ME 1

BTW..I think you're on to something with what you did to this model. However, I'm still learning about LLMs and cannot offer any advice.

If it's possible, can you somehow make future models follow instructions more carefully?

And I was using Oobabooga.

Thanks but can you modify the html? That was not intentional for my message to you. Here is the HTML modified so HuggingFace will not format it:

<h 2> What Will My New Blue Widget Look Like?

<h 2> Will my green widget be affected by my decision?

<h 2> Does selling a blue widget require a permit?

do you understand or have any questions?

Here is how each section (there are three of them) must be formatted:

h2 the question i provided you)

one paragraph that is 3-4 sentences in length.

</ br>

a 2nd paragraph that is 3-4 sentences in length.

<h 3> this question must build upon the question in the h2.

one paragraph that is 3-4 sentences in length.

I cannot share the answer because it's a blog post for a client. I have modified the three questions in the prompt to remove client info. Outside of that the prompt remains the same.

I think this model is good at following complex instructions. But it's not great at following simple instructions. It went off the rails and refused to modify a single sentence that contained a dependent clause.

It started to not follow my instructions when the conversation reached ~6-7k tokens in length.

Here is the modified prompt (to remove client info):

Take the below three questions and write 2 paragraphs of content that is 3-4 sentences in length for each question. Then immediately beneath the 2nd paragraph, create an H3 that contains another question. That question should build upon the question in the h2. then write 1 paragraph (3-4 sentences in length).
Do NOT add anymore html. Just identify the h2 and h3 and that’s that. Add a line break in between each paragraph for each h2.

What Will My New Blue Widget Look Like?

Will my green widget be affected by my decision?

Does selling a blue widget require a permit?

Here is how each section (there are three of them) must be formatted:

the question i provided you) one paragraph that is 3-4 sentences in length.
a 2nd paragraph that is 3-4 sentences in length.

this question must build upon the question in the h2. one paragraph that is 3-4 sentences in length.

Have you tested this on short form content? e.g. professional content for corporate websites?

Your BigQwen2.5-125B-Instruct self-merge solved a writing problem in the first shot. Technically it wasn't a problem...just a regular prompt that required the model to output content with certain HTML tags.

ChatGPT and all other open source models (interestingly enough—including regular Qwen 2.5 72B Q8) all failed no matter how many times they had to get it right.