CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-09 10:17:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into add-fh1-rebased

Browse Source
This commit is contained in:
Younes B
2025-07-08 10:37:13 +02:00
committed by GitHub
parent 58e3866d02 8f22dc0a53
commit d28c31a90c
63 changed files with 2874 additions and 670 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
gguf-py/gguf/constants.py
View File

@@ -359,6 +359,7 @@ class MODEL_ARCH(IntEnum):
DOTS1 = auto()
ARCEE = auto()
ERNIE4_5 = auto()
HUNYUAN_MOE = auto()
class VISION_PROJECTOR_TYPE(IntEnum):
@@ -663,6 +664,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.ARCEE: "arcee",
MODEL_ARCH.ERNIE4_5: "ernie4_5",
MODEL_ARCH.FALCON_H1: "falcon_h1",
MODEL_ARCH.HUNYUAN_MOE: "hunyuan-moe",
}
VISION_PROJECTOR_TYPE_NAMES: dict[VISION_PROJECTOR_TYPE, str] = {
@@ -2248,6 +2250,27 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.OUTPUT_NORM, # Final layer norm
MODEL_TENSOR.OUTPUT, # Output projection (lm_head)
],
MODEL_ARCH.HUNYUAN_MOE: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_Q_NORM,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_K_NORM,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_GATE_INP,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE_EXP,
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
MODEL_TENSOR.FFN_GATE_SHEXP,
MODEL_TENSOR.FFN_DOWN_SHEXP,
MODEL_TENSOR.FFN_UP_SHEXP,
],
# TODO
}

6
gguf-py/gguf/tensor_mapping.py
View File

@@ -305,6 +305,7 @@ class TensorNameMap:
"model.layers.{bid}.block_sparse_moe.router.layer", # granitemoe
"model.layers.{bid}.feed_forward.router", # llama4
"encoder.layers.{bid}.mlp.router.layer", # nomic-bert-moe
"model.layers.{bid}.mlp.gate.wg", # hunyuan
),
MODEL_TENSOR.FFN_GATE_INP_SHEXP: (
@@ -365,6 +366,7 @@ class TensorNameMap:
"model.layers.{bid}.mlp.shared_experts.up_proj", # deepseek deepseek2
"model.layers.{bid}.feed_forward.shared_expert.up_proj", # llama4
"model.layers.{bid}.feed_forward.down_proj",
"model.layers.{bid}.mlp.shared_mlp.up_proj", # hunyuan
),
# AWQ-activation gate
@@ -401,6 +403,7 @@ class TensorNameMap:
"model.layers.{bid}.mlp.shared_expert.gate_proj", # qwen2moe
"model.layers.{bid}.mlp.shared_experts.gate_proj", # deepseek deepseek2
"model.layers.{bid}.feed_forward.shared_expert.gate_proj", # llama4
"model.layers.{bid}.mlp.shared_mlp.gate_proj", # hunyuan
),
# Feed-forward down
@@ -450,11 +453,13 @@ class TensorNameMap:
"model.layers.{bid}.mlp.shared_experts.down_proj", # deepseek deepseek2
"model.layers.{bid}.feed_forward.shared_expert.down_proj", # llama4
"model.layers.{bid}.shared_mlp.output_linear", # granitemoe
"model.layers.{bid}.mlp.shared_mlp.down_proj", # hunyuan
),
MODEL_TENSOR.ATTN_Q_NORM: (
"language_model.encoder.layers.{bid}.self_attention.q_layernorm",
"model.layers.{bid}.self_attn.q_layernorm", # persimmon
"model.layers.{bid}.self_attn.query_layernorm", # hunyuan
"model.layers.{bid}.self_attn.q_norm", # cohere olmoe chameleon olmo2
"transformer.blocks.{bid}.attn.q_ln", # sea-lion
"encoder.layer.{bid}.attention.self.layer_norm_q", # jina-bert-v2
@@ -464,6 +469,7 @@ class TensorNameMap:
MODEL_TENSOR.ATTN_K_NORM: (
"language_model.encoder.layers.{bid}.self_attention.k_layernorm",
"model.layers.{bid}.self_attn.k_layernorm", # persimmon
"model.layers.{bid}.self_attn.key_layernorm", # hunyuan
"model.layers.{bid}.self_attn.k_norm", # cohere olmoe chameleon olmo2
"transformer.blocks.{bid}.attn.k_ln", # sea-lion
"encoder.layer.{bid}.attention.self.layer_norm_k", # jina-bert-v2