LBS-27B vs Qwopus3.8-27B-Flash-V2

LBS-27B

A 27B Qwen3.8 fine-tune for GUI grounding that zooms into screenshots over a few turns before naming a point, and inclusionAI says it also helps with fine detail in large images. It needs about 96 GB at BF16 and a client that runs its crop actions. Every result so far is inclusionAI's own, with no safety results or injection guidance.

Qwopus3.8-27B-Flash-V2

A Qwen3.8-27B cousin tuned to think less per turn in long agent loops, with an MTP head for speculative decoding in llama.cpp. It fits 24 GB with Jackrong's 4-bit GGUF. The risks are one person's experimental release with no safety work, training traces taken from Claude and GPT, and a tool format that differs from Qwen's.

LBS-27B

1. Serve with vllm serve inclusionAI/LBS-27B --tensor-parallel-size 4 --max-model-len 65536, which is what the card's wrapper script runs; keep the context at 32,768 or more for multi-turn images
2. Build the system prompt with build_system_prompt from the LBS branch's zoomin_grounding_protocol.py, send the screenshot after image_path: <name>, and don't pass tools; the model answers in plain text
3. Parse <action>crop_image(image_path=..., x1=..., y1=..., x2=..., y2=...)</action>, cut the crop yourself and return it with its mapping to the original image; allow at most two crops per turn and three per episode
4. Read the final answer as [x.xx,y.yy] on a 0 to 1000 scale of the original image, and use your own executor to click
5. Pass reasoning_effort as xhigh, medium or low only, since the template raises an error for high, and don't count on thinking_budget, which the template ignores
6. Don't turn on MTP speculative decoding; the MTP weights aren't in the repository though config.json names one MTP layer

Qwopus3.8-27B-Flash-V2

1. Quickest start is llama-server -hf Jackrong/Qwopus3.8-27B-Flash-V2-GGUF:Q4_K_M --jinja, about 17 GB, and add mmproj-F32.gguf for images
2. The card's tests used Q5_K_M with the bundled MTP head and --spec-type draft-mtp in llama.cpp
3. The template asks for tool calls as JSON inside tool_call tags, so use a Hermes-style parser (vLLM's hermes), not Qwen3.8's qwen3_coder
4. For coding the card recommends enable_thinking true, reasoning_effort xhigh and temperature 0.3, which you must pass yourself, since generation_config.json says 1.0
5. reasoning_effort takes xhigh (default), medium or low, and any other value makes the template raise an error
6. Earlier reasoning stays in the context unless preserve_thinking is false
7. Cap max_tokens, since a report says it loops more than V1