Commit
•
a9d8c75
1
Parent(s):
afe61ae
End of training
Browse files
README.md
ADDED
@@ -0,0 +1,168 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
library_name: peft
|
4 |
+
tags:
|
5 |
+
- axolotl
|
6 |
+
- generated_from_trainer
|
7 |
+
base_model: mistralai/Mistral-7B-v0.1
|
8 |
+
model-index:
|
9 |
+
- name: Skanderbeg-mistral-alpaca
|
10 |
+
results: []
|
11 |
+
---
|
12 |
+
|
13 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
14 |
+
should probably proofread and complete it, then remove this comment. -->
|
15 |
+
|
16 |
+
[<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)
|
17 |
+
<details><summary>See axolotl config</summary>
|
18 |
+
|
19 |
+
axolotl version: `0.4.0`
|
20 |
+
```yaml
|
21 |
+
base_model: mistralai/Mistral-7B-v0.1
|
22 |
+
model_type: MistralForCausalLM
|
23 |
+
tokenizer_type: LlamaTokenizer
|
24 |
+
is_mistral_derived_model: true
|
25 |
+
|
26 |
+
load_in_8bit: false
|
27 |
+
load_in_4bit: true
|
28 |
+
strict: false
|
29 |
+
|
30 |
+
lora_fan_in_fan_out: false
|
31 |
+
data_seed: 49
|
32 |
+
seed: 49
|
33 |
+
|
34 |
+
datasets:
|
35 |
+
- path: alpaca_synth_queries_healed.jsonl
|
36 |
+
type: sharegpt
|
37 |
+
conversation: alpaca
|
38 |
+
dataset_prepared_path: last_run_prepared
|
39 |
+
val_set_size: 0.1
|
40 |
+
output_dir: ./qlora-alpaca-out
|
41 |
+
hub_model_id: Skanderbeg/Skanderbeg-mistral-alpaca
|
42 |
+
|
43 |
+
adapter: qlora
|
44 |
+
lora_model_dir:
|
45 |
+
|
46 |
+
sequence_len: 896
|
47 |
+
sample_packing: false
|
48 |
+
pad_to_sequence_len: true
|
49 |
+
|
50 |
+
lora_r: 32
|
51 |
+
lora_alpha: 16
|
52 |
+
lora_dropout: 0.05
|
53 |
+
lora_target_linear: true
|
54 |
+
lora_fan_in_fan_out:
|
55 |
+
lora_target_modules:
|
56 |
+
- gate_proj
|
57 |
+
- down_proj
|
58 |
+
- up_proj
|
59 |
+
- q_proj
|
60 |
+
- v_proj
|
61 |
+
- k_proj
|
62 |
+
- o_proj
|
63 |
+
|
64 |
+
wandb_project: honeycomb-finetune
|
65 |
+
wandb_entity: vinaydaredevil27
|
66 |
+
|
67 |
+
gradient_accumulation_steps: 4
|
68 |
+
micro_batch_size: 16
|
69 |
+
eval_batch_size: 16
|
70 |
+
num_epochs: 1
|
71 |
+
optimizer: adamw_bnb_8bit
|
72 |
+
lr_scheduler: cosine
|
73 |
+
learning_rate: 0.0002
|
74 |
+
max_grad_norm: 1.0
|
75 |
+
adam_beta2: 0.95
|
76 |
+
adam_epsilon: 0.00001
|
77 |
+
save_total_limit: 12
|
78 |
+
|
79 |
+
train_on_inputs: false
|
80 |
+
group_by_length: false
|
81 |
+
bf16: true
|
82 |
+
fp16: false
|
83 |
+
tf32: false
|
84 |
+
|
85 |
+
gradient_checkpointing: true
|
86 |
+
early_stopping_patience:
|
87 |
+
resume_from_checkpoint:
|
88 |
+
local_rank:
|
89 |
+
logging_steps: 1
|
90 |
+
xformers_attention:
|
91 |
+
flash_attention: true
|
92 |
+
|
93 |
+
loss_watchdog_threshold: 5.0
|
94 |
+
loss_watchdog_patience: 3
|
95 |
+
|
96 |
+
warmup_steps: 20
|
97 |
+
evals_per_epoch: 4
|
98 |
+
eval_table_size:
|
99 |
+
eval_table_max_new_tokens: 128
|
100 |
+
saves_per_epoch: 6
|
101 |
+
debug:
|
102 |
+
weight_decay: 0.0
|
103 |
+
fsdp:
|
104 |
+
fsdp_config:
|
105 |
+
special_tokens:
|
106 |
+
bos_token: "<s>"
|
107 |
+
eos_token: "</s>"
|
108 |
+
unk_token: "<unk>"
|
109 |
+
save_safetensors: true
|
110 |
+
```
|
111 |
+
|
112 |
+
</details><br>
|
113 |
+
|
114 |
+
# Skanderbeg-mistral-alpaca
|
115 |
+
|
116 |
+
This model is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) on the None dataset.
|
117 |
+
It achieves the following results on the evaluation set:
|
118 |
+
- Loss: 0.0357
|
119 |
+
|
120 |
+
## Model description
|
121 |
+
|
122 |
+
More information needed
|
123 |
+
|
124 |
+
## Intended uses & limitations
|
125 |
+
|
126 |
+
More information needed
|
127 |
+
|
128 |
+
## Training and evaluation data
|
129 |
+
|
130 |
+
More information needed
|
131 |
+
|
132 |
+
## Training procedure
|
133 |
+
|
134 |
+
### Training hyperparameters
|
135 |
+
|
136 |
+
The following hyperparameters were used during training:
|
137 |
+
- learning_rate: 0.0002
|
138 |
+
- train_batch_size: 16
|
139 |
+
- eval_batch_size: 16
|
140 |
+
- seed: 49
|
141 |
+
- distributed_type: multi-GPU
|
142 |
+
- num_devices: 2
|
143 |
+
- gradient_accumulation_steps: 4
|
144 |
+
- total_train_batch_size: 128
|
145 |
+
- total_eval_batch_size: 32
|
146 |
+
- optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
|
147 |
+
- lr_scheduler_type: cosine
|
148 |
+
- lr_scheduler_warmup_steps: 20
|
149 |
+
- num_epochs: 1
|
150 |
+
|
151 |
+
### Training results
|
152 |
+
|
153 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
154 |
+
|:-------------:|:------:|:----:|:---------------:|
|
155 |
+
| 1.1418 | 0.0011 | 1 | 1.1640 |
|
156 |
+
| 0.0691 | 0.25 | 225 | 0.0725 |
|
157 |
+
| 0.0518 | 0.5 | 450 | 0.0441 |
|
158 |
+
| 0.033 | 0.75 | 675 | 0.0367 |
|
159 |
+
| 0.0384 | 1.0 | 900 | 0.0357 |
|
160 |
+
|
161 |
+
|
162 |
+
### Framework versions
|
163 |
+
|
164 |
+
- PEFT 0.10.0
|
165 |
+
- Transformers 4.40.2
|
166 |
+
- Pytorch 2.2.2+cu121
|
167 |
+
- Datasets 2.19.1
|
168 |
+
- Tokenizers 0.19.1
|