woohello/olmo3-190m-zh-nano-continue

持续预训练版本:基于 woohello/olmo3-190m-zh-nano 继续训练, 学习 Wikipedia-zh 等新语料,在保持原有知识基础上扩展能力。

训练配置

  • Base model: woohello/olmo3-190m-zh-nano (26M, OLMo3 arch, SDPA)
  • 数据:42ailab/llm101-v3.1-data/tokenized/full_v31.bin (Wikipedia-zh 继续)
  • LR: 2e-4(比 pretrain 1e-3 低 5x,防止灾难性遗忘)
  • Warmup: **10%**(比 pretrain 2% 长 5x,平滑过渡)
  • 训练: RTX 3090 (24GB), bf16, attn_implementation=sdpa

用法

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("woohello/olmo3-190m-zh-nano-continue", attn_implementation="sdpa")
tok = AutoTokenizer.from_pretrained("woohello/olmo3-190m-zh-nano-continue")

input_ids = tok("从前有座山,山里有座庙,", return_tensors="pt").input_ids
out = model.generate(input_ids, max_new_tokens=100, do_sample=True, temperature=0.8)
print(tok.decode(out[0], skip_special_tokens=True))
Downloads last month
8
Safetensors
Model size
22M params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for woohello/olmo3-190m-zh-nano-continue

Unable to build the model tree, the base model loops to the model itself. Learn more.

Space using woohello/olmo3-190m-zh-nano-continue 1