feat(sense): T-G4 线性头(离线 softmax 回归 + 纯 Python 推理)
- classifier.py:LinearHead.load(JSON 工件,缺失/损坏 -> ArtifactMissing)+ predict(纯 Python 点积 + 稳定 softmax,D-G5 serving 零依赖);LoraRemote 预留 - scripts/train_tier_head.py:numpy softmax 回归(类别权重均衡/L2), 70/15/15 切分,head.json + metrics.json(accuracy/macro-F1/逐档 F1)+ 工件登记; --synthetic 合成自测链路;CLI 出口 UTF-8 reconfigure - requirements-ml.txt 登记 numpy(离线训练用,D-P6/D-G5 说明理由) - 测试 +5(黄金向量 softmax 一致/argmax/缺失/损坏/合成训练-推理链路),全量 400 passed
This commit is contained in:
+15
-11
@@ -1,11 +1,15 @@
|
||||
# 可选:接入真实开源小模型(RTX 4060 8GB 可跑 0.5B-4B 量化模型)
|
||||
# 安装:pip install -r requirements-ml.txt
|
||||
torch>=2.2
|
||||
transformers>=4.40
|
||||
accelerate>=0.30
|
||||
peft>=0.11
|
||||
sentencepiece
|
||||
protobuf
|
||||
bitsandbytes
|
||||
# 轻量推理引擎(可选)
|
||||
# llama-cpp-python␍
|
||||
# 可选:接入真实开源小模型(RTX 4060 8GB 可跑 0.5B-4B 量化模型)
|
||||
# 安装:pip install -r requirements-ml.txt
|
||||
torch>=2.2
|
||||
transformers>=4.40
|
||||
accelerate>=0.30
|
||||
peft>=0.11
|
||||
sentencepiece
|
||||
protobuf
|
||||
bitsandbytes
|
||||
# 轻量推理引擎(可选)
|
||||
# llama-cpp-python
|
||||
|
||||
# 语义分析器离线训练(T-G4):仅 scripts/train_tier_head.py 使用;
|
||||
# serving 路径零新依赖(D-G5:线性头推理纯 Python 点积)
|
||||
numpy>=1.26
|
||||
|
||||
Reference in New Issue
Block a user