LLaMPPL
LLaMPPL copied to clipboard
A domain-specific probabilistic programming language for modeling and inference with language models
[I put your changes of llama.cpp into the most recent llama.cpp](https://github.com/Phylliida/llama.cpp) Then I had to modify LLaMPPL/llamppl/llama_cpp.py to use the new code from llama_cpp_python, you can see the new file...
I will be playing with this, as I have a vested interest in seeing proper constraints implemented in LLMs which solve the issues that you outline with the technique I...
In models that sample tokens from the prior, it is unnecessary to actually run the LLM on the newly sampled token unless the particle survives the next resampling step. Maybe...
When to resample in SMC? Currently particles are aligned by number-of-tokens, so when we resample, all particles have the same number of tokens (unless some have already hit EOS). But...