Steel-LLM is a fully open 1B-parameter Chinese-centric LLM trained from scratch on ~1T tokens with 8 GPUs, reaching 41.9 C-Eval and 36.1 CMMLU after SFT and DPO.
Infinity instruct
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Steel-LLM:From Scratch to Open Source -- A Personal Journey in Building a Chinese-Centric LLM
Steel-LLM is a fully open 1B-parameter Chinese-centric LLM trained from scratch on ~1T tokens with 8 GPUs, reaching 41.9 C-Eval and 36.1 CMMLU after SFT and DPO.