add native hf auth check

Signed-off-by: Vladimir Mandic <mandic00@live.com>

modules/model_hidream.py

@@ -21,6 +21,13 @@ def load_hidream(checkpoint_info, diffusers_load_config={}):
     login = modelloader.hf_login()
     repo_id = sd_models.path_to_repo(checkpoint_info.name)
 
+    from huggingface_hub import auth_check
+    try:
+        auth_check(shared.opts.model_h1_llama_repo)
+    except Exception as e:
+        shared.log.error(f'Load model: type=HiDream te4="{shared.opts.model_h1_llama_repo}" login={login} {e}')
+        return False
+
     load_args, quant_args = model_quant.get_dit_args(diffusers_load_config, module='Transformer', device_map=True)
     shared.log.debug(f'Load model: type=HiDream transformer="{repo_id}" quant="{model_quant.get_quant_type(quant_args)}" args={load_args}')
     transformer = diffusers.HiDreamImageTransformer2DModel.from_pretrained(
@@ -47,25 +54,22 @@ def load_hidream(checkpoint_info, diffusers_load_config={}):
 
     load_args, quant_args = model_quant.get_dit_args(diffusers_load_config, module='LLM', device_map=True)
     shared.log.debug(f'Load model: type=HiDream te4="{shared.opts.model_h1_llama_repo}" quant="{model_quant.get_quant_type(quant_args)}" args={load_args}')
-    try:
-        text_encoder_4 = transformers.LlamaForCausalLM.from_pretrained(
-            shared.opts.model_h1_llama_repo,
-            output_hidden_states=True,
-            output_attentions=True,
-            cache_dir=shared.opts.hfcache_dir,
-            **load_args,
-            **quant_args,
-        )
-        tokenizer_4 = transformers.PreTrainedTokenizerFast.from_pretrained(
-            shared.opts.model_h1_llama_repo,
-            cache_dir=shared.opts.hfcache_dir,
-            **load_args,
-        )
-        if shared.opts.diffusers_offload_mode != 'none':
-            text_encoder_4 = text_encoder_4.to(devices.cpu)
-    except Exception as e:
-        shared.log.error(f'Load model: type=HiDream te4="{shared.opts.model_h1_llama_repo}" {e}')
-        shared.log.warning(f'Load model: type=HiDream te4="{shared.opts.model_h1_llama_repo}" login={login} verify access to gated model')
+
+    text_encoder_4 = transformers.LlamaForCausalLM.from_pretrained(
+        shared.opts.model_h1_llama_repo,
+        output_hidden_states=True,
+        output_attentions=True,
+        cache_dir=shared.opts.hfcache_dir,
+        **load_args,
+        **quant_args,
+    )
+    tokenizer_4 = transformers.PreTrainedTokenizerFast.from_pretrained(
+        shared.opts.model_h1_llama_repo,
+        cache_dir=shared.opts.hfcache_dir,
+        **load_args,
+    )
+    if shared.opts.diffusers_offload_mode != 'none':
+        text_encoder_4 = text_encoder_4.to(devices.cpu)
 
     load_args, _quant_args = model_quant.get_dit_args(diffusers_load_config, module='Model')
     shared.log.debug(f'Load model: type=HiDream model="{checkpoint_info.name}" repo="{repo_id}" offload={shared.opts.diffusers_offload_mode} dtype={devices.dtype} args={load_args}')

modules/sd_models.py

@@ -549,6 +549,9 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
         # load with custom loader
         if sd_model is None:
             sd_model = load_diffuser_force(model_type, checkpoint_info, diffusers_load_config, op)
+            if sd_model is not None and not sd_model:
+                shared.log.error(f'Load {op}: type="{model_type}" pipeline="{pipeline}" not loaded')
+                return
 
         # load from hf folder-style
         if sd_model is None: