manaestras commited on
Commit
01b8c47
·
verified ·
1 Parent(s): f5ce4c7

Upload angelslim_config.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. angelslim_config.json +68 -0
angelslim_config.json ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_config": {
3
+ "name": "HunyuanDense",
4
+ "model_path": "/",
5
+ "trust_remote_code": true,
6
+ "torch_dtype": "auto",
7
+ "device_map": "cpu",
8
+ "low_cpu_mem_usage": true,
9
+ "use_cache": false
10
+ },
11
+ "compression_config": {
12
+ "name": "PTQ",
13
+ "quantization": {
14
+ "name": "int4_gptq",
15
+ "bits": 4,
16
+ "quant_method": {
17
+ "weight": "per-group",
18
+ "group_size": 128
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
+ "mlp.gate.wg"
30
+ ],
31
+ "quant_analyse": false
32
+ }
33
+ },
34
+ "dataset_config": {
35
+ "name": "TextDataset",
36
+ "data_path": "./dataset/PTQ_data.jsonl",
37
+ "max_seq_length": 8192,
38
+ "num_samples": 128,
39
+ "batch_size": 1,
40
+ "shuffle": false
41
+ },
42
+ "global_config": {
43
+ "save_path": "/",
44
+ "max_seq_length": 8192,
45
+ "hidden_size": 1024,
46
+ "model_arch_type": "hunyuan_v1_dense",
47
+ "deploy_backend": "vllm"
48
+ },
49
+ "debug_info": {
50
+ "python": "3.10.14 (main, Mar 6 2025, 18:16:47) [GCC 11.4.0]",
51
+ "angelslim": {
52
+ "name": "angelslim",
53
+ "version": "be35620eca8ad4e6bf8263e84cee69d90c7045a4",
54
+ "source": "git"
55
+ },
56
+ "torch": {
57
+ "name": "torch",
58
+ "version": "2.7.0",
59
+ "source": "pip"
60
+ },
61
+ "transformers": {
62
+ "name": "transformers",
63
+ "version": "4.52.4",
64
+ "source": "pip"
65
+ },
66
+ "torch_cuda_version": "12.6"
67
+ }
68
+ }