[Speculative Decoding] EAGLE Implementation with Top-1 proposer (#6830)

2024-08-22 15:12:24 +05:30
parent b3856bef7d
commit a3fce56b88
17 changed files with 854 additions and 83 deletions
--- a/vllm/model_executor/models/init.py
+++ b/vllm/model_executor/models/init.py
@@ -60,6 +60,7 @@ _GENERATION_MODELS = {
    "XverseForCausalLM": ("xverse", "XverseForCausalLM"),
    "Phi3SmallForCausalLM": ("phi3_small", "Phi3SmallForCausalLM"),
    "MedusaModel": ("medusa", "Medusa"),
+    "EAGLEModel": ("eagle", "EAGLE"),
    "MLPSpeculatorPreTrainedModel": ("mlp_speculator", "MLPSpeculator"),
    "JambaForCausalLM": ("jamba", "JambaForCausalLM"),
 }