WebGPU LLM benchmark
Inspecting environment…
Environment
Model
Model
LiquidAI LFM2.5 1.2B Instruct
Qwen3.5 2B ONNX OPT (comparison)
dtype
q4f16
q4
vendor sampling
Load model
Run prompts
Verify palindrome
Long-prompt probe
Model weights are large and may be downloaded from Hugging Face. This benchmark never selects a CPU/wasm backend.
Fixed prompts
Long-prompt probe
Export
Copy JSON