File size: 3,248 Bytes
c64f61e
 
 
68e4b32
c64f61e
 
 
 
 
 
 
d20c687
c64f61e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d20c687
c64f61e
d20c687
c64f61e
 
 
 
 
 
d20c687
c64f61e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
---
library_name: transformers
license: apache-2.0
base_model: THUDM/GLM-4-32B-0414
tags:
- roleplay
- conversational
- axolotl
- qwen
---

# 残响 GLM4 32B(第一系列)

[English](./README.md) | [简体中文](./README-cn.md)

*空气中飘浮着一缕尘埃。它仿佛来自某个逝去的时代,但你无从追溯。它落在你的舌尖,滋味奇妙。*

![image/png](https://cdn-uploads.huggingface.co/production/uploads/634262af8d8089ebaefd410e/_ovgodU331FO4YAqFGCnk.png)

「残响」是一系列专注于SFW与NSFW角色扮演及对话的微调大语言模型。

## 量化版本
GGUF:
- 待补充!

EXL3:
- 待补充!

EXL2:
- 待补充!

其他格式:
- 待补充!

## 推荐参数
对话模板: GLM4  
采样器设置:
- 温度值 `1.0`
- 最小概率阈值 `0.1`

## 致谢
特别感谢Allura和ilya <3  
衷心感谢以下项目的开发者:  
- Axolotl(训练框架)  
- 智谱AI(基础模型)  
- Prime Intellect(算力支持)  
- 以及我的银行(资金支持)

## 其他信息

[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>查看Axolotl配置</summary>

axolotl版本: `0.10.0.dev0`
```yaml
# === 模型配置 ===
base_model: THUDM/GLM-4-32B-0414  # e.g. "mistralai/Mistral-Small-24B-Instruct-2501"
load_in_8bit: false
load_in_4bit: true

# === 训练设置 ===
num_epochs: 2
micro_batch_size: 3
gradient_accumulation_steps: 2
sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

# === 超参数配置 ===
optimizer: adamw_8bit
# Apollo-mini配置:
#optim_args: "proj=random,rank=1,scale=128.0,scale_type=tensor,update_proj_gap=200"
# 标准Apollo配置:
# optim_args: 
#optim_target_modules: all_linear
learning_rate: 1e-5
lr_scheduler: rex
weight_decay: 0.01
warmup_ratio: 0.05

# === LoRA ===
adapter: qlora
lora_r: 16
lora_alpha: 32
lora_dropout: 0.25
lora_target_modules:
lora_target_linear: true

# === 数据配置 ===
datasets:
  - path: allura-org/inkmix-v3.0
    type: chat_template
    split: train
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    train_on_eos: all

dataset_prepared_path: last_run_prepared
chat_template: jinja
chat_template_jinja: |
  [gMASK]<sop>{%- for msg in messages %}{%- if msg.role == 'system' %}<|system|>
  {{ msg.content }}{%- elif msg.role == 'user' %}<|user|>
  {{ msg.content }}{%- elif msg.role == 'assistant' %}<|assistant|>
  {{ msg.content }}{%- endif %}{%- endfor %}{% if add_generation_prompt %}<|assistant|>{% else %}<|user|>{% endif %}

# === 插件 ===
plugins:
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

# === 硬件优化 ===
gradient_checkpointing: offload
gradient_checkpointing_kwargs:
  use_reentrant: false
cut_cross_entropy: true
wandb_project: glm4-32b-inkmix-v3

# === Wandb追踪 ===
wandb_project: qwen3-8b-inkmix-v3

# === 检查点 ===
saves_per_epoch: 2
save_total_limit: 3

# === 高级设置 ===
output_dir: /workspace/ckpts
bf16: auto
flash_attention: true
train_on_inputs: false
group_by_length: false
logging_steps: 1
trust_remote_code: true

```

</details>