mikuhhn1239

qwen3-8b-scene-segmentation-lora

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more
Container

Run this model inference with full control and performance in your environment.

Learn more

Get help setting up a custom Dedicated Endpoints.

Talk with our engineer to get a quote for reserved GPU instances with discounts.

README

License: apache-2.0

任务

  • 输入: 编号段落 [P1]...[P2]...
  • 输出: {"boundaries": [N]} — N 为切分位置(在段落 N 之后切)

示例

markdown

输入:
[P1] 下课铃响,教室里热闹起来。
[P2] 她低头收拾书包。
[P3] 我犹豫了一下,还是叫住了她。
[P4] 十分钟后,我们并肩走在校门外的街道上。
输出:
{"boundaries": [3]} ← P3后切scene(教室→校门外)

加载

python

from transformers import AutoModelForCausalLM
from peft import PeftModel
base = AutoModelForCausalLM.from_pretrained(
"mikuhhn1239/qwen3-8b-novel-base-sft",
torch_dtype="auto", device_map="auto",
)
model = PeftModel.from_pretrained(
base, "mikuhhn1239/qwen3-8b-scene-boundary-lora"
)

训练

markdown

基座: Qwen3-8B-Novel-Base-SFT (Stage1 全参 SFT, 72K 小说续写数据)
方法: LoRA (r=64, α=128, dropout=0.05)
数据: 350 条 (280 train / 35 val / 35 test),Clean 标注
框架: transformers Trainer + PEFT
优化器: AdamW (adamw_torch_fused), cosine schedule, warmup=5%
epoch: 5 | LR: 1e-4 | batch: 1×16(accum) | bf16 | max_length: 4096

版本历史

Table
版本数据量epochsLRJSON解析关键指标
零基座 (Qwen3-8B)2.9%F1 0%
+Stage1 (全参SFT)0%F1 0%
v18232e-466.7%F1 33.3%
v2~15032e-488.9%F1 53.3%
v3 (LoRA)28032e-497.1%F1 19.6%
v3.1 (LoRA)28051e-4100%F1 28.6%

结论

  • 零基座 / +Stage1 全 0%:不做 Agent SFT 就不会场景边界检测 ✅
  • scene-boundary: 3→5 epochs 后 F1 从 19.6% → 28.6%(+9pp
  • v2 F1 53.3% 但任务定义不同(含原因输出),不可直接对比

Model provider

mikuhhn1239

Model tree

Base

mikuhhn1239/qwen3-8b-novel-base-sft

Adapter

this model

Modalities

Input

Text

Output

Text

Pricing

Dedicated Endpoints

View details

Supported Functionality

Model APIs

Dedicated Endpoints

Container

More information

Explore FriendliAI today