Afshin Oroojlooy

Thanks for the reply. I added `--interval=0.3` (it was `2` by default) and added `--turn` too. But, still it only includes **four** vehicle records. Any suggestion?

Yes, you are right. I got a new file `flow_turn.json`, and I was looking into `flow.json`. But, still it generates 12 flows. How can I get 1000 flow?

How the values for them are obtained? For example why in `"x": 6.96, "y": -4.499999999999999` how `x` is `6.96`? What happens if I change it to `6.92`? How about `5`?...

I am using an old version of cityflow, ([this one](https://github.com/gjzheng93/frap-pub/blob/master/engine.cpython-36m-x86_64-linux-gnu.so)). Are points used in that version?

@sgugger I already have posted it there, and it seems that the issue is not on TRL side.

`torch.autograd.set_detect_anomaly(True)` reports that the root of issue might be in line 201 in `site-packages/transformers/models/gpt2/modeling_gpt2.py`

Turned out that modifying line 201 as below solves the issue. `attn_weights = torch.where(causal_mask.clone(), attn_weights.to(attn_weights.dtype).clone(), mask_value)` Remember that it was: `attn_weights = torch.where(causal_mask, attn_weights.to(attn_weights.dtype), mask_value)` @sgugger Do you know if...

The implementation does not support 3D states, i.e., image inputs. You need to call `Breakout-ram-v0`. > After installing all the prerequisites and trying to run the random_agent.cpp in > cpp...

Same here. It just keeps generating until it gets to its max-gen-limit.