The real comparison is memory. There's always a turning point where they get overwhelmed and lose track of what they are doing. The introduction in this situation is a problem, it's chewing up tokens unrelated to the task at hand. Same for code comments, wasting tokens, reducing memory. Thank you for the detailed comparison!
I found it was similar to other open source models, focused on coding, but still greatly lacking in comprehension skills, it cannot make sense of a code base you already have. You need llama 3.1 or 3.2 to do comprehension, and don't use ollama, most models are crippled in ollama, especially llama 3.1, which is still the best open source model for AGI, just use it properly, build it with python and vllm.
![เปิด อาณาจักร หมื่นล้าน "สาระตั้ม" l [Nickynachat]](http://i.ytimg.com/vi/3XE0nPF4YkQ/mqdefault.jpg)
The real comparison is memory. There's always a turning point where they get overwhelmed and lose track of what they are doing.
The introduction in this situation is a problem, it's chewing up tokens unrelated to the task at hand. Same for code comments, wasting tokens, reducing memory.
Thank you for the detailed comparison!
I found it was similar to other open source models, focused on coding, but still greatly lacking in comprehension skills, it cannot make sense of a code base you already have. You need llama 3.1 or 3.2 to do comprehension, and don't use ollama, most models are crippled in ollama, especially llama 3.1, which is still the best open source model for AGI, just use it properly, build it with python and vllm.
What do you mean with crippled in ollama?