CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-01 09:01:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

gguf.py : fix for falcon 40b

Browse Source
This commit is contained in:
klosax
2023-08-19 01:01:52 +02:00
committed by GitHub
parent 281d6d1105
commit bd5a57901b
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
gguf.py
View File

@@ -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)