lm_head.weight is missing
#3
by
DeepMount00
- opened
the final lm_head.weight is missing, why?
They've seemingly uploaded the underlying Qwen2Model instead of the Qwen2ForCausalLM.
yeah
This model sets tie_word_embeddings=True, which shares the weight between "model.embed_tokens" and "lm_head". Therefore, use the transpose of "model.embed_tokens.weight" as "lm_head.weight" directly.