Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

model : BailingMoE3 Support (#26608)

Browse Source 
* Adding support for bailingmoe3

* Adds speculative decoding support

* Make BailingMoE3 safe gate metadata optional

* bailingmoe3: apply trained SwiGLU clamps

* common: fix Bailing V3 tool argument parsing

* llama-model-saver, instantiate float vector metadata writer

* bailingmoe3: support Q-LoRA (Ling-3.0-tiny)

Ling-3.0-flash sets q_lora_rank: None and projects Q directly, so the current
implementation loads a single ATTN_Q tensor. Ling-3.0-tiny sets q_lora_rank: 256
and routes Q through a LoRA bottleneck instead:

    q_a_proj -> q_a_layernorm -> q_b_proj

Conversion therefore failed with:

    ValueError: Can not map tensor 'model.layers.3.attention.q_a_layernorm.weight'

Add the missing path, mirroring the existing deepseek2 MLA implementation:

  * constants.py       - add ATTN_Q_A / ATTN_Q_B / ATTN_Q_A_NORM to BAILINGMOE3
  * tensor_mapping.py  - map model.layers.{bid}.attention.q_{a,b}_proj and
                         q_a_layernorm
  * conversion         - emit attention.q_lora_rank when the config has it
  * bailingmoe3.cpp    - read n_lora_q; create the Q-LoRA tensors and build Q
                         through the bottleneck when q_lora_rank > 0

Everything is gated on q_lora_rank > 0. Ling-3.0-flash's config has no
q_lora_rank, the converter only emits the key when present, hparams.n_lora_q
defaults to 0, and get_key(..., required=false) leaves the target untouched when
the key is absent - so flash keeps taking the existing direct-Q branch.

The LoRA path produces the same shape as the direct projection, so the
nope/rope split, RoPE application and wk_b absorption downstream are unchanged.

* small mtp change

* bailingmoe3: support separate MTP GGUF and Q-LoRA MTP

* gguf: remove duplicate add_kda_gate_lower_bound definition

---------

Co-authored-by: bloomer <bloomer@booper.brushtail.me>
Co-authored-by: Dyluhn <dylanranejohnston1@gmail.com>
This commit is contained in:
Toby
2026-08-17 09:49:49 +02:00
committed by GitHub
co-authored by bloomer Dyluhn
parent 4197155add
commit 3733366720
18 changed files with 938 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gguf-py/gguf/constants.py
+50 -1
View File
@@ -261,6 +261,7 @@ class Keys:
class KDA:
HEAD_DIM = "{arch}.kda.head_dim"
SAFE_GATE = "{arch}.kda.safe_gate"
GATE_LOWER_BOUND = "{arch}.kda.gate_lower_bound"
class WKV:
@@ -552,6 +553,7 @@ class MODEL_ARCH(IntEnum):
PLM = auto()
BAILINGMOE = auto()
BAILINGMOE2 = auto()
BAILINGMOE3 = auto()
DOTS1 = auto()
ARCEE = auto()
AFMOE = auto()
@@ -1267,6 +1269,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.PLM: "plm",
MODEL_ARCH.BAILINGMOE: "bailingmoe",
MODEL_ARCH.BAILINGMOE2: "bailingmoe2",
MODEL_ARCH.BAILINGMOE3: "bailingmoe3",
MODEL_ARCH.DOTS1: "dots1",
MODEL_ARCH.ARCEE: "arcee",
MODEL_ARCH.AFMOE: "afmoe",
@@ -4234,6 +4237,50 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.NEXTN_SHARED_HEAD_NORM,
MODEL_TENSOR.LAYER_OUT_NORM,
],
MODEL_ARCH.BAILINGMOE3: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_Q_A,
MODEL_TENSOR.ATTN_Q_B,
MODEL_TENSOR.ATTN_Q_A_NORM,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_GATE,
MODEL_TENSOR.ATTN_KV_A_MQA,
MODEL_TENSOR.ATTN_KV_B,
MODEL_TENSOR.ATTN_K_B,
MODEL_TENSOR.ATTN_V_B,
MODEL_TENSOR.ATTN_KV_A_NORM,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.FFN_GATE_INP,
MODEL_TENSOR.FFN_GATE_EXP,
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
MODEL_TENSOR.FFN_GATE_SHEXP,
MODEL_TENSOR.FFN_DOWN_SHEXP,
MODEL_TENSOR.FFN_UP_SHEXP,
MODEL_TENSOR.FFN_EXP_PROBS_B,
MODEL_TENSOR.SSM_CONV1D_Q,
MODEL_TENSOR.SSM_CONV1D_K,
MODEL_TENSOR.SSM_CONV1D_V,
MODEL_TENSOR.SSM_F_A,
MODEL_TENSOR.SSM_BETA,
MODEL_TENSOR.SSM_A,
MODEL_TENSOR.SSM_G_A,
MODEL_TENSOR.SSM_DT,
MODEL_TENSOR.SSM_NORM,
MODEL_TENSOR.NEXTN_EH_PROJ,
MODEL_TENSOR.NEXTN_ENORM,
MODEL_TENSOR.NEXTN_HNORM,
MODEL_TENSOR.LAYER_OUT_NORM,
],
MODEL_ARCH.DOTS1: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
@@ -5487,7 +5534,9 @@ KEY_SSM_GROUP_COUNT = Keys.SSM.GROUP_COUNT
KEY_SSM_DT_B_C_RMS = Keys.SSM.DT_B_C_RMS
# KDA
KEY_KDA_HEAD_DIM = Keys.KDA.HEAD_DIM
KEY_KDA_HEAD_DIM = Keys.KDA.HEAD_DIM
KEY_KDA_SAFE_GATE = Keys.KDA.SAFE_GATE
KEY_KDA_GATE_LOWER_BOUND = Keys.KDA.GATE_LOWER_BOUND
# tokenization
KEY_TOKENIZER_MODEL = Keys.Tokenizer.MODEL