Remek/basal-1.0-4.5B
Text Generation • 5B • Updated • 1.16k • 5
basal-1.0
Calibrated typed-decision models for Polish (and English): one forward pass, probabilities over the allowed answers. Engine: github.com/rkinas/basal
Note Main model, bf16, with early-exit heads
Remek/basal-1.0-1.5B
Text Generation • 2B • Updated • 523Note Lite model: 2.5x throughput
Remek/basal-1.0-4.5B-FP8
Text Generation • 5B • Updated • 36 • 1Note ModelOpt FP8 for vLLM
Remek/basal-1.0-4.5B-NVFP4
Text Generation • 2B • Updated • 235Note ModelOpt NVFP4 for vLLM (Blackwell, throughput)
Remek/basal-1.0-1.5B-FP8
Text Generation • 2B • Updated • 22Note ModelOpt FP8 for vLLM
Remek/basal-1.0-1.5B-NVFP4
Text Generation • 0.8B • Updated • 263Note ModelOpt NVFP4 for vLLM (Blackwell)
jev-pl-benchmark
⚖1Explore benchmark scores for Polish and English decision models
Note Scores only (Polish, English, speed)