chore: update Qwen model params #2892
Conversation
dosubot
bot
added
the
size:L
api/core/model_runtime/model_providers/tongyi/llm/qwen-max-1201.yaml
Outdated
Show resolved
Hide resolved
| max: 2.0 | ||
| help: | ||
| zh_Hans: 用于控制随机性和多样性的程度。具体来说，temperature值控制了生成文本时对每个候选词的概率分布进行平滑的程度。较高的temperature值会降低概率分布的峰值，使得更多的低概率词被选择，生成结果更加多样化；而较低的temperature值则会增强概率分布的峰值，使得高概率词更容易被选择，生成结果更加确定。 | ||
| en_US: Used to control the degree of randomness and diversity. Specifically, the temperature value controls the degree to which the probability distribution of each candidate word is smoothed when generating text. A higher temperature value will reduce the peak value of the probability distribution, allowing more low-probability words to be selected, and the generated results will be more diverse; while a lower temperature value will enhance the peak value of the probability distribution, making it easier for high-probability words to be selected. , the generated results are more certain. | ||
| - name: max_tokens | ||
| use_template: max_tokens | ||
| required: false |
There was a problem hiding this comment.
It's quite confused when setting all these common parameters (max_tokens, top_p and etc.) into not required by default.
These configs are not just for calling the LLM APIs, but also for the balance of productive and deliverable app constructions.
There was a problem hiding this comment.
What I hope is that when the front-end is displayed, these options can be unchecked by default.
You are more familiar with the project, so I will follow your suggestion.
|
Btw, as you're updating Qwen model's specification, use |
soulteary
changed the title
| @@ -8,54 +8,65 @@ model_properties: | |||
| parameter_rules: | |||
| - name: temperature | |||
| use_template: temperature | |||
| default: 1.0 | |||
| type: float | |||
| default: 0.85 | |||
There was a problem hiding this comment.
Btw, the default temperature here seems too high. In our practice on Qwen, it will make LLM's answer unrelated to the recalled knowledge. Would prefer setting relatively lower in 0.1 for more stable answer.
Description
Updated model calling parameters based on official website information and SDK parameters.
https://help.aliyun.com/zh/dashscope/developer-reference/api-details?disableWebsiteRedirect=true
https://help.aliyun.com/zh/dashscope/developer-reference/api-details
Type of Change
How Has This Been Tested?
Suggested Checklist:
dev/reformat(backend) andcd web && npx lint-staged(frontend) to appease the lint godsoptionalI have made corresponding changes to the documentationoptionalI have added tests that prove my fix is effective or that my feature worksoptionalNew and existing unit tests pass locally with my changes