Organize the model guide around use, inputs, and reproduction

README.md

@@ -73,20 +73,24 @@ configuration and training recipe:
 ```bash
 hf download risenyard/egms-qa-dataset --repo-type dataset \
     --local-dir release/egms-qa-dataset
-python -m egms_qa.release install \
+python -m egms_encoder.install_data \
     --release-dir release/egms-qa-dataset --target-root .
 hf download risenyard/egms-qa-encoder --include '*.json' \
-    --local-dir release/egms-qa-encoder
-python -m egms_qa.reproduce encoder \
-    --model-dir release/egms-qa-encoder --output-dir outputs/encoder_pretrain
+    --local-dir data/encoder/checkpoint
+python -m egms_encoder.pretrain \
+    --output-dir outputs/my_encoder --device cuda:0
 ```
 
-The runner starts training from scratch using `config.json`,
-`training_args.json`, and the train-fitted `normalization.json`. These files
-specify the architecture, masking, sampling, loss, optimization, and validation
-settings. Add `--dry-run` to inspect the resolved command. Checkpoint selection
-uses validation loss, and numerical reproduction depends on the software,
-hardware, and random state.
+Training reads `config.json`, `training_args.json`, and the train-fitted
+`normalization.json` from `data/encoder/checkpoint/`. These files specify the
+architecture, masking, sampling, loss, optimization, and validation settings.
+The output includes `best.safetensors` for inference, `latest.pt` for resuming,
+and the matching configuration and normalization files. The
+[Encoder guide](https://github.com/risenyard/egms-qa/blob/main/src/egms_encoder/README.md)
+provides resume and token-extraction commands for the trained model.
+
+Checkpoint selection uses validation loss. Numerical reproduction depends on
+the software, hardware, and random state.
 
 ## Evaluation