection in virtual interiors.

Key features

  • Handles complex spatial tasks: object distance, keeping targets in frame, occlusion, projective relationships, screen occupancy ratio, and equidistance.
  • Runs in a multi-turn conversational loop (up to 12 turns), receiving a fresh camera view and pose at each step while preserving full conversation history.
  • Outputs discrete control commands inside a tag (supporting multiple actions per turn): move_forward, move_backward, move_left, move_right, turn_left, turn_right.
  • Weights converted from fp32 to bf16 with the original Qwen3-VL tokenizer, processor, and chat template left intact.

What you can do

  • Hook the model up to the InteriorGS indoor simulator to automatically search for camera angles matching a target object layout.
  • Deploy it as a baseline navigation agent in 3D interiors that require step-by-step camera movement driven by visual feedback.
  • Benchmark spatial planning gains after SFT and PPO against the stock Qwen3-VL-8B-Instruct model.

Example

import torch
from transformers import AutoProcessor, Qwen3VLForConditionalGeneration

repo = "yinbq/qwen3vl8b-interiorgs-reward-rw_s2tcsw3_100-bf16"
processor = AutoProcessor.from_pretrained(repo)
model = Qwen3VLForConditionalGeneration.from_pretrained(repo, dtype=torch.bfloat16, device_map="auto")
Read moreDetails
Training: SFT on trajectories generated by Qwen3-VL-8B-Instruct, followed by RL (PPO with step-level scoring, seed 20261021, step 100 checkpoint).
Metrics (InteriorGS v10, greedy, success rate): validation (413 tasks) — 53.75%, test (292 tasks) — 47.60%, overall (705 tasks) — 51.21%.
Inference: chat template without reasoning mode (no_think), temperature=0, top_p=1; unparseable outputs count as a failed step.
Compatibility: requires a transformers release with Qwen3-VL support (tested on 4.57).
Discuss Telegram 0 15

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 16 GB · ready files18 GB
  • FP8 / INT8recommendedweights 8.2 GB9.5 GB
  • 4-bitweights 4.6 GB5.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 3080 · BF16 — from RTX 4090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # yinbq/qwen3vl8b-interiorgs-reward-rw_s2tcsw3_100-bf16
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download yinbq/qwen3vl8b-interiorgs-reward-rw_s2tcsw3_100-bf16 --local-dir ~/models/qwen3vl8b-interiorgs-reward-rw_s2tcsw3_100-bf16
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Qwen3-VL-8B-Instruct (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 4All on Hugging Face
Merges 1All on Hugging Face

About the model

yinbq/qwen3vl8b-interiorgs-reward-rw_s2tcsw3_100-bf16
Author
yinbq
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

150

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments