DSSD Trained early exit head to be used with Dynamic Self-Speculative Decoding FlorianVal/DSSD-Qwen3-0.6B Updated Jan 8 • 31 FlorianVal/DSSD-Llama3-8B Updated Jan 8 • 13
DSSD Trained early exit head to be used with Dynamic Self-Speculative Decoding FlorianVal/DSSD-Qwen3-0.6B Updated Jan 8 • 31 FlorianVal/DSSD-Llama3-8B Updated Jan 8 • 13