ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 17:20:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix convert error for ernie 4.5 (#774)

Browse Source
This commit is contained in:
firecoperana
2025-09-11 00:59:24 -05:00
committed by GitHub
parent 8403308d8e
commit de97c33b40
1 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
convert_hf_to_gguf.py
View File

@@ -320,7 +320,7 @@ class Model:
gguf.LlamaFileType.MOSTLY_Q4_0,
gguf.LlamaFileType.MOSTLY_Q4_1,
):
data_qtype = gguf.GGMLQuantizationType.Q5_0
data_qtype = gguf.GGMLQuantizationType.Q5_0
elif self.ftype in (
gguf.LlamaFileType.MOSTLY_Q5_0,
gguf.LlamaFileType.MOSTLY_Q5_1,
@@ -418,7 +418,7 @@ class Model:
logger.info("Set model quantization version")
self.gguf_writer.add_quantization_version(gguf.GGML_QUANT_VERSION)
logger.info("***********************************************************************************************")
logger.info("** Converting to `q4_0`,`q4_1`,`q5_0`, `q5_1` or `q6_0` is not equiv to using `llama-quantize`!")
logger.info("** Ftype `q4_0`,`q4_1` are here converting embeddings, output, attn_k and attn_v/qkv in q5_0.")
@@ -2195,8 +2195,8 @@ class Qwen3MoeModel(Qwen2MoeModel):
model_arch = gguf.MODEL_ARCH.QWEN3MOE
@ModelBase.register("Ernie4_5_ForCausalLM", "Ernie4_5ForCausalLM")
class Ernie4_5Model(TextModel):
@Model.register("Ernie4_5_ForCausalLM", "Ernie4_5ForCausalLM")
class Ernie4_5Model(Model):
model_arch = gguf.MODEL_ARCH.ERNIE4_5
def set_vocab(self):
@@ -2242,7 +2242,7 @@ class Ernie4_5Model(TextModel):
return [(self.map_tensor_name(name), data_torch)]
@ModelBase.register("Ernie4_5_MoeForCausalLM")
@Model.register("Ernie4_5_MoeForCausalLM")
class Ernie4_5MoeModel(Ernie4_5Model):
model_arch = gguf.MODEL_ARCH.ERNIE4_5_MOE
_experts: list[dict[str, Tensor]] | None = None