LLM Parameter Settings
Parameter Settings
-
To best leverage the programming capabilities of LLM models, several key parameter settings need to be adjusted.
-
Previously, these were adjusted manually, but actually, we can let the LLM models recommend their own parameter settings.
-
After asking each LLM for its own parameter setting recommendations, the following settings are currently adopted:
| Model | context Length | temperature | topP |
|---|---|---|---|
| gemini1.5 pro | 4096 | 0.2 | 0.95 |
| gpt-4o | 4096 | 0.2 | 0.95 |
| Grok2 beta | 4096 | 0.2 | 0.95 |
| Mistral Large | 8192 | 0.1 | 0.9 |
| GPT-4o mini | 4096 | 0.2 | 0.95 |
| Llama3.2 90b | 4096 | 0.2 | 0.95 |
| Doubao | 4096 | 0.1 | 1 |
| Qwen2.5 72b | 4096 | 0.2 | 0.9 |
| Llama3.1 405b | 4096 | 0.2 | 0.9 |
- Among them, Mistral Large seems very confident about the length of its context window. However, judging from the responses of other models, such a large context window is only needed for slightly larger projects. Let’s try it out and see.