List your company

Code Generation on HumanEval (Speed/Throughput)

287.77Tokens/sPRISM
-8.494868.4201145.335222.2499Nov 16, 2024Jan 28, 2025Apr 12, 2025Jun 25, 2025Sep 7, 2025Nov 20, 2025Feb 2, 2026

Evaluation results

MethodLinks
PRISMModel=LLaMA-2, Temp=T = 02026.02287.77——5.26—
HASSModel=LLaMA-2, Temp=T = 02026.02271.59——5.12—
PRISMModel=LLaMA-3, Temp=T = 02026.02252.18——5.44—
EAGLE-2Model=LLaMA-2, Temp=T = 02026.02251.94——4.75—
PRISMModel=LLaMA-2, Temp=T = 12026.02250.55——4.88—
HASSModel=LLaMA-2, Temp=T = 12026.02233.54——4.74—
HASSModel=LLaMA-3, Temp=T = 02026.02230.42——5.16—
EAGLE-2Model=LLaMA-2, Temp=T = 12026.02218.69——4.48—
EAGLE-2Model=LLaMA-3, Temp=T = 02026.02198.19——4.57—
PRISMModel=LLaMA-3, Temp=T = 12026.02182.53——5.08—
StandardModel=LLaMA-3, Temp=T = 02026.02171.76——5.93—
HASSModel=LLaMA-3, Temp=T = 12026.02169.27——4.75—
EAGLE-2Model=LLaMA-3, Temp=T = 12026.02154.91——4.35—
StandardModel=LLaMA-3, Temp=T = 12026.02154.6——5.36—
StandardModel=LLaMA-2, Temp=T = 02026.02133.32——2.72—
StandardModel=LLaMA-2, Temp=T = 12026.02129.76——2.62—
VanillaModel=LLaMA-2, Temp=T = 02026.02103.57————
VanillaModel=LLaMA-2, Temp=T = 12026.02103.27————
Streaming-dLLMGen Length=256, Prompting Shot=0-shot2026.0193.30.4345.1——
VanillaModel=LLaMA-3, Temp=T = 02026.0291.51————
SAM-Decoding[E2]Model=Llama3-8B2024.1191.54.762.54——
VanillaModel=LLaMA-3, Temp=T = 12026.0291.02————
EAGLE-2Model=Llama3-8B2024.1185.584.742.37——
SAM-Decoding[E2]Model=Vicuna-13B2024.1178.965.123.54——
SAM-Decoding[T]Model=Llama3-8B2024.1178.042.772.16——
EAGLE-2Model=Vicuna-13B2024.1177.855.153.49——
Streaming-dLLMGen Length=512, Prompting Shot=0-shot2026.0175.80.4456.4——
Fast-dLLMGen Length=256, Prompting Shot=0-shot2026.0171.50.4333.9——
Token RecyclingModel=Llama3-8B2024.1171.492.931.99——
SAM-DecodingModel=Llama3-8B2024.1164.382.061.79——
Fast-dLLMGen Length=512, Prompting Shot=0-shot2026.01520.4454.4——
SAM-Decoding[T]Model=Vicuna-13B2024.1150.872.792.28——
SAGESetup=Llama3 8B-1B2026.0150.2—2.67—4.89
SAM-DecodingModel=Vicuna-13B2024.1148.922.422.2——
Token RecyclingModel=Vicuna-13B2024.1146.032.792.07——
Native-SDSetup=Llama3 8B-1B2026.0144.89—2.39—3.48
PLDModel=Llama3-8B2024.1142.391.31.18——
Streaming-dLLMGen Length=256, shots=0-shot2026.0134.10.395.3——
PLDModel=Vicuna-13B2024.1132.061.541.44——
Prefix-CacheGen Length=256, Prompting Shot=0-shot2026.0130.60.4271.7——
SAM-Decoding[E2]Model=Vicuna-33B2024.1129.563.613.14——
EAGLE-2Model=Vicuna-33B2024.1128.183.533——
Streaming-dLLMGen Length=512, shots=0-shot2026.0126.70.4029.2——
SAM-Decoding[T]Model=Vicuna-33B2024.1122.442.732.39——
Prefix-CacheGen Length=512, Prompting Shot=0-shot2026.0121.80.4571.8——
Token RecyclingModel=Vicuna-33B2024.1119.642.642.09——
Fast-dLLMGen Length=256, shots=0-shot2026.0119.10.3723——
SAM-DecodingModel=Vicuna-33B2024.1119.082.052.03——
dKV-CacheGen Length=256, Prompting Shot=0-shot2026.0118.90.3961——
VanillaSetup=Llama3 8B-1B2026.0118.8————
LLaDAGen Length=256, Prompting Shot=0-shot2026.0118.40.4151——
dKV-CacheGen Length=512, Prompting Shot=0-shot2026.0114.30.4511.2——
PLDModel=Vicuna-33B2024.1114.181.581.51——
Fast-dLLMGen Length=512, shots=0-shot2026.0113.60.3984.7——
LLaDAGen Length=512, Prompting Shot=0-shot2026.0111.90.4391——
Prefix-CacheGen Length=256, shots=0-shot2026.0110.90.3841.7——
dKV-CacheGen Length=256, shots=0-shot2026.016.60.4021——
LLaDA-1.5Gen Length=256, shots=0-shot2026.016.40.4391——
Prefix-CacheGen Length=512, shots=0-shot2026.014.80.3781.7——
dKV-CacheGen Length=512, shots=0-shot2026.013.30.4021.1——
LLaDA-1.5Gen Length=512, shots=0-shot2026.012.90.4051——
FullModel=LLaDA-8B, Inference Strategy=Full2026.03—30.5———
FullModel=Dream-7B, Inference Strategy=Full2026.03—54.9———
LSPModel=LLaDA-8B, Inference Strategy=LSP2026.03—29.31.22——
LSPModel=Dream-7B, Inference Strategy=LSP2026.03—55.51.46——