[ghstack-poisoned]

We plan to run the dynamo produced bytecode to retrace the pytree flatten/unflatten calls into fx nodes. One issue with this approach is that we would silently replay side effects. In export, we don't want to do that. So we add a config in dynamo to turn off replaying side effects. Potential users of this config are:
1. export 
2. VLLM (cc: zhxchen17)


cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames Lucaskabela

[ghstack-poisoned]

There are two motivating use cases for this change:
1) export (when we trace pytree calls into a graph, we don't want to accidentally trace the side effect bytecode which will pollute the initial state) -> We want to warn about side effects and don't want to actually apply them 
2) VLLM -> They want to detect side effects and error out. 

We implement this with two configs where one config controls whether we want to apply side effects (by default yes) and the warning level for side effects (warning for export and error for VLLM). We intentionally ignore input side effects, because they are captured in the graph and export would never trace the actual dynamo graph module when tracing the pytree calls). 


cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames Lucaskabela

[ghstack-poisoned]

There are two motivating use cases for this change:
1) export (when we trace pytree calls into a graph, we don't want to accidentally trace the side effect bytecode which will pollute the initial state) -> We want to warn about side effects and don't want to actually apply them 
2) VLLM -> They want to detect side effects and error out. 

We implement this with two configs where one config controls whether we want to apply side effects (by default yes) and the warning level for side effects (warning for export and error for VLLM). We intentionally ignore input side effects, because they are captured in the graph and export would never trace the actual dynamo graph module when tracing the pytree calls). 


cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames Lucaskabela

[ghstack-poisoned]

There are two motivating use cases for this change:
1) export (when we trace pytree calls into a graph, we don't want to accidentally trace the side effect bytecode which will pollute the initial state) -> We want to warn about side effects and don't want to actually apply them 
2) VLLM -> They want to detect side effects and error out. 

We implement this with two configs where one config controls whether we want to apply side effects (by default yes) and the warning level for side effects (warning for export and error for VLLM). We intentionally ignore input side effects, because they are captured in the graph and export would never trace the actual dynamo graph module when tracing the pytree calls). 


cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames Lucaskabela

[ghstack-poisoned]

There are two motivating use cases for this change:
1) export (when we trace pytree calls into a graph, we don't want to accidentally trace the side effect bytecode which will pollute the initial state) -> We want to warn about side effects and don't want to actually apply them 
2) VLLM -> They want to detect side effects and error out. 

We implement this with two configs where one config controls whether we want to apply side effects (by default yes) and the warning level for side effects (warning for export and error for VLLM). We intentionally ignore input side effects, because they are captured in the graph and export would never trace the actual dynamo graph module when tracing the pytree calls). 


cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames Lucaskabela

[ghstack-poisoned]

There are two motivating use cases for this change:
1) export (when we trace pytree calls into a graph, we don't want to accidentally trace the side effect bytecode which will pollute the initial state) -> We want to warn about side effects and don't want to actually apply them 
2) VLLM -> They want to detect side effects and error out. 

We implement this with two configs where one config controls whether we want to apply side effects (by default yes) and the warning level for side effects (warning for export and error for VLLM). We intentionally ignore input side effects, because they are captured in the graph and export would never trace the actual dynamo graph module when tracing the pytree calls). 


cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames Lucaskabela

[ghstack-poisoned]

There are two motivating use cases for this change:
1) export (when we trace pytree calls into a graph, we don't want to accidentally trace the side effect bytecode which will pollute the initial state) -> We want to warn about side effects and don't want to actually apply them 
2) VLLM -> They want to detect side effects and error out. 

We implement this with two configs where one config controls whether we want to apply side effects (by default yes) and the warning level for side effects (warning for export and error for VLLM). We intentionally ignore input side effects, because they are captured in the graph and export would never trace the actual dynamo graph module when tracing the pytree calls). 


cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames Lucaskabela

[ghstack-poisoned]