Yanan

文档错误

1

https://paddlepedia.readthedocs.io/en/latest/tutorials/deep_learning/metrics/perplexity.html 这里 > 一般来讲，用于评判语言模型的测试集均是合理的、高质量的语料，只要语言模型在测试集上的困惑度越高，则代表语言模型训练地越好，反之亦然。 应该是测试集上困惑度越低， 越好.

Are the bag of words case-sensitive?

1

Hello, I find that some words are cased while some are uncased. They have different word ids in the vocab of tokenizer of GPT. What is the appropriate way to...

Dev loss does not drop when finetuning

1

hello, I am reusing the code for finetuning the model , such as CBERT. However, the dev loss does not drop while the train loss on train set seems normal....

Test TracIn's effectiveness in text classification

3

Hello, I adopt the code from https://github.com/frederick0329/TracIn/blob/master/imagenet/resnet50_imagenet_proponents_opponents.ipynb to text classification. The primary goal of my task is to rank the training samples based on their positive or negative impacts on...

I want to implement an idea using tensorflow and gpt2(huggingface transformers version). For each iteration, I just let the gpt2 produce some sentences, and these sentences are fed into a...

How to liberate the gpt2 from reference model?

8

Hi, We know that KL is used in the loss as a constraint for the difference between the original gpt2 and the active gpt2 which produces responses for rewards feedbacks....

hi there. for the code https://github.com/huggingface/trl/blob/main/examples/scripts/ppo.py#L115, the ref model is set to None when not using peft. This seems to cause error below since the None ref_model is passed to...

[BUG]output tensor must have the same type as input tensor in PPO training script of TRL

6

hi there, i train the model with TRL - ppo following https://github.com/huggingface/trl/blob/main/examples/scripts/ppo.py with the accelerate config: deepspeed_zero3.yaml: ``` compute_environment: LOCAL_MACHINE debug: false deepspeed_config: deepspeed_multinode_launcher: standard gradient_accumulation_steps: 4 offload_optimizer_device: cpu offload_param_device:...

[Bug]: NameError: name 'vllm_ops' is not defined

8

### Your current environment vllm version: 0.4.0.post1 ### 🐛 Describe the bug `CUDA_VISIBLE_DEVICES="4" python -u -m vllm.entrypoints.openai.api_server --model mistralai/Mistral-7B-Instruct-v0.2 --dtype auto --api-key yanan --tensor-parallel-size 1 --port 1703 --host 0.0.0.0 --worker-use-ray...

Hi, We know that KL is used in the loss as a constraint for the difference between the original gpt2 and the active gpt2 which produces responses for rewards feedbacks....