初始化项目,由ModelHub XC社区提供模型
Model: ayh015/myLightningOPD Source: Original Platform
This commit is contained in:
49
configs/models/qwen3-30B-A3B.sh
Normal file
49
configs/models/qwen3-30B-A3B.sh
Normal file
@@ -0,0 +1,49 @@
|
||||
NLAYERS=48
|
||||
FIRST_K_DENSE_REPLACE=0
|
||||
|
||||
arr=()
|
||||
for ((i=0; i<NLAYERS; i++)); do
|
||||
if (( i < FIRST_K_DENSE_REPLACE )); then
|
||||
arr+=(0)
|
||||
else
|
||||
arr+=(1)
|
||||
fi
|
||||
done
|
||||
|
||||
printf -v MOE_LAYER_FREQ "[%s]" "$(IFS=', '; echo "${arr[*]}")"
|
||||
|
||||
|
||||
MODEL_ARGS=(
|
||||
--disable-bias-linear
|
||||
--qk-layernorm
|
||||
--group-query-attention
|
||||
--num-attention-heads 32
|
||||
--num-query-groups 4
|
||||
--kv-channels 128
|
||||
--num-layers 48
|
||||
--hidden-size 2048
|
||||
--ffn-hidden-size 6144
|
||||
|
||||
--normalization RMSNorm
|
||||
--position-embedding-type rope
|
||||
--norm-epsilon 1e-6
|
||||
--rotary-percent 1.0
|
||||
--swiglu
|
||||
--untie-embeddings-and-output-weights
|
||||
--vocab-size 151936
|
||||
|
||||
--rotary-base 1000000
|
||||
|
||||
# moe
|
||||
--moe-ffn-hidden-size 768
|
||||
--moe-router-score-function softmax
|
||||
--moe-token-dispatcher-type alltoall
|
||||
--moe-router-topk 8
|
||||
--moe-layer-freq $MOE_LAYER_FREQ
|
||||
--num-experts 128
|
||||
--moe-grouped-gemm
|
||||
--moe-token-drop-policy probs
|
||||
--moe-router-dtype fp32
|
||||
--moe-permute-fusion
|
||||
--moe-aux-loss-coeff 0
|
||||
)
|
||||
20
configs/models/qwen3-4B.sh
Normal file
20
configs/models/qwen3-4B.sh
Normal file
@@ -0,0 +1,20 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
MODEL_ARGS=(
|
||||
--swiglu
|
||||
--num-layers 36
|
||||
--hidden-size 2560
|
||||
--ffn-hidden-size 9728
|
||||
--num-attention-heads 32
|
||||
--group-query-attention
|
||||
--num-query-groups 8
|
||||
--use-rotary-position-embeddings
|
||||
--disable-bias-linear
|
||||
--normalization "RMSNorm"
|
||||
--norm-epsilon 1e-6
|
||||
--rotary-base "${MODEL_ARGS_ROTARY_BASE:-1000000}"
|
||||
--vocab-size 151936
|
||||
--kv-channels 128
|
||||
--qk-layernorm
|
||||
)
|
||||
21
configs/models/qwen3-8B.sh
Normal file
21
configs/models/qwen3-8B.sh
Normal file
@@ -0,0 +1,21 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
MODEL_ARGS=(
|
||||
--swiglu
|
||||
--num-layers 36
|
||||
--hidden-size 4096
|
||||
--ffn-hidden-size 12288
|
||||
--num-attention-heads 32
|
||||
--group-query-attention
|
||||
--num-query-groups 8
|
||||
--use-rotary-position-embeddings
|
||||
--disable-bias-linear
|
||||
--normalization "RMSNorm"
|
||||
--norm-epsilon 1e-6
|
||||
--rotary-base 1000000
|
||||
--vocab-size 151936
|
||||
--kv-channels 128
|
||||
--qk-layernorm
|
||||
--untie-embeddings-and-output-weights
|
||||
)
|
||||
Reference in New Issue
Block a user