upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-23 11:10:55 -05:00
Code Issues Packages Projects Releases Wiki Activity

spec: add EAGLE3 speculative decoding support (#18039)

Browse Source 
* llama : enable layer input extraction

* spec: support eagle3

* eagle3: fix params bug

* eagle3: support Gemma4 eagle3 from RedHatAI

* eagle3: set sync when get features from target

Co-authored-by: tnhnyzc <115956684+tnhnyzc@users.noreply.github.com>

* eagle3 : fix ubatch handling in embd_layer_inp extraction and encoder

Co-authored-by: Doğaç Eldenk <dogacel@gmail.com>

* eagle3: adapt to upstream changes

* eagle3: fix rebase issues and adapt to upstream changes

* eagle3:exclude the eagle3 arch from test-llama-archs

* eagle3: fix editorconfig check failures

* eagle3: fix multi-seq issue in d2t vocab mapping

* cont : minor style / clean-up

* spec : remove `common_speculative_setup_draft_model()`

* llama : clean-up unused API

* eagle3: set d2t vocab mapping in decode graph

* cont : assert layer inputs are configured

* hparams : use n_embd_inp instead of n_embd_target_features

* eagle3: make output.weight optional and inherit from target model when needed

* haparams : generic norm-before-residual param

* llama-ext : consistent names

* cont : fix

* hparams : remove target_hidden_size

* cparams : rename output_layer_inp -> embeddings_layer_inp

* arch : reuse ATTN_NORM_2 instead of adding new hidden norm

* llama : clean-up names

* cont : add assert + comment

* Update conversion/llama.py

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: tnhnyzc <115956684+tnhnyzc@users.noreply.github.com>
Co-authored-by: Doğaç Eldenk <dogacel@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Ruixiang Wang
2026-06-12 09:21:06 +02:00
committed by GitHub
parent 85f99dca8b
commit 88a39274ec
27 changed files with 1161 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
gguf-py/gguf/constants.py
View File

@@ -154,6 +154,9 @@ class Keys:
HIDDEN_ACT = "{arch}.hidden_activation"
DENSE_FEAT_IN_SIZE = "{arch}.{dense}_feat_in"
DENSE_FEAT_OUT_SIZE = "{arch}.{dense}_feat_out"
TARGET_LAYERS = "{arch}.target_layers"
TARGET_HIDDEN_SIZE = "{arch}.target_hidden_size"
NORM_BEFORE_RESIDUAL = "{arch}.norm_before_residual"
class Attention:
HEAD_COUNT = "{arch}.attention.head_count"
@@ -511,6 +514,7 @@ class MODEL_ARCH(IntEnum):
RND1 = auto()
PANGU_EMBED = auto()
MISTRAL3 = auto()
EAGLE3 = auto()
MISTRAL4 = auto()
PADDLEOCR = auto()
MIMO2 = auto()
@@ -901,14 +905,17 @@ class MODEL_TENSOR(IntEnum):
A_PER_DIM_K_SCALE = auto() # gemma4
A_PER_DIM_SCALE = auto() # gemma4
# nextn/mtp
NEXTN_PROJ_PRE = auto()
NEXTN_PROJ_POST = auto()
NEXTN_EH_PROJ = auto()
NEXTN_EMBED_TOKENS = auto()
NEXTN_ENORM = auto()
NEXTN_HNORM = auto()
NEXTN_PROJ_PRE = auto()
NEXTN_PROJ_POST = auto()
NEXTN_EH_PROJ = auto()
NEXTN_EMBED_TOKENS = auto()
NEXTN_ENORM = auto()
NEXTN_HNORM = auto()
NEXTN_SHARED_HEAD_HEAD = auto()
NEXTN_SHARED_HEAD_NORM = auto()
# eagle3
FC = auto() # feature fusion layer
D2T = auto() # draft to target vocabulary mapping
# lfm2 audio
A_ENC_NORM_CONV = auto()
A_ENC_LINEAR_POS = auto()
@@ -1063,6 +1070,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.RND1: "rnd1",
MODEL_ARCH.PANGU_EMBED: "pangu-embedded",
MODEL_ARCH.MISTRAL3: "mistral3",
MODEL_ARCH.EAGLE3: "eagle3",
MODEL_ARCH.MISTRAL4: "mistral4",
MODEL_ARCH.PADDLEOCR: "paddleocr",
MODEL_ARCH.MIMO2: "mimo2",
@@ -1095,8 +1103,8 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.POS_EMBD: "position_embd",
MODEL_TENSOR.OUTPUT_NORM: "output_norm",
MODEL_TENSOR.OUTPUT: "output",
MODEL_TENSOR.DENSE_2_OUT: "dense_2", # embeddinggemma 2_Dense
MODEL_TENSOR.DENSE_3_OUT: "dense_3", # embeddinggemma 2_Dense
MODEL_TENSOR.DENSE_2_OUT: "dense_2", # embeddinggemma 2_Dense
MODEL_TENSOR.DENSE_3_OUT: "dense_3", # embeddinggemma 2_Dense
MODEL_TENSOR.ROPE_FREQS: "rope_freqs",
MODEL_TENSOR.ROPE_FACTORS_LONG: "rope_factors_long",
MODEL_TENSOR.ROPE_FACTORS_SHORT: "rope_factors_short",
@@ -1488,6 +1496,8 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.NEXTN_HNORM: "blk.{bid}.nextn.hnorm",
MODEL_TENSOR.NEXTN_SHARED_HEAD_HEAD: "blk.{bid}.nextn.shared_head_head",
MODEL_TENSOR.NEXTN_SHARED_HEAD_NORM: "blk.{bid}.nextn.shared_head_norm",
MODEL_TENSOR.FC: "fc",
MODEL_TENSOR.D2T: "d2t",
}
MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
@@ -4028,6 +4038,24 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.EAGLE3: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_NORM_2,
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_TENSOR.FC,
MODEL_TENSOR.D2T,
],
MODEL_ARCH.MISTRAL4: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,