List your company

Code Generation on MBPP 3-shot (test)

64.1Speedup RatioStreaming-dLLM
-1.52415.51332.5549.587Sep 27, 2025Nov 6, 2025Dec 17, 2025Jan 27, 2026Mar 8, 2026Apr 18, 2026May 29, 2026

Evaluation results

MethodLinks
Streaming-dLLMGen Length=512, Evaluation Protocol=3-shot2026.0164.1——40.81.1——
d2CacheModel=LLaDA-Inst, Few-shot=3-shot, Gen. Len.=512, Decoding Strategy=Semi-AR parallel decoding2025.0948.3———4.64119.8140.2
d2CacheModel=Dream-Inst, Few-shot=3-shot, Gen. Len.=512, Decoding Strategy=Semi-AR parallel decoding2025.0939.7———5.11108.3952.8
Streaming-dLLMGen Length=256, Evaluation Protocol=3-shot2026.0132.7——420.9——
TSPDBackbone=LLaDA-8B-Instruct, KV Cache=true, Generation Length=2562026.0511.8——29.2—68.4—
Fast-dLLMGen Length=512, Evaluation Protocol=3-shot2026.019.3——13.87.6——
Fast-dLLMGen Length=256, Evaluation Protocol=3-shot2026.017.7——28.23.8——
Fast-dLLMBackbone=LLaDA-8B-Instruct, KV Cache=true, Generation Length=2562026.057.2——28.2—41.8—
Credit DecodingBackbone=LLaDA-8B-Instruct, KV Cache=true, Generation Length=2562026.057——29.3—40.6—
TSPDBackbone=LLaDA-8B-Instruct, KV Cache=false, Generation Length=2562026.055.6——29.5—32.5—
Learn2PDBackbone=LLaDA-8B-Instruct, KV Cache=false, Generation Length=2562026.055.1——29.2—29.6—
Prefix-CacheGen Length=256, Evaluation Protocol=3-shot2026.013.4——29.68.6——
Prefix-CacheGen Length=512, Evaluation Protocol=3-shot2026.013——13.423.6——
dKV-CacheGen Length=256, Evaluation Protocol=3-shot2026.011.7——26.217.6——
dKV-CacheGen Length=512, Evaluation Protocol=3-shot2026.011.6——14.643.3——
ProphetBackbone=LLaDA-8B-Instruct, KV Cache=false, Generation Length=2562026.051.4——29.2—8.1—
LLaDAGen Length=256, Evaluation Protocol=3-shot2026.011——29.429.4——
LLaDAGen Length=512, Evaluation Protocol=3-shot2026.011——14.870.5——
VanillaBackbone=LLaDA-8B-Instruct, KV Cache=true, Generation Length=2562026.051——29.4—5.8—
diffu-GRPO(d1)Sequence Length=128, Few-shot=32025.12—34.8—————
diffu-GRPO(d1)Sequence Length=256, Few-shot=32025.12—36.6—————
diffu-GRPO(d1)Sequence Length=512, Few-shot=32025.12—38—————
diffu-GRPO(d1)Few-shot=32025.12——36.5————
DualCacheBackbone=LLaDA-8B-Base, Speedup=12.6×2026.03—38.6———120.12—
ES-dLLMBackbone=LLaDA-8B-Base, Speedup=16.7×2026.03—38———158.81—
ESPOSequence Length=128, Few-shot=32025.12—47.4—————
ESPOSequence Length=256, Few-shot=32025.12—44.6—————
ESPOSequence Length=512, Few-shot=32025.12—44.2—————
ESPOFew-shot=32025.12——45.4————
LLaDASequence Length=128, Few-shot=32025.12—38.2—————
LLaDASequence Length=256, Few-shot=32025.12—37—————
LLaDASequence Length=512, Few-shot=32025.12—38.2—————
LLaDAFew-shot=32025.12——37.8————
LLaDABackbone=LLaDA-8B-Base, Speedup=1.0×2026.03—39.6———9.53—
LLaDA-1.5Sequence Length=128, Few-shot=32025.12—39.6—————
LLaDA-1.5Sequence Length=256, Few-shot=32025.12—39.9—————
LLaDA-1.5Sequence Length=512, Few-shot=32025.12—38.8—————
LLaDA-1.5Few-shot=32025.12——39.4————
wd1Sequence Length=128, Few-shot=32025.12—38—————
wd1Sequence Length=256, Few-shot=32025.12—37.2—————
wd1Sequence Length=512, Few-shot=32025.12—34.4—————
wd1Few-shot=32025.12——36.5————