Unverified50% confidenceReleaseExact time
Thinking Machines disclosed the training method for Inkling 2 Small: forward computation used 4-bit precision throughout pretraining, pretraining took only 68 hours, and the weights have been open-sourced.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/15/2026
First Seen
Valid until: 2/13/2027
Sources
Related Entities
Related Claims
UnverifiedMiniMind covers the complete pipeline including Tokenizer training, pre-training, supervised fine-tuning (SFT), LoRA fine-tuning, DPO reinforcement learning, and model distillation.71% similarUnverifiedThe ByteDance team said the training pipeline for this Agent foundation model was completed with only two stages of supervised fine-tuning plus one stage of reinforcement learning, and the training code and about 18,000 data entries were released together.71% similarUnverifiedThe currently public 4-bit training results have mostly been achieved on small- to medium-scale models or specific tasks, and there is still a gap before general large-model training.70% similarUnverifiedInkling 2 Small's training efficiency breakthrough is mainly attributed to improved data scheduling, parallel computing, and optimizer design.70% similarUnverifiedThe research team trained a batch of small models of about 1B parameters, some trained with tokens and some trained with bytes.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1171565API
curl https://kongchang.com/api/v1/knowledge/claims/1171565MCP
get_claim(id=1171565)