feat: add OLMoE grokking region based on arxiv 2506.21551 Figure 1

feat: add OLMoE-1B-7B-0924 training dynamics tab

fix 动态检测：该step是否有分片文件

fix. remove debug

debug data same issue. 2.8 using sharded.

debug data same issue

debug data same issue

debug all layers data are same issue

debug for 2.8b get the same data issue.

fix: rename Q_ssr to ssr in pythia_scanner.py, tab_pythia.py, and migrate existing CSVs

fix: support sharded safetensors and F16 dtype for Pythia 1.4b/2.8b

按架构写JSON，补充1B、1.4B、2.8B

feat: add Pythia training dynamics tab

改动汇总

Models analyzed after app startup did not appear in the Tab 6 model selector.

Two independent changes in one commit:

The condition number κ has extreme dynamic range across layers:

Three UX issues reported on Tab 5 Interactive view:

Add `core/plotter.py` and `ui/tab_plot.py` to generate Nature/PRL-standard

fix copy 2 times error

feat: add global debug switch and unified debug logging system

debug for gemma-4-31b-it

debug gemma-4-31b-it 2 code ,2 result issue

add debug info for gemma-4-31b-it

debug for gemma-4-31b-it layer 0, KV head 0

def build_tab_analyze():

feat: restructure into modular architecture with auto-inference engine