Assume tied weights if lm_head/output weights is missing.

Copy the embd_tokens weights as GGUF has no direct support for tied weights.
This is to support model configurations with "tie_word_embeddings" set to true.

Author: dmahurin

convert.py

@@ -1467,6 +1467,10 @@ def main(args_in: list[str] | None = None) -> None:
 
     model   = model_plus.model
     model   = convert_model_names(model, params, args.skip_unknown)
+
+    if 'output.weight' not in model:
+        model['output.weight'] = model['token_embd.weight']
+
     ftype   = pick_output_type(model, args.outtype)
     model   = convert_to_output_type(model, ftype)
     outfile = args.outfile or default_outfile(model_plus.paths, ftype)