Issues
- 0
- 0
- 0
Add tests for `Trajectories.to_transitions()`
#113 opened by saleml - 0
Explore ways of vectorizing the flow matching loss
#116 opened by saleml - 0
- 0
Add conditional LogZ calculation
#150 opened by josephdviviano - 5
About having graph in the States
#153 opened by wang422003 - 0
`Transitions` container should then have some `estimator_outputs` attribute to avoid duplicate computation.
#156 opened by josephdviviano - 0
- 0
Replace TorchMPI with CCL in `multinode`
#171 opened by josephdviviano - 0
Documentation of Common Operations Needed
#178 opened by josephdviviano - 0
API Change -- Always Recalculate Everything
#179 opened by josephdviviano - 1
Fix `intro_gfn_continuous_line.ipynb`
#172 opened by saleml - 0
typeguard / CI
#133 opened by saleml - 1
Uniform backward policy is state independent
#180 opened by tristandeleu - 1
Guidelines on finetuning LLMs as policy models
#170 opened by Saltychtao - 6
Conserve LogZ value while using epsilon
#148 opened by OMalenfantThuot - 0
Environment composer
#131 opened by vict0rsch - 6
[help wanted] Traning a DiscreteEBM ends up with "Log probabilities are inf. This should not happen."
#136 opened by ermiaetemadi - 0
Replay buffer broken in `train_hypergrid.py`
#168 opened by josephdviviano - 4
- 2
- 1
Add stack method to States
#161 opened by saleml - 1
`maskless_step` and `maskless_backward_step` should throw an error if you don't return a tensor
#145 opened by josephdviviano - 0
Clear out TODOs in codebase.
#104 opened by josephdviviano - 0
Review and complete all docstrings.
#105 opened by josephdviviano - 0
Ensure that external users who are relying on this library retain all functionality
#107 opened by josephdviviano - 0
Add a test for Trajectories extend function
#108 opened by saleml - 0
Function to revert backward trajectories
#109 opened by saleml - 0
Add type annotations to the file `box_utils.py`
#110 opened by saleml - 0
- 0
Implement flow matching for non discrete environments
#115 opened by saleml - 0
Renaming `batch_dim`
#118 opened by josephdviviano - 1
Manually create more involved documentation
#123 opened by josephdviviano - 0
Sponsored by Foundation?
#164 opened by josephdviviano - 1
Explore renaming `maskless_step` into `step` and `step` into `maskless_step`.
#103 opened by josephdviviano - 1
Environment's should be removed from `GFlowNets` in a way that allows one to easily specify multiple environments (e.g., Multi Objective GFN).
#124 opened by josephdviviano - 1
- 1
Nit: `log_reward_clip_min` isn't defined in `gflownet/base.py` - it depends on definitions in the subclasses.
#155 opened by josephdviviano - 1
Nit: Tensor padding utility function (DRY)
#154 opened by josephdviviano - 2
Rethinking the License
#151 opened by josephdviviano - 2
`test_scripts` failing by small deltas
#138 opened by josephdviviano - 0
Make loss debgging easier
#146 opened by josephdviviano - 0
GFN Object Typing still isn't quite right.
#140 opened by josephdviviano - 3
- 1
- 0
- 0
Implement a uniform forward policy as a random baseline method for easy benchmarking.
#129 opened by josephdviviano - 0
Let's build some black box design examples using DesignBench and using a gflownet (on a simple problem).
#127 opened by josephdviviano - 1