WIP

.github/workflows/test-pytorch-xla-tpu-tgi.yml

@@ -29,13 +29,18 @@ jobs:
         uses: actions/checkout@v4
 
       - name: Build and test TGI server
+        # run: |
+        #   HF_TOKEN=${{ secrets.HF_TOKEN_OPTIMUM_TPU_CI }} make tgi_test
         run: |
-          HF_TOKEN=${{ secrets.HF_TOKEN_OPTIMUM_TPU_CI }} make tgi_test
+          make test_installs tgi_server
+          find text-generation-inference/ -name "text_generation_server-*whl" -exec python -m pip install {} \;
+
 
       # Use a different step to test the Jetstream Pytorch version, to avoid conflicts with torch-xla[tpu]
       - name: Install and test TGI server (Jetstream Pytorch)
         run: |
-          pip install -e .[jetstream-pt] \
+          pip show text-generation-server
+          pip install -U .[jetstream-pt] \
             -f https://storage.googleapis.com/jax-releases/jax_nightly_releases.html \
             -f https://storage.googleapis.com/jax-releases/jaxlib_nightly_releases.html \
             -f https://storage.googleapis.com/libtpu-releases/index.html