Back to leaderboardLlama 3.2 1BMeta · 1B parameters · browser LLM configuration and published measurements.Runtime and model configurationConfigured weight formats: 4-bit weights · FP16 compute; 4-bit weights · FP32 compute; Unquantized FP16 weights. Inference uses WebLLM and WebGPU in a dedicated browser worker.Publisher model card · Run benchmarkPublished benchmark resultsq4f16_1 · WebLLM 0.2.82Median generation speed163.6 tok/sMedian first-output latency26.3 msMedian response time0.42 sSampled peak Chrome footprint3.82 GBCompleted measured requests1Browser and deviceChrome 155.0.0.0 · M5 Pro · 48 GB · dedicated ChromeDownload JSON reportWarmup is excluded from performance medians. Tokenizers and output lengths differ across models. These measurements do not evaluate task quality. Measurement definitions
