Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
LLM verified with Monte Carlo Tree Search
| Date | Stars |
|---|---|
| 2026-07-31 | 292 |
| 2026-08-06 | 292 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
[](https://arxiv.org/abs/2402.08147) # LLM verified with Monte Carlo Tree Search This prototype synthesizes verified code with an LLM. Using Monte Carlo Tree Search (MCTS), it explores the space of possible generation of a verified program, and it checks at every step that it's on the right track by calling the verifier. This prototype uses Dafny, Coq, Lean, Scala or Rust. Logs for example runs can be found in the [log](log) directory. Scroll to the very end of a log to see [a chosen solution](https://github.com/namin/llm-verified-with-monte-carlo-tree-search/blob/main/log/opt0_alt.txt#L7661). Note that the linked solution is optimal for the problem. By using this technique, weaker models that might not even know the generated language all that well can compete with stronger models. [Draft](https://arxiv.org/abs/2402.08147) (Outstanding Paper Award at Math-AI workshop co-located with NeurIPS 2024) ## Running This project relies on GPU access. It has been tested on a multi-GPU machine with two NVIDIA A100s. ### Setup Clone the github repo. Note that it has linked submodules so it should be cloned with the following command: ``` git clone --recurse-submodules https://github.com/namin/llm-verified-with-monte-carlo-tree-search.git ``` Note that you will be prompted to paste your huggingface authentication token. ``` conda create --name llm-verified python=3.10 conda activate llm-verified pip install -r requirements.txt huggingface-cli login ``` (If you want to use the singularity sandbox) Download the `llm-verified` docker image (created by the Dockerfile here and pushed to the [hub](https://hub.docker.com/r/namin/llm-verified)) and put it in `~/singularity`. ``` mkdir -p ~/singularity cd ~/singularity singularity pull docker://namin/llm-verified ``` (If you want to use the docker alternative) Download the Docker image: `docker pull namin/llm-verified`. Use the command line `--docker_sandbox`. (If you want to use Dafny) Install Dafny: Download a binary [here](https://github.com/dafny-lang/dafny/releases/latest). Please confirm that Dafny verification works before experimenting, for example by using the `okdafny.py` script we provide. We have sometimes run into issues where z3 was not set up correctly, and Dafny skipped the verification process without returning an error code. Should you run into these problems, compile z3 from source and place the executable into your PATH. (If you want to use Coq) Install Coq: [Install opam](https://opam.ocaml.org/doc/Install.html), then: ``` opam init opam install coq opam install "coq-serapi>=8.10.0+0.7.0" opam repo add coq-released https://coq.inria.fr/opam/released opam install coq-hammer ``` (If you want to use Lean) Install Lean: See detailed instructions [here](https://leanprover-community.github.io/get_started.html). Then, ``` git clone https://github.com/leanprover-community/repl.git ``` and add `require mathlib from git "https://github.com/leanprover-community/mathlib4"` to the file `repl/lakefile.lean` and run ``` curl https://raw.githubusercontent.com/leanprover-community/mathlib4/master/lean-toolchain -o lean-toolchain lake update lake build ``` in the `repl` directory. ### Execution #### Baseline To run the default experiment configuration, which uses the `Phind-CodeLlama-34B-v2` LLM to solve the `problem_opt0` as specified in `prompts.py` in Dafny, do: ``` python run.py ``` Note that `cmdline.py` handles the usage of command line arguments for the codebase. To see what arguments are available, do: ``` python cmdline.py --help ``` To run with specific arguments, simply add a `--[argument name] [desired argument value]` for each argument after the original command to run. For example, to execute `run.py` as before but with Coq instead of Dafny, do: ``` python run.py --language Coq ``` #### Interactive For the run that interacts with the user, do: ``` python run_user.py ``` #### Verifie
Excerpt of 7,700 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:430d82de58fcebeb, topic:llm