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

ggml : full ALiBi support (#7192)

Browse source 
* ggml : full ALiBi support

* ggml : update ggml_soft_max_ext() CUDA, SYCL

* ggml : ggml_flash_attn_ext() support ALiBi (CPU)

* ggml : ggml_flash_attn_ext() support ALiBi (Metal)

* ggml : fix warning

* ggml : ggml_flash_attn_ext() support ALiBi (CUDA)

ggml-ci

* ggml : fix assert message

* vulkan : add dev notes

* ggml : require mask when using ALiBi

ggml-ci

* convert : fix convert for refact models
This commit is contained in:
Georgi Gerganov 2024-05-11 10:32:41 +03:00 • committed by GitHub
parent e849648888
commit 9cb317f77e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
16 changed files with 350 additions and 825 deletions
Download patch file Download diff file Expand all files Collapse all files

4
gguf-py/gguf/tensor_mapping.py
View file

@ -137,6 +137,7 @@ class TensorNameMap:
"layers.{bid}.attention.wk", # llama-pth
"encoder.layer.{bid}.attention.self.key", # bert
"transformer.h.{bid}.attn.k_proj", # gpt-j
"transformer.h.{bid}.attn.k", # refact
"model.layers.layers.{bid}.self_attn.k_proj", # plamo
"model.layers.{bid}.attention.wk", # internlm2
"transformer.decoder_layer.{bid}.multi_head_attention.key" # Grok
@ -148,6 +149,7 @@ class TensorNameMap:
"layers.{bid}.attention.wv", # llama-pth
"encoder.layer.{bid}.attention.self.value", # bert
"transformer.h.{bid}.attn.v_proj", # gpt-j
"transformer.h.{bid}.attn.v", # refact
"model.layers.layers.{bid}.self_attn.v_proj", # plamo
"model.layers.{bid}.attention.wv", # internlm2
"transformer.decoder_layer.{bid}.multi_head_attention.value" # Grok
@ -229,6 +231,7 @@ class TensorNameMap:
"layers.{bid}.feed_forward.w3", # llama-pth
"encoder.layer.{bid}.intermediate.dense", # bert
"transformer.h.{bid}.mlp.fc_in", # gpt-j
"transformer.h.{bid}.mlp.linear_3", # refact
"language_model.encoder.layers.{bid}.mlp.dense_h_to_4h", # persimmon
"model.layers.{bid}.mlp.dense_h_to_4h", # persimmon
"transformer.h.{bid}.mlp.w1", # qwen
@ -266,6 +269,7 @@ class TensorNameMap:
"model.layers.layers.{bid}.mlp.gate_proj", # plamo
"model.layers.{bid}.feed_forward.w1", # internlm2
"encoder.layers.{bid}.mlp.fc12", # nomic-bert
"transformer.h.{bid}.mlp.linear_1", # refact
),
MODEL_TENSOR.FFN_GATE_EXP: (