CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into compilade/mamba2

Browse Source
This commit is contained in:
Francis Couture-Harpin
2024-10-01 13:09:40 -04:00
parent 273e7a495a f1b8c42711
commit 7d6cb36895
105 changed files with 8055 additions and 5231 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
gguf-py/gguf/tensor_mapping.py
View File

@@ -87,6 +87,9 @@ class TensorNameMap:
"rope.freqs", # llama-pth
"rotary_pos_emb.inv_freq", # chatglm
),
MODEL_TENSOR.ROPE_FACTORS_LONG: (),
MODEL_TENSOR.ROPE_FACTORS_SHORT: (),
}
block_mappings_cfg: dict[MODEL_TENSOR, tuple[str, ...]] = {
@@ -251,11 +254,12 @@ class TensorNameMap:
),
MODEL_TENSOR.FFN_GATE_INP: (
"layers.{bid}.feed_forward.gate", # mixtral
"model.layers.{bid}.block_sparse_moe.gate", # mixtral
"model.layers.{bid}.mlp.gate", # qwen2moe olmoe
"transformer.decoder_layer.{bid}.router", # Grok
"transformer.blocks.{bid}.ffn.router.layer", # dbrx
"layers.{bid}.feed_forward.gate", # mixtral
"model.layers.{bid}.block_sparse_moe.gate", # mixtral
"model.layers.{bid}.mlp.gate", # qwen2moe olmoe
"transformer.decoder_layer.{bid}.router", # Grok
"transformer.blocks.{bid}.ffn.router.layer", # dbrx
"model.layers.{bid}.block_sparse_moe.router.layer", # granitemoe
),
MODEL_TENSOR.FFN_GATE_INP_SHEXP: (
@@ -364,10 +368,11 @@ class TensorNameMap:
),
MODEL_TENSOR.FFN_DOWN_EXP: (
"layers.{bid}.feed_forward.experts.w2", # mixtral (merged)
"transformer.decoder_layer.{bid}.moe.linear_1", # Grok (merged)
"transformer.blocks.{bid}.ffn.experts.mlp.w2", # dbrx
"model.layers.{bid}.mlp.experts.down_proj", # qwen2moe olmoe (merged)
"layers.{bid}.feed_forward.experts.w2", # mixtral (merged)
"transformer.decoder_layer.{bid}.moe.linear_1", # Grok (merged)
"transformer.blocks.{bid}.ffn.experts.mlp.w2", # dbrx
"model.layers.{bid}.mlp.experts.down_proj", # qwen2moe olmoe (merged)
"model.layers.{bid}.block_sparse_moe.output_linear", # granitemoe
),
MODEL_TENSOR.FFN_DOWN_SHEXP: (
@@ -378,7 +383,7 @@ class TensorNameMap:
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.q_norm", # cohere olmoe
"model.layers.{bid}.self_attn.q_norm", # cohere olmoe chameleon
"transformer.blocks.{bid}.attn.q_ln", # sea-lion
"encoder.layer.{bid}.attention.self.layer_norm_q", # jina-bert-v2
"transformer.layers.{bid}.attn.q_norm", # openelm
@@ -387,7 +392,7 @@ 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.k_norm", # cohere olmoe
"model.layers.{bid}.self_attn.k_norm", # cohere olmoe chameleon
"transformer.blocks.{bid}.attn.k_ln", # sea-lion
"encoder.layer.{bid}.attention.self.layer_norm_k", # jina-bert-v2
"transformer.layers.{bid}.attn.k_norm", # openelm
@@ -681,6 +686,15 @@ class TensorNameMap:
MODEL_TENSOR.ENC_OUTPUT_NORM: (
"encoder.final_layer_norm", # t5
),
MODEL_TENSOR.CLS: (
"classifier", # jina
"classifier.dense", # roberta
),
MODEL_TENSOR.CLS_OUT: (
"classifier.out_proj", # roberta
),
}
# architecture-specific block mappings