diff --git a/README.md b/README.md index de2864e758f527f5aa16a91ab1738b79acb9d675..4f77effa29669c91331ad7b1a3ce72fc1ec5f2c4 100644 --- a/README.md +++ b/README.md @@ -41,12 +41,14 @@ DeepSparkHub甄选上百个应用算法和模型,覆盖AI和通用计算各领 | [Llama3-8B](nlp/llm/llama3_8b/pytorch) | PyTorch | Megatron-DeepSpeed | Bookcorpus | 4.1.1 | | [Llama3-8B](nlp/llm/llama3_8b/megatron-lm) | PyTorch | Megatron-LM | GPT Small-117M | 4.3.0 | | [Llama3-8B SFT](nlp/llm/llama3_8b_sft/pytorch) | PyTorch | ColossalAI | school_math_0.25M | 4.1.1 | +| [Llama3-8B SFT](nlp/llm/llama3_8b/openrlhf) | PyTorch | OpenRLHF | Meta-Llama-3-8B | 4.3.0 | | [Llama3-8B PPO](nlp/llm/llama3_8b/openrlhf) | PyTorch | OpenRLHF | Llama-3-8b-sft-mixture | 4.2.0 | | [Llama3-8B DPO](nlp/llm/llama3_8b/openrlhf) | PyTorch | OpenRLHF | Llama-3-8b-sft-mixture | 4.2.0 | | [Llama3-8B KTO](nlp/llm/llama3_8b/openrlhf) | PyTorch | OpenRLHF | Llama-3-8b-sft-mixture | 4.2.0 | | [Mamba-2](nlp/llm/mamba-2/pytorch) | PyTorch | Megatron-LM | GPT Small-117M | 4.1.1 | | [MiniCPM](nlp/llm/minicpm/pytorch) | PyTorch | DeepSpeed | MiniCPM-2B-sft-bf16 | 4.2.0 | | [Mixtral 8x7B](nlp/llm/mixtral/pytorch) | PyTorch | Megatron-LM | GPT Small-117M | 4.1.1 | +| [Mixtral 8x7B](nlp/llm/mixtral/openrlhf) | PyTorch | OpenRLHF | Mixtral-8x7B-v0.1 | 4.3.0 | | [Phi-3](nlp/llm/phi-3/pytorch) | PyTorch | Torchrun | Phi-3-mini-4k-instruct | 4.2.0 | | [QWen-7B](nlp/llm/qwen-7b/pytorch) | PyTorch | Firefly | qwen-7b | 3.4.0 | | [QWen1.5-7B](nlp/llm/qwen1.5-7b/pytorch) | PyTorch | Firefly | school_math | 4.1.1 | diff --git a/README_en.md b/README_en.md index 3116386737b60d693bde9627ba883a1c87b74a59..3aba08f0a1e1aa13d6fb69ef4730e9cea36ac8c5 100644 --- a/README_en.md +++ b/README_en.md @@ -43,12 +43,14 @@ individuals, healthcare, education, communication, energy, and more. | [Llama3-8B](nlp/llm/llama3_8b/pytorch) | PyTorch | Megatron-DeepSpeed | Bookcorpus | 4.1.1 | | [Llama3-8B](nlp/llm/llama3_8b/megatron-lm) | PyTorch | Megatron-LM | GPT Small-117M | 4.3.0 | | [Llama3-8B SFT](nlp/llm/llama3_8b_sft/pytorch) | PyTorch | ColossalAI | school_math_0.25M | 4.1.1 | +| [Llama3-8B SFT](nlp/llm/llama3_8b/openrlhf) | PyTorch | OpenRLHF | Meta-Llama-3-8B | 4.3.0 | | [Llama3-8B PPO](nlp/llm/llama3_8b/openrlhf) | PyTorch | OpenRLHF | Llama-3-8b-sft-mixture | 4.2.0 | | [Llama3-8B DPO](nlp/llm/llama3_8b/openrlhf) | PyTorch | OpenRLHF | Llama-3-8b-sft-mixture | 4.2.0 | | [Llama3-8B KTO](nlp/llm/llama3_8b/openrlhf) | PyTorch | OpenRLHF | Llama-3-8b-sft-mixture | 4.2.0 | | [Mamba-2](nlp/llm/mamba-2/pytorch) | PyTorch | Megatron-LM | GPT Small-117M | 4.1.1 | | [MiniCPM](nlp/llm/minicpm/pytorch) | PyTorch | DeepSpeed | MiniCPM-2B-sft-bf16 | 4.2.0 | | [Mixtral 8x7B](nlp/llm/mixtral/pytorch) | PyTorch | Megatron-LM | GPT Small-117M | 4.1.1 | +| [Mixtral 8x7B](nlp/llm/mixtral/openrlhf) | PyTorch | OpenRLHF | Mixtral-8x7B-v0.1 | 4.3.0 | | [Phi-3](nlp/llm/phi-3/pytorch) | PyTorch | Torchrun | Phi-3-mini-4k-instruct | 4.2.0 | | [QWen-7B](nlp/llm/qwen-7b/pytorch) | PyTorch | Firefly | qwen-7b | 3.4.0 | | [QWen1.5-7B](nlp/llm/qwen1.5-7b/pytorch) | PyTorch | Firefly | school_math | 4.1.1 | diff --git a/RELEASE.md b/RELEASE.md index 30575c0b666935783ae702b6a3aeec37fa81fc29..de9449591beac6accee4d1a72854b59020bee2b8 100644 --- a/RELEASE.md +++ b/RELEASE.md @@ -3,6 +3,56 @@ # DeepSparkHub Release Notes +## DeepSparkHub 25.09 Release Notes + +### 模型与算法 + +* 新增了10个大模型强化学习微调示例,使用了[verl](https://github.com/volcengine/verl)、[OpenRLHF](https://github.com/OpenRLHF/OpenRLHF)、[Megatron-LM](https://github.com/NVIDIA/Megatron-LM)、[Colossal-AI](https://github.com/hpcaitech/ColossalAI)、[deepspeed](https://github.com/deepspeedai/DeepSpeed)工具箱 + + + + + + + + + + + + + + + + + + + + + + + + +
大模型
Qwen2-7B GRPO (verl)Qwen2.5-VL-7B GRPO (verl)Qwen3-8B GRPO (verl)
DeepSeek-LLM-7B PPO (verl)Gemma-2-2B-IT PPO (verl)Llama-3-8B SFT (OpenRLHF)
Mixtral-8x7B-v0.1 SFT (OpenRLHF)Llama-3-8B (Megatron-LM)Qwen2.5-3B (Colossal-AI)
CosyVoice2-0.5B (deepspeed)
+ +### 修复更新 + +* 更新了AlexNet、GoogLeNet等模型所支持的最高SDK版本。 +* 同步了Colossal-AI、Megatron-Deepspeed、deepspeed框架的最新代码。 +* 同步了SSD模型的最新代码。 +* 修复了README文件中markdownlint的问题。 + +### 版本关联 + +DeepSparkHub 25.09对应天数软件栈4.3.0/4.4.0版本。 + +### 贡献者 + +感谢以下社区贡献者 + +sanghui_ilu,李一力,郝燕龙,胡方健,lsy789,张汉涛,majorli6,honglyua。 + +欢迎以任何形式为DeepSparkHub项目贡献。 + ## DeepSparkHub 25.06 Release Notes ### 模型与算法 diff --git a/tests/model_info.json b/tests/model_info.json index 12bb3d93905b5a0b53886273cda704ba317f7022..9f0ed785a2b1ee9559b582acee7a8b67c257be66 100644 --- a/tests/model_info.json +++ b/tests/model_info.json @@ -7535,6 +7535,54 @@ "github_branch": "", "github_path": "", "priority": "P4" + }, + { + "model_name": "llama3_8b", + "framework": "openrlhf", + "release_version": "25.09", + "release_sdk": "4.3.0", + "release_gpgpu": "BI-V150", + "latest_sdk": "", + "latest_gpgpu": "", + "category": "nlp/llm", + "toolbox": "", + "mdims": "", + "dataset": "", + "license": "", + "model_path": "deepsparkhub/nlp/llm/llama3_8b/openrlhf/", + "readme_file": "deepsparkhub/nlp/llm/llama3_8b/openrlhf/README.md", + "bitbucket_repo": "", + "bitbucket_branch": "", + "bitbucket_path": "", + "develop_owner": "", + "github_repo": "", + "github_branch": "", + "github_path": "", + "priority": "P4" + }, + { + "model_name": "mixtral", + "framework": "openrlhf", + "release_version": "25.09", + "release_sdk": "4.3.0", + "release_gpgpu": "BI-V150", + "latest_sdk": "", + "latest_gpgpu": "", + "category": "nlp/llm", + "toolbox": "", + "mdims": "", + "dataset": "", + "license": "", + "model_path": "deepsparkhub/nlp/llm/mixtral/openrlhf/", + "readme_file": "deepsparkhub/nlp/llm/mixtral/openrlhf/README.md", + "bitbucket_repo": "", + "bitbucket_branch": "", + "bitbucket_path": "", + "develop_owner": "", + "github_repo": "", + "github_branch": "", + "github_path": "", + "priority": "P4" } ] } \ No newline at end of file