CofeAI
/

FLM-101B

@@ -113,7 +113,8 @@ It uses a 3D(DP+TP+PP) parallelism approach and distributed optimizer.
 <!-- This section is meant to convey both technical and sociotechnical limitations. -->
 Although we've made extensive efforts to thoroughly clean and filter the training corpus for the model, due to the open nature of the dataset, the model may still have picked up on some unsafe examples. Consequently, the model may still generate unexpected content, including but not limited to discrimination, bias, or offensive language. We would like to strongly advise users not to spread any unsafe content generated by the model. The project developers cannot be held responsible for any repercussions stemming from the dissemination of harmful information.
 ## Citation

 <!-- This section is meant to convey both technical and sociotechnical limitations. -->
 Although we've made extensive efforts to thoroughly clean and filter the training corpus for the model, due to the open nature of the dataset, the model may still have picked up on some unsafe examples. Consequently, the model may still generate unexpected content, including but not limited to discrimination, bias, or offensive language. We would like to strongly advise users not to spread any unsafe content generated by the model. The project developers cannot be held responsible for any repercussions stemming from the dissemination of harmful information.
+At the current stage of training, FLM-101B has a relatively low token count, leaving significant room for improvement in knowledge, especially in specialized domains. Additionally, the model's inference process is not yet optimized, leading to high resource usage and limited speed. We will soon introduce support for Flash Attention.
+If you have suggestions for improvement in these areas or any other aspects, please feel free to open an issue on GitHub, and we will respond promptly. Thank you!
 ## Citation

README_zh.md CHANGED Viewed

@@ -9,9 +9,9 @@
 # FLM-101B
-FLM-101B是一个开源的decoder-only架构的语言模型,参数规模101B.训练过程采用模型生长技术，通过训练前期在小规模模型上快速学习知识，后期将模型逐步生长成大模型的方式，实现了千亿规模模型的低成本(~$100K)训练。
 FLM-101B支持中英双语，训练上下文窗口长度为2048，得益于使用了xPos旋转位置编码，推理时窗口大小可进行良好的拓展。
-为推动千亿规模LLM技术发展,FLM-101B现已全面开源.
 ## 为什么使用FLM-101B
@@ -96,11 +96,11 @@ print(tokenizer.decode(generated.cpu()[0], skip_special_tokens=True))
 #### 硬件
-FLM-101B在24节点DGX-A800 GPU(8×80G)集群上完成的训练,总耗时近26天.基于模型生长策略,我们依次在该集群上进行了16B, 51B和101B的模型的训练和生长.
 #### 软件
-FLM-101B的训练代码Megatron-FLM基于Megatron-LM框架修改,将在近期开源.
 框架支持3D并行策略以及分布式优化器.
 ## 偏见、风险与限制
@@ -109,6 +109,9 @@ FLM-101B的训练代码Megatron-FLM基于Megatron-LM框架修改,将在近期开
 尽管我们已经尽最大努力对模型训练语料进行了清洗过滤，但由于训练语料的开放性，模型仍有可能在一些不安全的语料上进行过学习。因此模型仍有可能生成不符合预期的文本，包括但不限于歧视、偏见、谩骂等。我们在此提醒模型使用者，请勿传播模型可能生成的不安全内容。由于传播不良信息导致的任何后果，本项目开发者不承担责任。
 ## Citation
 <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

 # FLM-101B
+FLM-101B是一个开源的decoder-only架构的语言模型，参数规模101B.训练过程采用模型生长技术，通过训练前期在小规模模型上快速学习知识，后期将模型逐步生长成大模型的方式，实现了千亿规模模型的低成本(~$100K)训练。
 FLM-101B支持中英双语，训练上下文窗口长度为2048，得益于使用了xPos旋转位置编码，推理时窗口大小可进行良好的拓展。
+为推动千亿规模LLM技术发展，FLM-101B现已全面开源.
 ## 为什么使用FLM-101B
 #### 硬件
+FLM-101B在24节点DGX-A800 GPU(8×80G)集群上完成的训练，总耗时近26天.基于模型生长策略，我们依次在该集群上进行了16B， 51B和101B的模型的训练和生长.
 #### 软件
+FLM-101B的训练代码Megatron-FLM基于Megatron-LM框架修改，将在近期开源.
 框架支持3D并行策略以及分布式优化器.
 ## 偏见、风险与限制
 尽管我们已经尽最大努力对模型训练语料进行了清洗过滤，但由于训练语料的开放性，模型仍有可能在一些不安全的语料上进行过学习。因此模型仍有可能生成不符合预期的文本，包括但不限于歧视、偏见、谩骂等。我们在此提醒模型使用者，请勿传播模型可能生成的不安全内容。由于传播不良信息导致的任何后果，本项目开发者不承担责任。
+FLM-101B 现阶段训练的 token 数比较少，在知识（特别是专业知识）方面有较大进步空间。另一方面，模型的推理目前没有进行优化因此推理资源占用较高，速度受限。我们很快将支持Flash Attention。
+如果您对这两方面，或其他方面有改进需求，请在 github 提issue，我们会尽快响应。谢谢！
 ## Citation
 <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->