LiLiCorr
A draft model supplied by NVIDIA alongside Nemotron 3.5 Bolt for speculative decoding. It is implemented through DFlash and predicts seven speculative tokens per step by default.
Timeline (last 90 days)
The DGX Spark (GB10) has 128 GB of unified memory and can run Nemotron 3.5 Super VL using the official NVFP4 + vLLM + LiLiCorr drafter.
On the DGX Spark, the official Quick Start uses vllm/vllm-openai:v0.32.0, with key parameters including --max-num-seqs 1, --gpu-memory-utilization 0.75, speculative decoding with the LiLiCorr drafter configured with 7 draft tokens, and --reasoning-parser nemotron_v3.
NVIDIA released the Nemotron-3.5-Super-VL-120B-A12B model family, covering a base version as well as multiple quantization formats including NVFP4 and BF16, and introduced the LiLiCorr speculative decoding scheme.
Nemotron 3.5 Bolt comes with a LiLiCorr drafter for speculative decoding, implemented via DFlash, with 7 speculative tokens by default in the vLLM configuration.
All Facts (4)
The DGX Spark (GB10) has 128 GB of unified memory and can run Nemotron 3.5 Super VL using the official NVFP4 + vLLM + LiLiCorr drafter.
50%UnverifiedOn the DGX Spark, the official Quick Start uses vllm/vllm-openai:v0.32.0, with key parameters including --max-num-seqs 1, --gpu-memory-utilization 0.75, speculative decoding with the LiLiCorr drafter configured with 7 draft tokens, and --reasoning-parser nemotron_v3.
50%UnverifiedNemotron 3.5 Bolt comes with a LiLiCorr drafter for speculative decoding, implemented via DFlash, with 7 speculative tokens by default in the vLLM configuration.
50%UnverifiedNVIDIA released the Nemotron-3.5-Super-VL-120B-A12B model family, covering a base version as well as multiple quantization formats including NVFP4 and BF16, and introduced the LiLiCorr speculative decoding scheme.
50%