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

llama : add support for Chameleon (#8543)

Browse source 
* convert chameleon hf to gguf

* add chameleon tokenizer tests

* fix lint

* implement chameleon graph

* add swin norm param

* return qk norm weights and biases to original format

* implement swin norm

* suppress image token output

* rem tabs

* add comment to conversion

* fix ci

* check for k norm separately

* adapt to new lora implementation

* fix layer input for swin norm

* move swin_norm in gguf writer

* add comment regarding special token regex in chameleon pre-tokenizer

* Update src/llama.cpp

Co-authored-by: compilade <git@compilade.net>

* fix punctuation regex in chameleon pre-tokenizer (@compilade)

Co-authored-by: compilade <git@compilade.net>

* fix lint

* trigger ci

---------

Co-authored-by: compilade <git@compilade.net>
This commit is contained in:
nopperl 2024-09-28 12:08:43 +00:00 • committed by GitHub
parent 43bcdd9703
commit 9a913110cf
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
10 changed files with 505 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

19
gguf-py/gguf/constants.py
View file

@ -94,6 +94,7 @@ class Keys:
DECODER_START_TOKEN_ID = "{arch}.decoder_start_token_id"
ATTN_LOGIT_SOFTCAPPING = "{arch}.attn_logit_softcapping"
FINAL_LOGIT_SOFTCAPPING = "{arch}.final_logit_softcapping"
SWIN_NORM = "{arch}.swin_norm"
RESCALE_EVERY_N_LAYERS = "{arch}.rescale_every_n_layers"
TIME_MIX_EXTRA_DIM = "{arch}.time_mix_extra_dim"
TIME_DECAY_EXTRA_DIM = "{arch}.time_decay_extra_dim"
@ -236,6 +237,7 @@ class MODEL_ARCH(IntEnum):
EXAONE = auto()
GRANITE = auto()
GRANITE_MOE = auto()
CHAMELEON = auto()
class MODEL_TENSOR(IntEnum):
@ -394,6 +396,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.EXAONE: "exaone",
MODEL_ARCH.GRANITE: "granite",
MODEL_ARCH.GRANITE_MOE: "granitemoe",
MODEL_ARCH.CHAMELEON: "chameleon",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@ -1260,6 +1263,22 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.CHAMELEON: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
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_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
# TODO
}

3
gguf-py/gguf/gguf_writer.py
View file

@ -670,6 +670,9 @@ class GGUFWriter:
def add_expert_weights_scale(self, value: float) -> None:
self.add_float32(Keys.LLM.EXPERT_WEIGHTS_SCALE.format(arch=self.arch), value)
def add_swin_norm(self, value: bool) -> None:
self.add_bool(Keys.LLM.SWIN_NORM.format(arch=self.arch), value)
def add_rescale_every_n_layers(self, count: int) -> None:
self.add_uint32(Keys.LLM.RESCALE_EVERY_N_LAYERS.format(arch=self.arch), count)

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

@ -380,7 +380,7 @@ 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
@ -389,7 +389,7 @@ 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