yuchenlin/LLM-Blender
[ACL2023] We introduce LLM-Blender, an innovative ensembling framework to attain consistently superior performance by leveraging the diverse strengths of multiple open-source LLMs. LLM-Blender cut the weaknesses through ranking and integrate the strengths through fusing generation to enhance the capability of LLMs.
PythonApache-2.0
Issues
- 2
How to blend generations from different LLMs?
#26 opened by Sunt-ing - 3
- 0
How to use gen_fuser alone for merging?
#25 opened by 20191864218 - 2
Training the GenFuser
#24 opened by jonsaadfalcon - 2
Make `llm-blender` available on PyPI
#23 opened by lewtun - 6
- 9
GPU Memory Requirement on Retraining PairRM
#20 opened by harshyadav17 - 1
How to change models to train ranker?
#21 opened by ashmalvayani - 1
How to act as reward model for RLHF
#16 opened by Andrewzh112 - 2
Training ranker on Unified-Feedback
#17 opened by yifei-he - 1
undefined symbol: _ZNK3c1010TensorImpl36is_contiguous_nondefault_policy_implENS_12MemoryFormatE
#15 opened by vermabasu - 4
Data Generation Code
#12 opened by tgyuan21 - 1
Replicate Experiments
#13 opened by RaccoonOnion - 1
Support for MPS device
#14 opened by fangyuan-ksgk - 4
Load PairRM ranker without Internet access failed
#11 opened by kai01ai - 3
- 3
name 'vllm' is not defined
#9 opened by siddiquelatif - 10
- 4
Issues with the split of dataset
#7 opened by Cascol-Chen - 14
Issue with calculating >= Vic and OA
#6 opened by Cascol-Chen - 4
Question about the fuser outputs
#5 opened by sai4july - 6
- 3
- 3
- 4
Why all the models used in the experiments are released after ACL2023 submission deadline?
#1 opened by Yuanhy1997