Higher LLM temperature settings yield the best English augmentation performance, with Llama temperature 2 the top generator overall¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Llama with temperature = 2 was the top performer among the five generator/temperature combinations for the English rainwater concepts. → Keith Cochran 2025 q2 i? For the multi-class French dataset, a lower Llama temperature of 1.5 produced the highest classifier performance. → Keith Cochran 2025 (2)

Evidence¶

Keith Cochran 2025¶

Keith Cochran, Clayton Cohn, Jean-François Rouet, Peter Hastings. (2025). Comparing Text Augmentation by GPT-3.5 and Llama3 for Evaluating Student Responses. International Journal of Artificial Intelligence in Education. https://doi.org/10.1007/s40593-025-00473-x

q2 · i? · causal · r2

Comparison of maximum micro-F1 per augmentation level across generator/temperature combinations for the English rainwater concepts, reported in Tables 9–13. Llama temperature 2 "was the top performer".

"Examining the maximum performance at each augmentation level revealed that Llama with temperature = 2 was the top performer among the five generator / temperature combinations used for student response generation."

Keith Cochran 2025 (2)¶

Keith Cochran, Clayton Cohn, Jean-François Rouet, Peter Hastings. (2025). Comparing Text Augmentation by GPT-3.5 and Llama3 for Evaluating Student Responses. International Journal of Artificial Intelligence in Education. https://doi.org/10.1007/s40593-025-00473-x

q2 · i? · causal · r2

Result for the four-label French opinion dataset, where the best configuration (Classla model, 89x, temperature 1.5) reached micro-F1 0.939 in Table 13. The authors state the diversity hypothesis was "only partially supported".

"However, the lower temperature = 1.5 for Llama produced the highest classifier performance for the French video games data set. For this reason, this hypothesis was only partially supported."

Discussion¶

Related Claims¶

• LLM-generated responses outperform self-augmentation when augmenting small educational classification datasets — related
• LLM temperature settings significantly affect generated-text length and complexity, with Llama3 producing more complex texts than GPT-3.5 — related
• Temperature-construct interactions: constructs with moderate theoretical coherence benefited from higher temperatures, while well-defined constructs required deterministic settings — related
• Hyperparameter setting with maximum token 53 and temperature 1 (Set C) yields the most accurate AGE grading performance