It will load from cache if downloaded once. Uses the Web-GPU TransformersJS DeepseekR1 model or other models:
Data warning ~1.4 GB saved to cache
Loading progress: 0%
Max tokens in the answer: kind of cuts off the reply
Top_p: closer to zero more focused, closer to 1 more variety
Temperature: Close to 0 more predictable, closer to 1 more diverse
Do_sample: when selected, picks and considers more token options making it varied and cretive, not selected keeps safest token
Chain_of_thought: AI explains more about it's token creation process.