-
你好!MiniCPM-v2已被证实在提取高分辨率图片中的文本信息时拥有令人惊叹的准确度,并且远强于ominillm-12b,是非常优秀的模型。然而,可能是llm参数量太小的原因,MiniCPM-v2遵循人类指令和生成较长文本的能力比较弱。非常期待更大参数规模版本的MiniCPM-v2。
-
### 起始日期 | Start Date
_No response_
### 实现PR | Implementation PR
_No response_
### 相关Issues | Reference Issues
_No response_
### 摘要 | Summary
按照大佬提供的finetune_ds.sh 训练发现loss为0,不知道哪里有问题
### 基本示例…
-
### 是否已有关于该错误的issue或讨论? | Is there an existing issue / discussion for this?
- [X] 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions
### 该问题是否在FAQ中有解答? | Is there an existing…
xyw1 updated
2 months ago
-
### Feature request / 功能建议
ollama怎么部署新版v2
-
### 是否已有关于该错误的issue或讨论? | Is there an existing issue / discussion for this?
- [X] 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions
### 该问题是否在FAQ中有解答? | Is there an existing ans…
-
### 是否已有关于该错误的issue或讨论? | Is there an existing issue / discussion for this?
- [X] 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions
### 该问题是否在FAQ中有解答? | Is there an existing ans…
-
web_demo.py 运行需要10G显存。为啥 vllm 启动api模式运行,需要23G显存啊?
python web_demo.py --device cuda --dtype bf16
Vllm运行命令如下
/services/srv/MiniCPM-vllm/venv/bin/python -m vllm.entrypoints.openai.api_server --mo…
-
## 🐛 Bug
MODEL_TYPE=minicpm_v
(tvm-build-venv)~/Documents/mlc-MiniCPM$ mlc_chat convert_weight --model-type ${MODEL_TYPE} ./dist/models/${MODEL_NAME}/ --quantization $QUANTIZATION -o dist/$MODEL_NAM…
-
### 是否已有关于该错误的issue或讨论? | Is there an existing issue / discussion for this?
- [X] 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions
### 该问题是否在FAQ中有解答? | Is there an existing ans…
-
Thanks for your great work! Can you support MiniCPM backbone?