Alibaba releases IndustryLLM for industrial procurement with 35B parameters
Researchers at Alibaba have introduced IndustryLLM, an open-weight language model designed for industrial procurement tasks. Built from Qwen3.5-35B-A3B-Base, the model uses 35 billion parameters with 3 billion activated per token and a frozen vision encoder. Its training focuses on failure-driven adaptation, combining continued pre-training (CPT) on a 100 billion-token corpus—including 5 billion…
Key points
- IndustryLLM is a 35B-parameter model trained on 100B tokens for industrial procurement, with 3B activated per token
- Offline tests show 2.97% improvement in query structuring; live A/B tests report 4.25% higher GMV and 8.3% more satisfied inquiries
- Latency reduced from 6–7 seconds to 1.5 seconds, with open weights released on Hugging Face
The model achieves 2.97 percentage points improvement in procurement-query structuring (95% CI: [2.11, 3.86]) in offline tests and 4.25% higher GMV and 8.3% more satisfied inquiries in live A/B experiments. Latency dropped from 6–7 seconds to 1.5 seconds. The weights and configurations are available on Hugging Face.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Anthropic's Claude begins 'dreaming' to reorganize memory in tests · 1 src
- John Langford quits Microsoft, declares AI Frontiers over · 1 src
- Anthropic releases study on Claude usage in scientific research · 3 src
- MIT and BASF release CheMeleon for low-data molecular prediction · 2 src
- Anthropic tests Claude agents in EVE Online trading and sabotage detection · 1 src
Comments
via GitHub Discussions