CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : support LiquidAI LFM2-MoE hybrid model (#16464)

Browse Source 
* llama : support LiquidAI LFM2-MoE hybrid model

Add support for [LiquidAI/LFM2-8B-A1B](https://huggingface.co/LiquidAI/LFM2-8B-A1B) model.
For more information about models, please read [the blog post](https://www.liquid.ai/company/news).

[HF PR](https://github.com/huggingface/transformers/pull/41401)
[GGUFs](https://huggingface.co/LiquidAI/LFM2-8B-A1B-GGUF)

* Do not use defaultdict

* Address PR feedback
This commit is contained in:
Tarek Dakhran
2025-10-07 20:03:35 +02:00
committed by GitHub
parent df1b612e29
commit aeaf8a36f0
7 changed files with 192 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
gguf-py/gguf/constants.py
View File

@@ -407,6 +407,7 @@ class MODEL_ARCH(IntEnum):
SMOLLM3 = auto()
GPT_OSS = auto()
LFM2 = auto()
LFM2MOE = auto()
DREAM = auto()
SMALLTHINKER = auto()
LLADA = auto()
@@ -749,6 +750,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.SMOLLM3: "smollm3",
MODEL_ARCH.GPT_OSS: "gpt-oss",
MODEL_ARCH.LFM2: "lfm2",
MODEL_ARCH.LFM2MOE: "lfm2moe",
MODEL_ARCH.DREAM: "dream",
MODEL_ARCH.SMALLTHINKER: "smallthinker",
MODEL_ARCH.LLADA: "llada",
@@ -2698,6 +2700,29 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.OUTPUT,
],
MODEL_ARCH.LFM2MOE: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.TOKEN_EMBD_NORM,
MODEL_TENSOR.SHORTCONV_CONV,
MODEL_TENSOR.SHORTCONV_INPROJ,
MODEL_TENSOR.SHORTCONV_OUTPROJ,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.ATTN_NORM, # operator_norm
MODEL_TENSOR.ATTN_Q_NORM,
MODEL_TENSOR.ATTN_K_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_GATE_INP,
MODEL_TENSOR.FFN_GATE_EXP,
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
MODEL_TENSOR.FFN_EXP_PROBS_B,
],
MODEL_ARCH.SMALLTHINKER: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,

2
gguf-py/gguf/tensor_mapping.py
View File

@@ -358,6 +358,7 @@ class TensorNameMap:
"model.layers.{bid}.mlp.router", # openai-moe
"model.layers.{bid}.mlp.gate.wg", # hunyuan
"model.layers.{bid}.block_sparse_moe.primary_router", # smallthinker
"model.layers.{bid}.feed_forward.gate", # lfm2moe
),
MODEL_TENSOR.FFN_GATE_INP_SHEXP: (
@@ -367,6 +368,7 @@ class TensorNameMap:
MODEL_TENSOR.FFN_EXP_PROBS_B: (
"model.layers.{bid}.mlp.gate.e_score_correction", # deepseek-v3 dots1
"model.layers.{bid}.mlp.moe_statics.e_score_correction", # ernie4.5-moe
"model.layers.{bid}.feed_forward.expert_bias", # lfm2moe
),
# Feed-forward up