async-profiler VS trl

We'll use async-profiler and flame graphs for profiling. To simplify the process, we'll run the code using JBang.

JIT'ed code is sadly poorly supported, but LLVM has had great hooks for noting each method that is produced and its address. So you can build a simple mixed-mode unwinder, pretty easily, but mostly in process.

I think Intel's DNN things dump their info out to some common file that perf can read instead, but because the *kernels* themselves reuse rbp throughout oneDNN, it's totally useless.

Finally, can any JVM folks explain this claim about DWARF info from the article:

> Doesn't exist for JIT'd runtimes like the Java JVM

that just sounds surprising to me. Is it off by default or literally not available? (Google searches have mostly pointed to people wanting to include the JNI/C side of a JVM stack, like https://github.com/async-profiler/async-profiler/issues/215).

debugging native calls in itself is also painful. I have switched to using async-profiler (https://github.com/async-profiler/async-profiler) instead of JFR for most of my usecases.

A. it tracks native calls by default

https://github.com/async-profiler/async-profiler#flame-graph...

Ok, Windows is not supported. But IntelliJ made a fork which works on Windows.

Hi, today I used async-profiler to check the CPU usage of my Spring Boot app (just a normal backend) in production. Surprisingly, Lettuce (Redis) + Mybatis (MySQL) take up most of the CPU time. I am not talking about wall time here, but CPU time, since I know database requests need to wait for milliseconds and thus wall time will be very long. Therefore, I wonder:

You can use async-profiler to see what is happening under the hood.

I have used visual vm quite a bit. https://github.com/async-profiler/async-profiler is also amazing... Throw the binary on the system and fire it up. It also profiles down into native code as well if you do that kind of thing.

I am using TRL for RLHF fine-tuning the Llama-2-7B model and getting an OOM error (even with batch_size=1). If anyone used TRL for RLHF can please tell me what I am doing wrong? Code details can be found in the GitHub issue.

SFTtrainer from huggingface

For lora - PEFT seems to work. I don't have patience to wait 5 hours, but modifying this example seems to work. You don't even need to modify that much, as their model just as neo-x uses query_key_value name for self-attention.

They have examples of making GPT output more positive (code) by using a sentiment model as reward. There are other examples about reducing toxicity, summarization here: https://github.com/lvwerra/trl/tree/main/examples . Should be fairly simple to modify the sentiment example and try the calculator reward you mentioned above.

You can use this -> https://github.com/lvwerra/trl/blob/main/examples/sentiment/scripts/gpt-neox-20b_peft/merge_peft_adapter.py

Just the hh directly. From the results it seems like it might possibly be enough but I might also try instruction tuning then running the whole process from that base. I will also be running the reinforcement learning by using a Lora using this as an example https://github.com/lvwerra/trl/tree/main/examples/sentiment/scripts/gpt-neox-20b_peft

This package is pretty simple to use! https://github.com/lvwerra/trl