[fix] fix expert_bias convert hf to gguf for ling mini 2.0

im0qianqian · im0qianqian · commit 94ec7dc841f2 · 2025-09-14T01:42:55.000+08:00
diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
@@ -7917,7 +7917,7 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
             'attention.dense': 'self_attn.dense',
             'attention.key_layernorm': 'self_attn.key_layernorm',
             'attention.query_layernorm': 'self_attn.query_layernorm',
-            'mlp.gate.expert_bias': 'mlp.gate.e_score_correction',
+            'mlp.gate.expert_bias': 'mlp.gate.e_score_correction.bias',
         }
         for k, v in pre_tensor_name_mapping.items():
             name = name.replace(k, v)

Original file line number	Diff line number	Diff line change
`@@ -7917,7 +7917,7 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int \| None) -> Iter`
`7917`	`7917`	`'attention.dense': 'self_attn.dense',`
`7918`	`7918`	`'attention.key_layernorm': 'self_attn.key_layernorm',`
`7919`	`7919`	`'attention.query_layernorm': 'self_attn.query_layernorm',`
`7920`		`- 'mlp.gate.expert_bias': 'mlp.gate.e_score_correction',`
	`7920`	`+ 'mlp.gate.expert_bias': 'mlp.gate.e_score_correction.bias',`
`7921`	`7921`	`}`
`7922`	`7922`	`for k, v in pre_tensor_name_mapping.items():`
`7923`	`7923`	`name = name.replace(k, v)`