Update

app.py

@@ -161,7 +161,7 @@ def create_app():
     gr.on(
         [run.click, prompt.submit],
         compute,
-        [image, prompt, model + "-text-model-q4_k_m.gguf", sampler],
+        [image, prompt, model, sampler],
         [highlighted_text, image, annotated_image],
     )
     clear.click(
@@ -244,8 +244,7 @@ if __name__ == '__main__':
     logging.info('environ["%s"] = %r', k, v)
 
   gradio_helpers.set_warmup_function(warmup)
-  for name, (repo, filename) in models.MODELS.items():
-    gradio_helpers.register_download(name + "-text-model-q4_k_m.gguf", repo, filename + "-text-model-q4_k_m.gguf")
-    gradio_helpers.register_download(name + "-mmproj-f16.gguf", repo, filename + "-mmproj-f16.gguf")
+  for name, (repo, filenames) in models.MODELS.items():
+    gradio_helpers.register_download(name, repo, filenames)
 
   create_app().queue().launch()

gradio_helpers.py

@@ -74,7 +74,7 @@ def _do_download():
       time.sleep(1)
       continue
 
-    name, (repo, filename, revision) = next(iter(_scheduled.items()))
+    name, (repo, filenames, revision) = next(iter(_scheduled.items()))
     logging.info('Downloading "%s" %s/%s/%s...', name, repo, filename, revision)
     with timed(f'downloading {name}', True) as t:
       if should_mock():
@@ -83,8 +83,7 @@ def _do_download():
         _done[name] = None
       else:
         try:
-          _done[name] = huggingface_hub.hf_hub_download(
-              repo_id=repo, filename=filename, revision=revision)
+          _done[name] = (huggingface_hub.hf_hub_download(repo_id=repo, filename=filename, revision=revision) for filename in filenames)
         except Exception as e:  # pylint: disable=broad-exception-caught
           logging.exception('Could not download "%s" from hub!', name)
           _failed[name] = str(e)
@@ -109,11 +108,11 @@ def _do_download():
       _scheduled.pop(name)
 
 
-def register_download(name, repo, filename, revision='main'):
+def register_download(name, repo, filenames, revision='main'):
   """Will cause download of `filename` from HF `repo` in background thread."""
   with _lock:
     if name not in _scheduled:
-      _scheduled[name] = (repo, filename, revision)
+      _scheduled[name] = (repo, filenames, revision)
 
 
 def _hms(secs):

models.py

@@ -20,7 +20,7 @@ MODELS = {
     **{
         model_name: (
             f'{ORGANIZATION}/{repo}',
-            f'{model_name}',
+            (f'{model_name}-text-model-q4_k_m.gguf', f'{model_name}-mmproj-f16.gguf'),
         )
         for repo, model_name in BASE_MODELS
     },
@@ -78,8 +78,7 @@ def generate(
   # with gradio_helpers.timed('computation', start_message=True):
   #   tokens = model.predict(params, batch, sampler=sampler)
 
-  model_path = gradio_helpers.get_paths()[model_name + "-text-model-q4_k_m.gguf"]
-  clip_path = gradio_helpers.get_paths()[model_name + "-mmproj-f16.gguf"]
+  model_path, clip_path = gradio_helpers.get_paths()[model_name]
   print(model_path)
   print(gradio_helpers.get_paths())
   model = llama_cpp.Llama(