From bd5a57901bc9699f929ff89e70b9dc885c80fcaa Mon Sep 17 00:00:00 2001 From: klosax <131523366+klosax@users.noreply.github.com> Date: Sat, 19 Aug 2023 01:01:52 +0200 Subject: [PATCH] gguf.py : fix for falcon 40b --- gguf.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/gguf.py b/gguf.py index c4290d64a..e5eb85ded 100644 --- a/gguf.py +++ b/gguf.py @@ -220,7 +220,7 @@ def get_tensor_name_map(arch: MODEL_ARCH, n_blocks: int) -> dict: tensor_map["transformer.h."+str(i)+".ln_1"] = mapped_to # gpt2 tensor_map["transformer.blocks."+str(i)+".norm_1"] = mapped_to # mpt tensor_map["transformer.h."+str(i)+".input_layernorm"] = mapped_to # falcon7b - tensor_map["transformer.h."+str(i)+".ln_attn"] = mapped_to # falcon40b + tensor_map["transformer.h."+str(i)+".ln_mlp"] = mapped_to # falcon40b tensor_map["model.layers."+str(i)+".input_layernorm"] = mapped_to # llama-hf tensor_map["layers."+str(i)+".attention_norm"] = mapped_to # llama-pth @@ -228,7 +228,7 @@ def get_tensor_name_map(arch: MODEL_ARCH, n_blocks: int) -> dict: mapped_to = MODEL_TENSOR_NAMES[arch].get(MODEL_TENSOR.ATTN_NORM_2, None) mapped_to = mapped_to.format(bid=i) if mapped_to is not None else None - tensor_map["transformer.h."+str(i)+".ln_mlp"] = mapped_to # falcon40b + tensor_map["transformer.h."+str(i)+".ln_attn"] = mapped_to # falcon40b # Attention query-key-value mapped_to = MODEL_TENSOR_NAMES[arch].get(MODEL_TENSOR.ATTN_QKV, None)