The problem: Whenever the attached tools to the ollama client are more than 0, it forces the model to call one of the attached function even if the user prompt does not require it to do so. I believe the reason is related to the default system prompt shown bellow: Here is the default system template used with llama3.2: Cutting Knowledge Date: December 2023 You are a helpful assistant with tool calling capabilities. When you receive a tool call response, use the output to format an answer to the original user question. Given the following functions, please respond with a JSON for a function call with its proper arguments that best answers the given prompt. {tools signatures injected here} Respond in the format {"name": function name, "parameters": dictionary of argument name and its value}. Do not use variables. As you can see, unless the model received a pre-called function response, ollama default llama3.2 system prompt is forcing the model to ALWAYS call a function even it the user prompt does not require the model to do so. The result is: When asking the model to categorize the items it will mostly, respond with tool_calls that will generate unavoidable error. Any way, thanks for the useful content. Have a nice day.
This is my first time seeing someone demonstrate using a python script to control A.I. Interesting. I'm a senior dev, and we only use shell (or compile a binary). This actually looks more verbose with python.
Good video, although incredibly dizzying as you're zoomed into the code, then jump up and down the code and also just paste random snippets of code. I get what you are trying to do to explain it, but it is hard to follow. Perhaps, It would be a good idea to share your code like other YT creators. a) It makes learning easier and b) gives some real value to users since, they don't get detered trying to fix problems that you haven't had/fixed already. Hide more sophisticated applications behind paid subscriptions, yes, but give yourself a chance by minimising the risk of potential subscribers jumping to your competitors because they've made things easier and allowed viewers to enjoy small wins. Just a suggestion for your consideration.
Thanks for your feedback. I really appreciate it. About the code, I forgot to include the link to the GitHub repo :(. The code link is in the descriptions now (or here:github.com/pdichone/ai-guild-ai-llama-categorizer). Again, thanks for your feedback :)
@@vincibits Thanks for taking on the feedback and sharing the code. It'll really help you scale your channel. Makes it more interactive and encourages subscription and engagement! I've left you my first 'issue' on Github. Your video makes more sense and the code works with one exception. Errors occur after the code has run and ayncio tries to close.
Join the AI Guild Community: 🚀 bit.ly/ai-guild-join 🚀
thankk you, very easy to follow
The problem: Whenever the attached tools to the ollama client are more than 0, it forces the model to call one of the attached function even if the user prompt does not require it to do so.
I believe the reason is related to the default system prompt shown bellow:
Here is the default system template used with llama3.2:
Cutting Knowledge Date: December 2023
You are a helpful assistant with tool calling capabilities.
When you receive a tool call response, use the output to format an answer to the original user question.
Given the following functions, please respond with a JSON for a function call with its proper arguments that best answers the given prompt.
{tools signatures injected here}
Respond in the format {"name": function name, "parameters": dictionary of argument name and its value}. Do not use variables.
As you can see, unless the model received a pre-called function response, ollama default llama3.2 system prompt is forcing the model to ALWAYS call a function even it the user prompt does not require the model to do so.
The result is: When asking the model to categorize the items it will mostly, respond with tool_calls that will generate unavoidable error.
Any way, thanks for the useful content. Have a nice day.
This is my first time seeing someone demonstrate using a python script to control A.I. Interesting.
I'm a senior dev, and we only use shell (or compile a binary). This actually looks more verbose with python.
Thank you for your comment! Yes, Python can be verbose, but not as verbose as Java :)
Nice
Thanks 😊
You’re welcome!!
Good video, although incredibly dizzying as you're zoomed into the code, then jump up and down the code and also just paste random snippets of code. I get what you are trying to do to explain it, but it is hard to follow. Perhaps, It would be a good idea to share your code like other YT creators. a) It makes learning easier and b) gives some real value to users since, they don't get detered trying to fix problems that you haven't had/fixed already.
Hide more sophisticated applications behind paid subscriptions, yes, but give yourself a chance by minimising the risk of potential subscribers jumping to your competitors because they've made things easier and allowed viewers to enjoy small wins.
Just a suggestion for your consideration.
Thanks for your feedback. I really appreciate it. About the code, I forgot to include the link to the GitHub repo :(. The code link is in the descriptions now (or here:github.com/pdichone/ai-guild-ai-llama-categorizer). Again, thanks for your feedback :)
Fully agree
@@kmwr57 Thank you so much for your feedback. I really appreciate it.
@@vincibits Thanks for taking on the feedback and sharing the code. It'll really help you scale your channel. Makes it more interactive and encourages subscription and engagement!
I've left you my first 'issue' on Github. Your video makes more sense and the code works with one exception. Errors occur after the code has run and ayncio tries to close.
2025? 24 almost over
Exactly that's how powerful it is 😂
Right? It’s like the secret sauce no one told us about!