Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

LM1B

Benchmarks

Task NameDataset NameSOTA ResultTrend
Language modellingLM1B (test)
Perplexity20.86
151
Language ModelingLM1B
PPL (Generalized)48.5
93
Text GenerationLM1B (test)
Entropy2.46
85
Language ModelingLM1B (val)
Perplexity16.57
67
Unconditional generationLM1B sequence length 128
Generation Perplexity (PPL)40.2
43
Language ModelingLM1B
Perplexity22.8
39
Unconditional generationLM1B
Generation Perplexity36.42
31
Unconditional Text GenerationLM1B
Entropy4.29
24
Text GenerationLM1B
Perplexity (PPL)68.11
24
Hyperparameter OptimizationPD1-LM1B (val)
Validation Error0.628
24
Language GenerationLM1B 1024 sequences of length 128
Generative PPL186.79
20
Language ModelingLM1B zero-shot
Perplexity51.25
20
Language ModelingLM1B L=128 (test)
NELBO PPL24.53
17
Language ModelingLM1B (test)
Block Efficiency8.94
15
Speculative DecodingLM1B (test)
BE7.88
10
Language ModelingLM1B GPT-2 small model size equivalent (test)
Perplexity20.53
10
Autoregressive Language ModelingLM1B
PPL21.5
7
Language ModelingLM1B GPT2
PPL65.629
4
Language ModelingLM1B ctx len. 128 (val)
PPL (Val)25.72
3
Text GenerationLM1B (val)
Perplexity51.25
1
Auto-regressive language modelingLM1B 1.0 (test)
Metric-
0
Showing 21 of 21 rows