Qianfan-VL: A Milestone Achievement in Chinese Multimodal AI with Domestic Chips
•
9
AI & ML interests
None defined yet.
Recent Activity
Papers
Less Noise, More Voice: Reinforcement Learning for Reasoning via Instruction Purification
LoongFlow: Directed Evolutionary Search via a Cognitive Plan-Execute-Summarize Paradigm
Articles
Update tokenizer_config.json
#14 opened 3 days ago
by
WYF3634076
vLLM on Docker: model-imported Python code requires the install of module `decord` via pip
1
#8 opened 3 months ago
by
didier-durand
WYF3634076
updated
a
model
3 days ago
Post
3630
LoongFlow Big News!!!
@all
We’ve put AI Agents into a production GPU cluster to handle GPU failure prediction.
Not as a demo. Not as AutoML.
But as an evolving system that designs and improves its own models.
On two GPU types:
– IT21HMDB01-B2: +30% prediction accuracy
– H800: +25% prediction accuracy
The resulting models already meet production standards and are being wired into the ops pipeline.
How it works:
• An ML agent designs the full ML pipeline from scratch
• A Math agent performs targeted evolutionary optimization
• The agents explore, discard, and iterate toward better modelsHumans don’t hand-tune parameters.
This is not offline analysis. GPU failure prediction means:
• heavy assets
• real incidents
• real operational risk
The agents now trigger maintenance before failures happen.
This feels like an early signal: AI agents are starting to take responsibility for infrastructure-level engineering decisions in production systems.
For ML Agent, you can check: https://github.com/baidu-baige/LoongFlow
We’ve put AI Agents into a production GPU cluster to handle GPU failure prediction.
Not as a demo. Not as AutoML.
But as an evolving system that designs and improves its own models.
On two GPU types:
– IT21HMDB01-B2: +30% prediction accuracy
– H800: +25% prediction accuracy
The resulting models already meet production standards and are being wired into the ops pipeline.
How it works:
• An ML agent designs the full ML pipeline from scratch
• A Math agent performs targeted evolutionary optimization
• The agents explore, discard, and iterate toward better modelsHumans don’t hand-tune parameters.
This is not offline analysis. GPU failure prediction means:
• heavy assets
• real incidents
• real operational risk
The agents now trigger maintenance before failures happen.
This feels like an early signal: AI agents are starting to take responsibility for infrastructure-level engineering decisions in production systems.
For ML Agent, you can check: https://github.com/baidu-baige/LoongFlow
- 6 replies
XiaofengWu1028
authored
2
papers
13 days ago
Zeyu
authored
2
papers
13 days ago
XiaofengWu1028
authored
a
paper
13 days ago
ChengCui
authored
a
paper
13 days ago
Zeyu
authored
a
paper
13 days ago
max-zhenyu-zhang
authored
a
paper
15 days ago
DesmonDay
authored
a
paper
15 days ago
Post
1084
Hello everyone, We are thrilled to announce that LoongFlow has officially launched: General Agent!
This iteration introduces three major features, bringing the capabilities of intelligent agents to new heights.
1. Claude Agent SDK Deep Integration 📷
* Integrated with Claude Agent SDK: Enhancing the framework’s extensibility
* Seamless integration with the Claude Skills ecosystem, sharing powerful tool capabilities
2. Breakthrough Multi-File System Support 📷
* Say goodbye to single-file limitations: Supports complex system development at a full project level
3. Support for AI Self-Evaluation Mode 📷
* Self-evaluation: Agents can assess the quality of their own solutions, saving you the hassle of building evaluation functions
📷 For more details, check out: https://github.com/baidu-baige/LoongFlow/tree/main/agents/general_agent.
Feel free to try it out, and let us know if you have any questions or feedback! ~
This iteration introduces three major features, bringing the capabilities of intelligent agents to new heights.
1. Claude Agent SDK Deep Integration 📷
* Integrated with Claude Agent SDK: Enhancing the framework’s extensibility
* Seamless integration with the Claude Skills ecosystem, sharing powerful tool capabilities
2. Breakthrough Multi-File System Support 📷
* Say goodbye to single-file limitations: Supports complex system development at a full project level
3. Support for AI Self-Evaluation Mode 📷
* Self-evaluation: Agents can assess the quality of their own solutions, saving you the hassle of building evaluation functions
📷 For more details, check out: https://github.com/baidu-baige/LoongFlow/tree/main/agents/general_agent.
Feel free to try it out, and let us know if you have any questions or feedback! ~
ChengCui
authored
a
paper
17 days ago
ChengCui
submitted
a
paper
to Daily Papers
18 days ago
DaisyGrace
authored
a
paper
24 days ago
Post
1452
Really thanks to the community for their help! LoongFlow now has its first Chinese hands-on video!
Version 2.0 is under intensive development. The next version will support Skills, making it easier for users to create their own expert agents to solve various challenging and complex real-world problems.
Meanwhile, we are also exploring whether we can automatically generate high-quality expert Skills after a task is completed, reducing the difficulty of writing Skills and letting the LoongFlow framework automatically output the best Skills for challenging scenarios!
Version 2.0 is under intensive development. The next version will support Skills, making it easier for users to create their own expert agents to solve various challenging and complex real-world problems.
Meanwhile, we are also exploring whether we can automatically generate high-quality expert Skills after a task is completed, reducing the difficulty of writing Skills and letting the LoongFlow framework automatically output the best Skills for challenging scenarios!
Post
1042
I'm thinking about the next iteration direction for LoongFlow, and I'd appreciate any suggestions. Should we evolve the Claude Code Skills? Or the Agent structure? Or what should we evolve?
huggingface
-
- 3 replies