Checkpoints for H-Spec: Parallel Speculative Decoding Without a Drafter-Side KV Cache (arXiv 2609.24197).
Weifan Jiang PRO
weifanjiang
AI & ML interests
Efficient LLM
Recent Activity
updated a collection about 15 hours ago
H-Spec updated a dataset 3 days ago
weifanjiang/spec-decode-profilings published a dataset 9 days ago
weifanjiang/spec-decode-profilingsOrganizations
None yet
models 12
weifanjiang/qwen3-4b.speculators.hspec
1B • Updated • 48
weifanjiang/qwen3-4b.speculators.dspark
1B • Updated • 45
weifanjiang/qwen3-4b.speculators.dflash
1B • Updated • 42
weifanjiang/qwen3-8b.speculators.hspec
2B • Updated • 41
weifanjiang/qwen3-8b.speculators.dspark
2B • Updated • 46
weifanjiang/qwen3-8b.speculators.dflash
2B • Updated • 39
weifanjiang/qwen3-8b.speculators.peagle
2B • Updated • 29
weifanjiang/llama3.1-8b-it.speculators.hspec
2B • Updated • 25
weifanjiang/llama3.1-8b-it.speculators.dspark
2B • Updated • 57
weifanjiang/llama3.1-8b-it.speculators.dflash
2B • Updated • 42