complexly/olmo3-190m-zh-sft

SFT(有监督微调)版本:基于complexly/olmo3-190m-zh-continue, 使用对话格式数据进行微调,学习指令遵循能力。

数据来源

  • 训练数据:cmz1024/llm101-olmo3-zh-demo-data
  • 子路径:sft/sft_t2t_mini.jsonl

训练配置

  • Learning Rate:5.0e-5
  • Warmup:5%
  • Epochs:3 epoch
  • Max Seq Length:2048
  • 使用 assistant_only_loss(仅对 assistant 部分计算 loss)
  • per_device_train_batch_size: 24
  • packing: true

用法

from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers import pipeline

model = AutoModelForCausalLM.from_pretrained("{target_repo}")
tok = AutoTokenizer.from_pretrained("{target_repo}")

# 使用 chat template
messages = [{{"role": "user", "content": "你好,请介绍一下北京"}}]
inputs = tok.apply_chat_template(messages, return_tensors="pt")
outputs = model.generate(inputs, max_new_tokens=200)
print(tok.decode(outputs[0]))
Downloads last month
28
Safetensors
Model size
0.2B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for complexly/olmo3-190m-zh-sft

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

Space using complexly/olmo3-190m-zh-sft 1