OKC0 comments

Results 11 comments of


                                            OKC0

请问预训练的lr多大呢 ---原始邮件--- 发件人: "Yiming ***@***.***> 发送时间: 2024年5月2日(周四) 上午9:51 收件人: ***@***.***>; 抄送: ***@***.******@***.***>; 主题: Re: [ymcui/Chinese-LLaMA-Alpaca-3] 预训练base时的学习率多少 (Issue #18) 我们SFT阶段使用的学习率是1e-4，bs=1K，你可以根据实际情况调整（观察开发集loss等手段）。 — Reply to this email directly, view it on GitHub, or...

space-t微调训练数据构造

space-t 表格sql模型

[Question]: Is there any way to support the deployment of a local embedding API and a reranker API?

LlamafileEmbedding can only load embeddings related to llamafile, right?How about the reranker API? How to solve it?

[Question]: Is there any way to support the deployment of a local embedding API and a reranker API?

not a good answer

微调大概需要多少样本？

此外，在自定义数据集训练，想知道七检测能力与yolo v10 哪个更好？

Flash Attention 2.0 only supports torch.float16 and torch.bfloat16 dtypes, but the current dype in Qwen2ForCausalLM is torch.float32.

> ![Image](https://github.com/user-attachments/assets/1b6dd5cf-2bd8-4a0e-bddc-aa4efa8d871c) same problem I guess you cloud add model_dtype in config to solve it ![Image](https://github.com/user-attachments/assets/a06e2fcd-d21c-473f-a4d3-7fdedc8caa62)

Flash Attention 2.0 only supports torch.float16 and torch.bfloat16 dtypes, but the current dype in Qwen2ForCausalLM is torch.float32.

> ![Image](https://github.com/user-attachments/assets/1b6dd5cf-2bd8-4a0e-bddc-aa4efa8d871c) same problem I guess you cloud add model_dtype in config to solve it Yeah，fixed , just fix the actor module

[Usage]: ValueError: The checkpoint you are trying to load has model type `qwen2_5_vl` but Transformers does not recognize this architecture

face the same issue,NOW

第二次启动发现不能分割

t同问