WIP test ci

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

@@ -2,7 +2,7 @@ name: Optimum TPU / Test TGI on TPU
 
 on:
   push:
-    branches: [ main ]
+    branches: [ quick-ci-test ]
     paths:
       - "text-generation-inference/**"
   pull_request:
@@ -30,14 +30,7 @@ jobs:
 
       - name: Build and test TGI server
         run: |
-          HF_TOKEN=${{ secrets.HF_TOKEN_OPTIMUM_TPU_CI }} make tgi_test
+          make test_installs tgi_server
+          find text-generation-inference/ -name "text_generation_server-*whl" -exec python -m pip install {} \;
+          HF_TOKEN=${{ secrets.HF_TOKEN_OPTIMUM_TPU_CI }} python -m pytest -sv text-generation-inference/tests -k gemma-2b
 
-      # 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 -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
-          JETSTREAM_PT=1 HF_TOKEN=${{ secrets.HF_TOKEN_OPTIMUM_TPU_CI }} python -m \
-            pytest -sv text-generation-inference/tests -k jetstream