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

Add Nemotron/Minitron GGUF Conversion & Inference Support (#8922)

Browse source 
* Add nemotron GGUF conversion & inference support

* Fix formatting issues

* Remove unnecessary write_tensors()

* Update convert_hf_to_gguf.py

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

* Update src/llama.cpp

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

* Address comments by @compilade

* Replace ggml_mul_mat()->llm_build_lora_mm()

* Remove mutable variable

* Use  for bias tensors

* Cover corner case for role_scaling not in config.json

---------

Co-authored-by: compilade <git@compilade.net>
This commit is contained in:
Yoshi Suhara 2024-08-15 19:23:33 -07:00 • committed by GitHub
parent e3f6fd56b1
commit 2a24c8caa6
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 271 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

21
gguf-py/gguf/constants.py
View file

@ -219,6 +219,7 @@ class MODEL_ARCH(IntEnum):
T5 = auto()
T5ENCODER = auto()
JAIS = auto()
NEMOTRON = auto()
class MODEL_TENSOR(IntEnum):
@ -347,6 +348,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.T5: "t5",
MODEL_ARCH.T5ENCODER: "t5encoder",
MODEL_ARCH.JAIS: "jais",
MODEL_ARCH.NEMOTRON: "nemotron",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@ -1065,6 +1067,21 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.NEMOTRON: [
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_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_ROT_EMBD,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
# TODO
}
@ -1105,6 +1122,10 @@ MODEL_TENSOR_SKIP: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_ARCH.CHATGLM: [
MODEL_TENSOR.ROPE_FREQS,
],
MODEL_ARCH.NEMOTRON: [
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
],
}
#