Upload inference/runner_config/tp32.yaml with huggingface_hub
Browse files
inference/runner_config/tp32.yaml
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# coding=utf-8
|
| 2 |
+
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All rights reserved.
|
| 3 |
+
|
| 4 |
+
model_name: "pangu_ultra_moe"
|
| 5 |
+
model_path: "./model"
|
| 6 |
+
exe_mode: "eager" # ["dynamo", "eager"]
|
| 7 |
+
|
| 8 |
+
model_config:
|
| 9 |
+
tokenizer_mode: default # ["default", "chat"]
|
| 10 |
+
mm_quant_mode: None
|
| 11 |
+
mla_backend: absorb # [native, absorb]
|
| 12 |
+
with_ckpt: 1 # [0, 1]
|
| 13 |
+
enable_profiler: 0 # [0, 1]
|
| 14 |
+
|
| 15 |
+
data_config:
|
| 16 |
+
input_max_len: 4096
|
| 17 |
+
max_new_tokens: 28000
|
| 18 |
+
batch_size: 1
|
| 19 |
+
max_position_embeddings: 32768
|
| 20 |
+
|
| 21 |
+
parallel_config:
|
| 22 |
+
attn_tp_size: 32
|
| 23 |
+
moe_tp_size: 32
|
| 24 |
+
embed_tp_size: 32
|
| 25 |
+
|
| 26 |
+
sampling_config:
|
| 27 |
+
top_n_sigma: 0.05
|
| 28 |
+
top_p: 1.0
|
| 29 |
+
temperature: 0.7
|
| 30 |
+
top_k: -1
|