Unverified50% confidenceFactExact time
CRN v2 has approximately 34 million trainable parameters, equivalent to 0.73% of the 4.65 billion-parameter text module.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe Qwen3 series training data reportedly exceeds 36 trillion tokens, covering code, mathematics, science, and multilingual text.61% similarUnverifiedLoRA使可训练参数量通常仅为全参微调的0.1%-1%，可在单张RTX 3090/4090（显存24GB）上微调7B乃至13B级别的模型60% similarUnverifiedThe vocabulary size obtained by the model after training the tokenizer was 3,266.59% similarUnverified作者推算Mythos Preview训练消耗算力约为5e26-1e27 FLOPs，是GPT-4训练算力(约2×10^25 FLOPs)的25到50倍59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924659API
curl https://kongchang.com/api/v1/knowledge/claims/924659MCP
get_claim(id=924659)