5 ms·
Aren't all Qwen models known to perform poorly with system prompt though?
by littlestymaar 2y ago
Aren't all Qwen models known to perform poorly with system prompt though?
- notfromhere 2y agoQwen does decently, DeepSeek doesn't like system prompts. For Qwen you really have to play with parameters
- simonw 2y agoI hadn't heard that, but it would certainly explain why the model made a mess of this task. Tried it again like this, using a regular prompt rather than a system prompt (with the https://github.com/simonw/llm-hacker-news https://github.com/simonw/llm-hacker-news plugin for the hn: prefix): llm -f hn:43825900 \ 'Summarize the themes of the opinions expressed here. For each theme, output a markdown header. Include direct "quotations" (with author attribution) where appropriate. You MUST quote directly from users when crediting them, with double quotes. Fix HTML entities. Output markdown. Go long. Include a section of quotes that illustrate opinions uncommon in the rest of the piece' \ -m qwen3:32b This worked much better! https://gist.github.com/simonw/3b7dbb2432814ebc8615304756395558 https://gist.github.com/simonw/3b7dbb2432814ebc8615304756395...
- littlestymaar 2y agoWow, it hallucinates quotes a lot!
- croemer 2y agoSeems to truncate the input to only 2048 input tokens
- simonw 2y agoOops! That's an Ollama default setting. You can fix that by increasing the num_ctx setting - I'll try running this again. The num_predict setting controls output size.