multimodal : add BakLLaVA conversion support (#3682)

Author: monatis

examples/llava/llava-surgery.py

@@ -16,13 +16,29 @@
 mm_tensors = [k for k, v in checkpoint.items() if k.startswith("model.mm_projector")]
 
 # store these tensors in a new dictionary and torch.save them
-projector = {name: checkpoint[name] for name in mm_tensors}
+projector = {name: checkpoint[name].float() for name in mm_tensors}
 torch.save(projector, f"{args.model}/llava.projector")
 
 # remove these tensors from the checkpoint and save it again
 for name in mm_tensors:
     del checkpoint[name]
 
+# BakLLaVA models contain CLIP tensors in it
+clip_tensors = [k for k, v in checkpoint.items() if k.startswith("model.vision_tower")]
+if len(clip_tensors) > 0:
+    clip = {name.replace("vision_tower.vision_tower.", ""): checkpoint[name].float() for name in clip_tensors}
+    torch.save(clip, f"{args.model}/llava.clip")
+
+    # remove these tensors
+    for name in clip_tensors:
+        del checkpoint[name]
+
+    # added tokens should be removed to be able to convert Mistral models
+    if os.path.exists(f"{args.model}/added_tokens.json"):
+        with open(f"{args.model}/added_tokens.json", "w") as f:
+            f.write("{}\n")
+
+
 torch.save(checkpoint, path)
 
 print("Done!")