Update app.py

app.py

@@ -23,6 +23,7 @@ MODELS = {
         "clip_prefix":  "mmproj-SmolVLM2-256M-Video-Instruct",
         "model_variants": ["f16", "Q8_0", "Q2_K", "Q4_K_M"],
         "clip_variants":  ["Q8_0", "f16"],
+        "model_separator": ".",  # Dot for SmolVLM model files
     },
     "450M": {
         "model_repo": "LiquidAI/LFM2-VL-450M-GGUF",
@@ -31,6 +32,7 @@ MODELS = {
         "clip_prefix": "mmproj-LFM2-VL-450M",
         "model_variants": ["F16", "Q4_0", "Q8_0"],
         "clip_variants": ["F16", "Q8_0"],
+        "model_separator": "-",  # Dash for LiquidAI model files
     },
     "500M": {
         "model_repo": "mradermacher/SmolVLM2-500M-Video-Instruct-GGUF",
@@ -39,19 +41,21 @@ MODELS = {
         "clip_prefix":  "mmproj-SmolVLM2-500M-Video-Instruct",
         "model_variants": ["f16", "Q4_K_M", "Q8_0", "Q2_K"],
         "clip_variants":  ["Q8_0", "f16"],
+        "model_separator": ".",  # Dot for SmolVLM model files
     },
     "1B": {
         "model_repo": "bartowski/OpenGVLab_InternVL3_5-1B-GGUF",
-        "clip_repo":  "bartowski/OpenGVLab_InternVL3_5-1B-GGUF",  # Same repo for CLIP files
-        "model_prefix": "InternVL3_5-1B",
-        "clip_prefix":  "mmproj-OpenGVLab_InternVL3_5-1B",  # Matches the CLIP file naming
+        "clip_repo":  "bartowski/OpenGVLab_InternVL3_5-1B-GGUF",
+        "model_prefix": "OpenGVLab_InternVL3_5-1B",  # Corrected prefix
+        "clip_prefix":  "mmproj-OpenGVLab_InternVL3_5-1B",
         "model_variants": [
             "bf16", "Q8_0", "Q6_K_L", "Q5_K_L", "Q6_K", "Q4_K_L", 
             "Q3_K_XL", "Q5_K_M", "Q5_K_S", "Q4_1", "Q2_K_L", "Q4_K_M", 
             "Q4_K_S", "Q4_0", "IQ4_NL", "IQ4_XS", "Q3_K_L", "Q3_K_M", 
             "IQ3_M", "Q3_K_S", "IQ3_XS", "IQ3_XXS", "Q2_K", "IQ2_M"
         ],
-        "clip_variants":  ["bf16", "f16"],  # Based on the available CLIP files
+        "clip_variants":  ["bf16", "f16"],
+        "model_separator": "-",  # Dash for InternVL3 model files
     },
     "2.2B": {
         "model_repo": "mradermacher/SmolVLM2-2.2B-Instruct-GGUF",
@@ -60,6 +64,7 @@ MODELS = {
         "clip_prefix":  "mmproj-SmolVLM2-2.2B-Instruct",
         "model_variants": ["f16", "Q4_K_M", "Q8_0", "Q2_K"],
         "clip_variants":  ["Q8_0", "f16"],
+        "model_separator": ".",  # Dot for SmolVLM model files
     },
 }
 
@@ -125,8 +130,11 @@ def update_llm(size, model_file, clip_file, verbose_mode, n_threads):
 # Build weight filename lists
 def get_weight_files(size):
     cfg = MODELS[size]
-    model_files = [f"{cfg['model_prefix']}.{v}.gguf" for v in cfg['model_variants']]
-    clip_files  = [f"{cfg['clip_prefix']}-{v}.gguf"  for v in cfg['clip_variants']]
+    # Use model_separator for model files (default to '.' if not specified)
+    model_sep = cfg.get("model_separator", ".")
+    model_files = [f"{cfg['model_prefix']}{model_sep}{v}.gguf" for v in cfg['model_variants']]
+    # CLIP files always use dash separator
+    clip_files = [f"{cfg['clip_prefix']}-{v}.gguf" for v in cfg['clip_variants']]
     return model_files, clip_files
 
 # Caption using cached llm with real-time debug logs