manaestras commited on
Commit
b0eb9e7
·
verified ·
1 Parent(s): 8975920

Upload angelslim_config.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. angelslim_config.json +67 -0
angelslim_config.json ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_config": {
3
+ "name": "HunyuanDense",
4
+ "model_path": "/",
5
+ "trust_remote_code": true,
6
+ "torch_dtype": "auto",
7
+ "device_map": "auto",
8
+ "low_cpu_mem_usage": true,
9
+ "use_cache": false
10
+ },
11
+ "compression_config": {
12
+ "name": "PTQ",
13
+ "quantization": {
14
+ "name": "fp8_static",
15
+ "bits": 8,
16
+ "quant_method": {
17
+ "weight": "per-tensor",
18
+ "activation": "per-tensor"
19
+ },
20
+ "quant_helpers": [],
21
+ "smooth_alpha": 0.5,
22
+ "low_memory": false,
23
+ "modules_to_quantize": [],
24
+ "zero_point": true,
25
+ "mse_range": false,
26
+ "ignore_layers": [
27
+ "lm_head",
28
+ "model.embed_tokens"
29
+ ],
30
+ "quant_analyse": false
31
+ }
32
+ },
33
+ "dataset_config": {
34
+ "name": "TextDataset",
35
+ "data_path": "./dataset/PTQ_data.jsonl",
36
+ "max_seq_length": 4096,
37
+ "num_samples": 512,
38
+ "batch_size": 1,
39
+ "shuffle": false
40
+ },
41
+ "global_config": {
42
+ "save_path": "/",
43
+ "max_seq_length": 4096,
44
+ "hidden_size": 2048,
45
+ "model_arch_type": "hunyuan_v1_dense",
46
+ "deploy_backend": "vllm"
47
+ },
48
+ "debug_info": {
49
+ "python": "3.10.14 (main, Mar 6 2025, 18:16:47) [GCC 11.4.0]",
50
+ "angelslim": {
51
+ "name": "angelslim",
52
+ "version": "be35620eca8ad4e6bf8263e84cee69d90c7045a4",
53
+ "source": "git"
54
+ },
55
+ "torch": {
56
+ "name": "torch",
57
+ "version": "2.7.0",
58
+ "source": "pip"
59
+ },
60
+ "transformers": {
61
+ "name": "transformers",
62
+ "version": "4.52.4",
63
+ "source": "pip"
64
+ },
65
+ "torch_cuda_version": "12.6"
66
+ }
67
+ }