kweinmeister
commited on
End of training
Browse files- README.md +41 -48
- adapter_model.bin +1 -1
README.md
CHANGED
@@ -5,8 +5,6 @@ base_model: google/gemma-2-27b-it
|
|
5 |
tags:
|
6 |
- axolotl
|
7 |
- generated_from_trainer
|
8 |
-
datasets:
|
9 |
-
- databricks/databricks-dolly-15k
|
10 |
model-index:
|
11 |
- name: gemma-2-27b-it-dolly-15k
|
12 |
results: []
|
@@ -18,17 +16,11 @@ should probably proofread and complete it, then remove this comment. -->
|
|
18 |
[<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)
|
19 |
<details><summary>See axolotl config</summary>
|
20 |
|
21 |
-
axolotl version: `0.
|
22 |
```yaml
|
23 |
base_model: google/gemma-2-27b-it
|
24 |
-
model_type: AutoModelForCausalLM
|
25 |
-
tokenizer_type: AutoTokenizer
|
26 |
hub_model_id: kweinmeister/gemma-2-27b-it-dolly-15k
|
27 |
|
28 |
-
# https://github.com/vllm-project/vllm/issues/10590
|
29 |
-
bnb_config_kwargs:
|
30 |
-
bnb_4bit_quant_storage: uint8
|
31 |
-
|
32 |
load_in_8bit: false
|
33 |
load_in_4bit: true
|
34 |
strict: false
|
@@ -39,27 +31,33 @@ datasets:
|
|
39 |
field_instruction: instruction
|
40 |
field_input: context
|
41 |
field_output: response
|
|
|
42 |
|
43 |
-
|
44 |
-
|
|
|
|
|
45 |
|
46 |
adapter: qlora
|
|
|
47 |
lora_r: 32
|
48 |
-
lora_alpha:
|
49 |
lora_dropout: 0.05
|
50 |
lora_target_linear: true
|
|
|
51 |
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
|
|
56 |
|
57 |
gradient_accumulation_steps: 4
|
58 |
-
micro_batch_size:
|
59 |
-
num_epochs:
|
60 |
optimizer: adamw_torch
|
61 |
lr_scheduler: cosine
|
62 |
-
learning_rate:
|
63 |
|
64 |
train_on_inputs: false
|
65 |
group_by_length: false
|
@@ -68,6 +66,8 @@ fp16:
|
|
68 |
tf32: true
|
69 |
|
70 |
gradient_checkpointing: true
|
|
|
|
|
71 |
early_stopping_patience:
|
72 |
resume_from_checkpoint:
|
73 |
local_rank:
|
@@ -75,16 +75,17 @@ logging_steps: 1
|
|
75 |
xformers_attention:
|
76 |
flash_attention: false
|
77 |
|
78 |
-
|
79 |
evals_per_epoch: 4
|
80 |
-
eval_max_new_tokens: 128
|
81 |
saves_per_epoch: 1
|
82 |
debug:
|
83 |
-
deepspeed: deepspeed_configs/
|
84 |
weight_decay: 0.0
|
85 |
-
|
86 |
fsdp:
|
87 |
fsdp_config:
|
|
|
|
|
|
|
88 |
|
89 |
```
|
90 |
|
@@ -92,9 +93,9 @@ fsdp_config:
|
|
92 |
|
93 |
# gemma-2-27b-it-dolly-15k
|
94 |
|
95 |
-
This model is a fine-tuned version of [google/gemma-2-27b-it](https://huggingface.co/google/gemma-2-27b-it) on the
|
96 |
It achieves the following results on the evaluation set:
|
97 |
-
- Loss: 1.
|
98 |
|
99 |
## Model description
|
100 |
|
@@ -113,42 +114,34 @@ More information needed
|
|
113 |
### Training hyperparameters
|
114 |
|
115 |
The following hyperparameters were used during training:
|
116 |
-
- learning_rate:
|
117 |
-
- train_batch_size:
|
118 |
-
- eval_batch_size:
|
119 |
- seed: 42
|
120 |
- distributed_type: multi-GPU
|
121 |
- num_devices: 2
|
122 |
- gradient_accumulation_steps: 4
|
123 |
-
- total_train_batch_size:
|
124 |
-
- total_eval_batch_size:
|
125 |
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
126 |
- lr_scheduler_type: cosine
|
127 |
-
- lr_scheduler_warmup_steps:
|
128 |
-
- num_epochs:
|
129 |
|
130 |
### Training results
|
131 |
|
132 |
| Training Loss | Epoch | Step | Validation Loss |
|
133 |
|:-------------:|:------:|:----:|:---------------:|
|
134 |
-
| 4.
|
135 |
-
|
|
136 |
-
| 1.
|
137 |
-
| 1.
|
138 |
-
| 1.7757 | 1.0129 | 156 | 1.5193 |
|
139 |
-
| 1.7768 | 1.2654 | 195 | 1.4965 |
|
140 |
-
| 1.3735 | 1.5178 | 234 | 1.4835 |
|
141 |
-
| 1.7285 | 1.7702 | 273 | 1.4744 |
|
142 |
-
| 1.6601 | 2.0259 | 312 | 1.4701 |
|
143 |
-
| 1.6477 | 2.2783 | 351 | 1.4657 |
|
144 |
-
| 1.3795 | 2.5307 | 390 | 1.4645 |
|
145 |
-
| 1.6575 | 2.7832 | 429 | 1.4649 |
|
146 |
|
147 |
|
148 |
### Framework versions
|
149 |
|
150 |
-
- PEFT 0.
|
151 |
-
- Transformers 4.
|
152 |
-
- Pytorch 2.
|
153 |
- Datasets 3.1.0
|
154 |
-
- Tokenizers 0.
|
|
|
5 |
tags:
|
6 |
- axolotl
|
7 |
- generated_from_trainer
|
|
|
|
|
8 |
model-index:
|
9 |
- name: gemma-2-27b-it-dolly-15k
|
10 |
results: []
|
|
|
16 |
[<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)
|
17 |
<details><summary>See axolotl config</summary>
|
18 |
|
19 |
+
axolotl version: `0.5.2`
|
20 |
```yaml
|
21 |
base_model: google/gemma-2-27b-it
|
|
|
|
|
22 |
hub_model_id: kweinmeister/gemma-2-27b-it-dolly-15k
|
23 |
|
|
|
|
|
|
|
|
|
24 |
load_in_8bit: false
|
25 |
load_in_4bit: true
|
26 |
strict: false
|
|
|
31 |
field_instruction: instruction
|
32 |
field_input: context
|
33 |
field_output: response
|
34 |
+
val_set_size: 0.05
|
35 |
|
36 |
+
sequence_len: 2048
|
37 |
+
sample_packing: true
|
38 |
+
eval_sample_packing: false
|
39 |
+
pad_to_sequence_len: true
|
40 |
|
41 |
adapter: qlora
|
42 |
+
lora_model_dir:
|
43 |
lora_r: 32
|
44 |
+
lora_alpha: 64
|
45 |
lora_dropout: 0.05
|
46 |
lora_target_linear: true
|
47 |
+
lora_fan_in_fan_out:
|
48 |
|
49 |
+
wandb_project: gemma-2-27b-it-dolly-15k
|
50 |
+
wandb_entity:
|
51 |
+
wandb_watch:
|
52 |
+
wandb_name:
|
53 |
+
wandb_log_model:
|
54 |
|
55 |
gradient_accumulation_steps: 4
|
56 |
+
micro_batch_size: 4
|
57 |
+
num_epochs: 1
|
58 |
optimizer: adamw_torch
|
59 |
lr_scheduler: cosine
|
60 |
+
learning_rate: 0.0001
|
61 |
|
62 |
train_on_inputs: false
|
63 |
group_by_length: false
|
|
|
66 |
tf32: true
|
67 |
|
68 |
gradient_checkpointing: true
|
69 |
+
gradient_checkpointing_kwargs:
|
70 |
+
use_reentrant: true
|
71 |
early_stopping_patience:
|
72 |
resume_from_checkpoint:
|
73 |
local_rank:
|
|
|
75 |
xformers_attention:
|
76 |
flash_attention: false
|
77 |
|
78 |
+
warmup_steps: 10
|
79 |
evals_per_epoch: 4
|
|
|
80 |
saves_per_epoch: 1
|
81 |
debug:
|
82 |
+
deepspeed: deepspeed_configs/zero2.json
|
83 |
weight_decay: 0.0
|
|
|
84 |
fsdp:
|
85 |
fsdp_config:
|
86 |
+
special_tokens:
|
87 |
+
output_dir: "/mnt/disks/gcs/training/runs/google--gemma-2-27b-it-20250101-192231/out/"
|
88 |
+
dataset_prepared_path: "/mnt/disks/gcs/training/datasets"
|
89 |
|
90 |
```
|
91 |
|
|
|
93 |
|
94 |
# gemma-2-27b-it-dolly-15k
|
95 |
|
96 |
+
This model is a fine-tuned version of [google/gemma-2-27b-it](https://huggingface.co/google/gemma-2-27b-it) on the None dataset.
|
97 |
It achieves the following results on the evaluation set:
|
98 |
+
- Loss: 1.5560
|
99 |
|
100 |
## Model description
|
101 |
|
|
|
114 |
### Training hyperparameters
|
115 |
|
116 |
The following hyperparameters were used during training:
|
117 |
+
- learning_rate: 0.0001
|
118 |
+
- train_batch_size: 4
|
119 |
+
- eval_batch_size: 4
|
120 |
- seed: 42
|
121 |
- distributed_type: multi-GPU
|
122 |
- num_devices: 2
|
123 |
- gradient_accumulation_steps: 4
|
124 |
+
- total_train_batch_size: 32
|
125 |
+
- total_eval_batch_size: 8
|
126 |
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
127 |
- lr_scheduler_type: cosine
|
128 |
+
- lr_scheduler_warmup_steps: 10
|
129 |
+
- num_epochs: 1
|
130 |
|
131 |
### Training results
|
132 |
|
133 |
| Training Loss | Epoch | Step | Validation Loss |
|
134 |
|:-------------:|:------:|:----:|:---------------:|
|
135 |
+
| 4.2291 | 0.0244 | 1 | 2.1246 |
|
136 |
+
| 2.1928 | 0.2683 | 11 | 1.6858 |
|
137 |
+
| 1.742 | 0.5366 | 22 | 1.5769 |
|
138 |
+
| 1.7213 | 0.8049 | 33 | 1.5560 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
139 |
|
140 |
|
141 |
### Framework versions
|
142 |
|
143 |
+
- PEFT 0.13.2
|
144 |
+
- Transformers 4.46.3
|
145 |
+
- Pytorch 2.4.1+cu124
|
146 |
- Datasets 3.1.0
|
147 |
+
- Tokenizers 0.20.3
|
adapter_model.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 456822394
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:34bb1599f6a859b0c63f13428fd8d11df5781227d292a053bffadb108b5fa623
|
3 |
size 456822394
|