From: Georgi Gerganov <redacted>
Date: Fri, 6 Dec 2024 19:33:15 +0000 (+0200)
Subject: convert : add custom attention mapping
X-Git-Tag: upstream/0.0.4488~210
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=c5ede3849fc021174862f9c0bf8273808d8f0d39;p=pkg%2Fggml%2Fsources%2Fllama.cpp

convert : add custom attention mapping
---

diff --git a/gguf-py/gguf/tensor_mapping.py b/gguf-py/gguf/tensor_mapping.py
index 1b6a3f4a..f0a7b647 100644
--- a/gguf-py/gguf/tensor_mapping.py
+++ b/gguf-py/gguf/tensor_mapping.py
@@ -146,6 +146,7 @@ class TensorNameMap:
         # Attention query
         MODEL_TENSOR.ATTN_Q: (
             "model.layers.{bid}.self_attn.q_proj",                       # llama-hf nemotron olmoe olmo2
+            "model.layers.{bid}.self_attn.q_proj_no_perm",               # llama-custom
             "layers.{bid}.attention.wq",                                 # llama-pth
             "encoder.layer.{bid}.attention.self.query",                  # bert
             "transformer.h.{bid}.attn.q_proj",                           # gpt-j
@@ -158,6 +159,7 @@ class TensorNameMap:
         # Attention key
         MODEL_TENSOR.ATTN_K: (
             "model.layers.{bid}.self_attn.k_proj",                     # llama-hf nemotron olmoe olmo2
+            "model.layers.{bid}.self_attn.k_proj_no_perm",             # llama-custom
             "layers.{bid}.attention.wk",                               # llama-pth
             "encoder.layer.{bid}.attention.self.key",                  # bert
             "transformer.h.{bid}.attn.k_proj",                         # gpt-j