diff options
| author | saood06 <saood05@gmail.com> | 2025-01-23 10:24:10 -0600 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-01-23 18:24:10 +0200 |
| commit | 2195632581c4f52707059b5963fe622ccead0dd2 (patch) | |
| tree | 34d46a344c5d32ff699126cea9255eb13fd3b38a /gguf-py/gguf/constants.py | |
| parent | c2624b2fd324ff98cc137397f5b0e1d22869cb58 (diff) | |
Deepseek V3 support added (#176)
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
Diffstat (limited to 'gguf-py/gguf/constants.py')
| -rw-r--r-- | gguf-py/gguf/constants.py | 9 |
1 files changed, 9 insertions, 0 deletions
diff --git a/gguf-py/gguf/constants.py b/gguf-py/gguf/constants.py index 1bea66aa..90d5efec 100644 --- a/gguf-py/gguf/constants.py +++ b/gguf-py/gguf/constants.py @@ -89,6 +89,8 @@ class Keys: EXPERT_USED_COUNT = "{arch}.expert_used_count" EXPERT_SHARED_COUNT = "{arch}.expert_shared_count" EXPERT_WEIGHTS_SCALE = "{arch}.expert_weights_scale" + EXPERT_WEIGHTS_NORM = "{arch}.expert_weights_norm" + EXPERT_GATING_FUNC = "{arch}.expert_gating_func" POOLING_TYPE = "{arch}.pooling_type" LOGIT_SCALE = "{arch}.logit_scale" DECODER_START_TOKEN_ID = "{arch}.decoder_start_token_id" @@ -257,6 +259,7 @@ class MODEL_TENSOR(IntEnum): FFN_GATE_SHEXP = auto() FFN_DOWN_SHEXP = auto() FFN_UP_SHEXP = auto() + FFN_EXP_PROBS_B = auto() ATTN_Q_NORM = auto() ATTN_K_NORM = auto() LAYER_OUT_NORM = auto() @@ -387,6 +390,7 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = { MODEL_TENSOR.FFN_GATE_EXP: "blk.{bid}.ffn_gate_exps", MODEL_TENSOR.FFN_DOWN_EXP: "blk.{bid}.ffn_down_exps", MODEL_TENSOR.FFN_UP_EXP: "blk.{bid}.ffn_up_exps", + MODEL_TENSOR.FFN_EXP_PROBS_B: "blk.{bid}.exp_probs_b", MODEL_TENSOR.LAYER_OUT_NORM: "blk.{bid}.layer_output_norm", MODEL_TENSOR.SSM_IN: "blk.{bid}.ssm_in", MODEL_TENSOR.SSM_CONV1D: "blk.{bid}.ssm_conv1d", @@ -978,6 +982,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = { MODEL_TENSOR.FFN_GATE_SHEXP, MODEL_TENSOR.FFN_DOWN_SHEXP, MODEL_TENSOR.FFN_UP_SHEXP, + MODEL_TENSOR.FFN_EXP_PROBS_B ], MODEL_ARCH.CHATGLM : [ MODEL_TENSOR.TOKEN_EMBD, @@ -1177,6 +1182,10 @@ class GGMLQuantizationType(IntEnum): IQ2_TN = 42, +class ExpertGatingFuncType(IntEnum): + SOFTMAX = 1 + SIGMOID = 2 + # TODO: add GGMLFileType from ggml_ftype in ggml.h |