Daniel

建议做一个python的直接安装用的轮子

7

方便普通用户直接用pip 进行安装使用，里面可以包括转好的模型或者把模型分开包安装。

6000 ADA has a capacity of 48GB memory . How do I tuning it with QLoRA? I guess QLora cannot download pre-trained model online, how should I do for it?

Errors happen during loading llama 65B for tuning.

5

ValueError: Target modules ['dense_h_to_4h', 'dense', 'dense_4h_to_h', 'query_key_value'] not found in the base model. Please check the target modules and try again. python qlora.py –learning_rate 0.0001 --model_name_or_path decapoda-research/llama-65b-hf which model can...

Could it load and tune falcon-40B ?

4

TIIUAC has released a promising model named falcon: https://huggingface.co/tiiuae . Can QLora load and tune it? It is totally free and can be used for commercial purposes.

Can GGML apply in CNN or RCNN ?

2

This quantization scheme can speed up the inference of neural networks, but there still is less example in CNN or RCNN, even RNN. Is these ones not easy on ggml...

It is definitely a milestone to the community of LLM.

4

The performance of InternLM is amazing, if you guys release the weight of InternLM to the community, it will proke the LM community advanced.

--- name: Pull Request title: '[PR]' assignees: 'BAAI-OpenPlatform,ftgreat' --- ### Description bugfix for: from torch._six import inf ModuleNotFoundError: No module named 'torch._six' ### Checklist - [ ] bug fixed -...

经过测试发现7B模型的中文能力不错，但是数理逻辑比较差（预料之中），有开放33B/30B级别的计划吗？

11

### Description 中文能力确实不错，与chatGLM 6B相当，但是数理逻辑表现比较低( 也与chatGLM6B)相当。 什么时候可以开放30B级别的模型？社区期待这方面的能力有所提升。 ### Alternatives _No response_

180B 测试结果

6

中文处理能力尚可，文本摘要比较稳定。 数理逻辑比较欠缺（相当于没有） 指令遵循能力尚可，存在听不懂的情况。 总结下：任重道远，相当于chatGLM 6B类似的水平。

Is there a tutorial for customing operators?

2

Need a crash course for customing GGML.