File size: 3,407 Bytes
d22975e
 
f2b3e89
 
 
 
 
aa1d543
d22975e
aa1d543
68f58eb
aa1d543
 
 
44302d3
 
 
 
 
 
 
 
42b70a6
44302d3
 
 
 
 
 
 
 
 
 
42b70a6
44302d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d22975e
44302d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
32ee7b6
 
 
 
 
 
44302d3
d22975e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44302d3
f2b3e89
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
137
138
139
140
141
142
---
library_name: peft
license: cc-by-nc-4.0
datasets:
- theblackcat102/evol-codealpaca-v1
language:
- en
pipeline_tag: text2text-generation
---
## llama-2-13b-code-alpaca
[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)

Trained for 3 epochs on `theblackcat102/evol-codealpaca-v1` dataset, scored decent on locally run perplexity at 4.36.

## Axolotl config used

```yaml
base_model: NousResearch/Llama-2-13b-hf
base_model_config: NousResearch/Llama-2-13b-hf
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
push_dataset_to_hub:
hub_model_id: 

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: theblackcat102/evol-codealpaca-v1
    type: alpaca
dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: ./checkpoints/llama-2-13b-qlora

adapter: qlora
lora_model_dir:

sequence_len: 4096
max_packed_sequence_len: 4096
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 2
num_epochs: 3
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.0001

train_on_inputs: false
group_by_length: true
bf16: true
fp16: false
tf32: true

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention: true
flash_attention:

warmup_steps: 10
eval_steps: 50
save_steps:
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"
```

And then merged with Axolotl via:

```
accelerate launch scripts/finetune.py configs/your_config.yml --merge_lora --lora_model_dir="./completed-model" --load_in_8bit=False --load_in_4bit=False
```

## Training procedure

The following `bitsandbytes` quantization config was used during training:
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16

The following `bitsandbytes` quantization config was used during training:
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16

The following `bitsandbytes` quantization config was used during training:
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16

The following `bitsandbytes` quantization config was used during training:
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16


### Framework versions
- PEFT 0.5.0.dev0