Weird repetition and rambling in the though process
Im testing your q6_k_xl quant rn, and after around 16k rambling without any real direction it started to just repeat "g" for ever.
Its probably a sampler issue, ive just used your sampling settings from your r1.1 page, so we probably have to find optimal settings before this model can be used to its full potential, with the normal qwen3 thinking preset it has a similar issue, a lot of meaningless ramble in the thought process and then repetition after 10k tokens.
These are the two presets i tested:
this btw happened any time ive tested the model, my prompt was always
Solve the following physics problem. Write your solution in LaTeX and enclose your final answer in a box. Problem Statement:
For the following problem, work under the assumption that interstellar matter is in local thermal equilibrium. The ratio of pressure to density is a constant ratio of $v_s^2$ , but the initial density $\rho(r)$ that has the unrealistic form $\rho(r) = \frac{k}{r}$, where $r$ is the distance to a point at $r = 0$ and $k$ is a constant.
- What is the initial radius of the smallest sphere centered at $r = 0$ that will undergo gravitational collapse?
- Verify your answer by determining both the kinetic and gravitational self-energy at the value of the radius found in part (a) and verify that the values you find satisfy the viral theorem.
Oh bruh ive just seen that this is the wrong model page /: