DSSD Trained early exit head to be used with Dynamic Self-Speculative Decoding valcore/DSSD-Qwen3-0.6B Updated Jan 8 • 5 valcore/DSSD-Llama3-8B Updated Jan 8 • 8
DSSD Trained early exit head to be used with Dynamic Self-Speculative Decoding valcore/DSSD-Qwen3-0.6B Updated Jan 8 • 5 valcore/DSSD-Llama3-8B Updated Jan 8 • 8