tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-18 14:18:50 -04:00
Code Issues Packages Projects Releases Wiki Activity

Add LLaDA 8b Diffusion model (#14771)

Browse Source 
* Add support for Llada-8b: diffusion model

* Add README

* Fix README and convert_hf_to_gguf

* convert_hf_to_gguf.py: address review comments

* Make everything in a single example

* Remove model-specific sampling

* Remove unused argmax

* Remove braced initializers, improve README.md a bit

* Add diffusion specific gguf params in set_vocab, remove setting rope_theta and rms_norm_eps

* Remove adding the mask token

* Move add_add_bos_token to set_vocab

* use add_bool in gguf_writer.py
This commit is contained in:
Aman Gupta
2025-07-31 19:49:09 +08:00
committed by GitHub
parent 11490b3672
commit 8a4a856277
12 changed files with 931 additions and 385 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
gguf-py/gguf/constants.py
View File

@@ -279,6 +279,9 @@ class Keys:
class Projector:
STACK_FACTOR = "clip.audio.projector.stack_factor"
class Diffusion:
SHIFT_LOGITS = "diffusion.shift_logits"
#
# recommended mapping of model tensor names for storage in gguf
#
@@ -377,6 +380,7 @@ class MODEL_ARCH(IntEnum):
LFM2 = auto()
DREAM = auto()
SMALLTHINKER = auto()
LLADA = auto()
class VISION_PROJECTOR_TYPE(IntEnum):
@@ -697,6 +701,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.LFM2: "lfm2",
MODEL_ARCH.DREAM: "dream",
MODEL_ARCH.SMALLTHINKER: "smallthinker",
MODEL_ARCH.LLADA: "llada",
}
VISION_PROJECTOR_TYPE_NAMES: dict[VISION_PROJECTOR_TYPE, str] = {
@@ -1318,6 +1323,21 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.LLADA: [
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.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.QWEN2VL: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,