vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

gguf-py : Moved non-conflicting block mappings from architecture-specific ARCTIC mappigs to general mappings.

Browse source
This commit is contained in:
Stanisław Szymczyk 2024-05-16 20:35:04 +02:00
parent 85263f0568
commit 5b2be25d9b
1 changed files with 5 additions and 50 deletions
Download patch file Download diff file Expand all files Collapse all files

55
gguf-py/gguf/tensor_mapping.py
View file

@ -244,6 +244,7 @@ class TensorNameMap:
"encoder.layers.{bid}.mlp.fc11", # nomic-bert "encoder.layers.{bid}.mlp.fc11", # nomic-bert
"model.layers.{bid}.mlp.c_fc", # starcoder2 "model.layers.{bid}.mlp.c_fc", # starcoder2
"encoder.layer.{bid}.mlp.gated_layers_v", # jina-bert-v2 "encoder.layer.{bid}.mlp.gated_layers_v", # jina-bert-v2
"model.layers.{bid}.residual_mlp.w3", # arctic
), ),
MODEL_TENSOR.FFN_UP_EXP: ( MODEL_TENSOR.FFN_UP_EXP: (
@ -272,6 +273,7 @@ class TensorNameMap:
"encoder.layers.{bid}.mlp.fc12", # nomic-bert "encoder.layers.{bid}.mlp.fc12", # nomic-bert
"encoder.layer.{bid}.mlp.gated_layers_w", # jina-bert-v2 "encoder.layer.{bid}.mlp.gated_layers_w", # jina-bert-v2
"transformer.h.{bid}.mlp.linear_1", # refact "transformer.h.{bid}.mlp.linear_1", # refact
"model.layers.{bid}.residual_mlp.w1", # arctic
), ),
MODEL_TENSOR.FFN_GATE_EXP: ( MODEL_TENSOR.FFN_GATE_EXP: (
@ -306,6 +308,7 @@ class TensorNameMap:
"encoder.layers.{bid}.mlp.fc2", # nomic-bert "encoder.layers.{bid}.mlp.fc2", # nomic-bert
"model.layers.{bid}.mlp.c_proj", # starcoder2 "model.layers.{bid}.mlp.c_proj", # starcoder2
"encoder.layer.{bid}.mlp.wo", # jina-bert-v2 "encoder.layer.{bid}.mlp.wo", # jina-bert-v2
"model.layers.{bid}.residual_mlp.w2", # arctic
), ),
MODEL_TENSOR.FFN_DOWN_EXP: ( MODEL_TENSOR.FFN_DOWN_EXP: (
@ -380,60 +383,14 @@ class TensorNameMap:
"model.layers.{bid}.out_proj", "model.layers.{bid}.out_proj",
"backbone.layers.{bid}.mixer.out_proj", "backbone.layers.{bid}.mixer.out_proj",
), ),
} }
# architecture-specific block mappings # architecture-specific block mappings
arch_block_mappings_cfg: dict[MODEL_ARCH, dict[MODEL_TENSOR, tuple[str, ...]]] = { arch_block_mappings_cfg: dict[MODEL_ARCH, dict[MODEL_TENSOR, tuple[str, ...]]] = {
MODEL_ARCH.ARCTIC: { MODEL_ARCH.ARCTIC: {
MODEL_TENSOR.TOKEN_EMBD: (
"model.embed_tokens",
),
MODEL_TENSOR.OUTPUT_NORM: (
"model.norm",
),
MODEL_TENSOR.OUTPUT: (
"lm_head",
),
MODEL_TENSOR.ATTN_NORM: (
"model.layers.{bid}.input_layernorm",
),
MODEL_TENSOR.ATTN_Q: (
"model.layers.{bid}.self_attn.q_proj",
),
MODEL_TENSOR.ATTN_K: (
"model.layers.{bid}.self_attn.k_proj",
),
MODEL_TENSOR.ATTN_V: (
"model.layers.{bid}.self_attn.v_proj",
),
MODEL_TENSOR.ATTN_OUT: (
"model.layers.{bid}.self_attn.o_proj",
),
MODEL_TENSOR.FFN_GATE_INP: (
"model.layers.{bid}.block_sparse_moe.gate",
),
MODEL_TENSOR.FFN_NORM: ( MODEL_TENSOR.FFN_NORM: (
"model.layers.{bid}.residual_layernorm", "model.layers.{bid}.residual_layernorm",
), ),
MODEL_TENSOR.FFN_GATE: (
"model.layers.{bid}.residual_mlp.w1",
),
MODEL_TENSOR.FFN_DOWN: (
"model.layers.{bid}.residual_mlp.w2",
),
MODEL_TENSOR.FFN_UP: (
"model.layers.{bid}.residual_mlp.w3",
),
MODEL_TENSOR.FFN_GATE_EXP: (
"layers.{bid}.feed_forward.experts.w1",
),
MODEL_TENSOR.FFN_DOWN_EXP: (
"layers.{bid}.feed_forward.experts.w2",
),
MODEL_TENSOR.FFN_UP_EXP: (
"layers.{bid}.feed_forward.experts.w3",
),
MODEL_TENSOR.FFN_NORM_EXP: ( MODEL_TENSOR.FFN_NORM_EXP: (
"model.layers.{bid}.post_attention_layernorm", "model.layers.{bid}.post_attention_layernorm",
), ),
@ -452,11 +409,9 @@ class TensorNameMap:
for key in keys: for key in keys:
self.mapping[key] = (tensor, tensor_name) self.mapping[key] = (tensor, tensor_name)
if arch in self.arch_block_mappings_cfg: if arch in self.arch_block_mappings_cfg:
block_mappings = self.arch_block_mappings_cfg[arch] self.block_mappings_cfg.update(self.arch_block_mappings_cfg[arch])
else:
block_mappings = self.block_mappings_cfg
for bid in range(n_blocks): for bid in range(n_blocks):
for tensor, keys in block_mappings.items(): for tensor, keys in self.block_mappings_cfg.items():
if tensor not in MODEL_TENSORS[arch]: if tensor not in MODEL_TENSORS[arch]:
continue continue
# TODO: make this configurable # TODO: make this configurable