Using with tensor map

Author: zRzRzRzRzRzRzR

convert_hf_to_gguf.py

@@ -4900,27 +4900,6 @@ def set_gguf_parameters(self):
                 self.gguf_writer.add_rope_scaling_factor(self.hparams["rope_scaling"]["factor"])
                 self.gguf_writer.add_rope_scaling_orig_ctx_len(self.hparams["rope_scaling"]["original_max_position_embeddings"])
 
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[Tuple[str, Tensor]]:
-        if "gate_up_proj" in name:
-            match = re.match(r"model\.layers\.(\d+)\.gate_up_proj\.weight", name)
-            if match:
-                bid = int(match.group(1))
-                return [(f"blk.{bid}.ffn_up.weight", data_torch)]
-
-        if "post_self_attn_layernorm" in name:
-            match = re.match(r"model\.layers\.(\d+)\.post_self_attn_layernorm\.weight", name)
-            if match:
-                bid = int(match.group(1))
-                return [(f"blk.{bid}.post_attn_norm.weight", data_torch)]
-
-        if "post_mlp_layernorm" in name:
-            match = re.match(r"model\.layers\.(\d+)\.post_mlp_layernorm\.weight", name)
-            if match:
-                bid = int(match.group(1))
-                return [(f"blk.{bid}.post_mlp_norm.weight", data_torch)]
-
-        return super().modify_tensors(data_torch, name, bid)
-
 @Model.register("GlmForCausalLM", "ChatGLMModel", "ChatGLMForConditionalGeneration")
 class ChatGLMModel(Model):
     model_arch = gguf.MODEL_ARCH.CHATGLM

gguf-py/gguf/constants.py

@@ -1577,6 +1577,8 @@ class MODEL_TENSOR(IntEnum):
         MODEL_TENSOR.FFN_NORM,
         MODEL_TENSOR.FFN_DOWN,
         MODEL_TENSOR.FFN_UP,
+        MODEL_TENSOR.ATTN_POST_NORM,
+        MODEL_TENSOR.FFN_POST_NORM,
     ],
     MODEL_ARCH.BITNET: [
         MODEL_TENSOR.ATTN_Q,

gguf-py/gguf/tensor_mapping.py

@@ -233,7 +233,8 @@ class TensorNameMap:
         ),
 
         MODEL_TENSOR.ATTN_POST_NORM: (
-            "model.layers.{bid}.post_attention_layernorm",     # gemma2 olmo2
+            "model.layers.{bid}.post_attention_layernorm",     # gemma2 olmo2    # ge
+            "model.layers.{bid}.post_self_attn_layernorm",     # glm-4-0414
         ),
 
         # Rotary embeddings
@@ -269,6 +270,7 @@ class TensorNameMap:
         # Post feed-forward norm
         MODEL_TENSOR.FFN_POST_NORM: (
             "model.layers.{bid}.post_feedforward_layernorm", # gemma2 olmo2
+            "model.layers.{bid}.post_mlp_layernorm", # glm-4-0414
         ),
 
         MODEL_TENSOR.FFN_GATE_INP: (