recast loralayer, norm, lmhead + embed token weights per original qlora (#393)

* recast loralayer, norm, lmhead + embed token weights per original qlora

* try again for the fix

* refactor torch dtype picking

* linter fixes

* missing import for LoraLayer

* fix install for tests now that peft is involved

.github/workflows/tests.yml

@@ -24,7 +24,7 @@ jobs:
 
       - name: Install dependencies
         run: |
-          pip install -e .
+          pip install -e .[peft]
           pip install -r requirements-tests.txt
 
       - name: Run tests