diabolic6045
commited on
Commit
•
ec25aaa
1
Parent(s):
bac728e
End of training
Browse files- README.md +14 -22
- adapter_model.bin +1 -1
README.md
CHANGED
@@ -1,21 +1,19 @@
|
|
1 |
---
|
2 |
-
|
3 |
library_name: peft
|
|
|
4 |
tags:
|
5 |
- axolotl
|
6 |
- generated_from_trainer
|
7 |
-
base_model: meta-llama/Meta-Llama-3-8B
|
8 |
model-index:
|
9 |
- name: Sanskrit-llama
|
10 |
results: []
|
11 |
-
datasets:
|
12 |
-
- diabolic6045/Sanskrit-llama
|
13 |
---
|
14 |
|
15 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
16 |
should probably proofread and complete it, then remove this comment. -->
|
17 |
|
18 |
-
[<img src="https://raw.githubusercontent.com/
|
19 |
<details><summary>See axolotl config</summary>
|
20 |
|
21 |
axolotl version: `0.4.1`
|
@@ -24,7 +22,7 @@ axolotl version: `0.4.1`
|
|
24 |
base_model: meta-llama/Meta-Llama-3-8B
|
25 |
model_type: AutoModelForCausalLM
|
26 |
tokenizer_type: AutoTokenizer
|
27 |
-
max_steps:
|
28 |
bnb_config_kwargs:
|
29 |
llm_int8_has_fp16_weight: false
|
30 |
bnb_4bit_quant_type: nf4
|
@@ -35,7 +33,7 @@ load_in_4bit: true
|
|
35 |
strict: false
|
36 |
|
37 |
datasets:
|
38 |
-
- path:
|
39 |
type: alpaca
|
40 |
dataset_prepared_path:
|
41 |
val_set_size: 0
|
@@ -46,7 +44,7 @@ hf_use_auth_token: true
|
|
46 |
adapter: qlora
|
47 |
lora_model_dir:
|
48 |
|
49 |
-
sequence_len:
|
50 |
sample_packing: true
|
51 |
eval_sample_packing: false
|
52 |
pad_to_sequence_len: true
|
@@ -56,21 +54,15 @@ lora_alpha: 16
|
|
56 |
lora_dropout: 0.05
|
57 |
lora_target_modules:
|
58 |
lora_target_linear: true
|
59 |
-
lora_fan_in_fan_out:
|
60 |
-
|
61 |
-
wandb_project: संस्कृतम्-llama
|
62 |
-
wandb_entity:
|
63 |
-
wandb_watch: all
|
64 |
-
wandb_name: संस्कृतम्-llama
|
65 |
-
wandb_log_model:
|
66 |
|
67 |
-
gradient_accumulation_steps:
|
68 |
micro_batch_size: 2
|
69 |
num_epochs: 1
|
70 |
optimizer: paged_adamw_8bit
|
71 |
lr_scheduler: cosine
|
72 |
cosine_min_lr_ratio: 0.2
|
73 |
-
learning_rate:
|
74 |
|
75 |
train_on_inputs: false
|
76 |
group_by_length: false
|
@@ -132,19 +124,19 @@ More information needed
|
|
132 |
### Training hyperparameters
|
133 |
|
134 |
The following hyperparameters were used during training:
|
135 |
-
- learning_rate:
|
136 |
- train_batch_size: 2
|
137 |
- eval_batch_size: 2
|
138 |
- seed: 42
|
139 |
- distributed_type: multi-GPU
|
140 |
- num_devices: 2
|
141 |
-
- gradient_accumulation_steps:
|
142 |
-
- total_train_batch_size:
|
143 |
- total_eval_batch_size: 4
|
144 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
145 |
- lr_scheduler_type: cosine
|
146 |
- lr_scheduler_warmup_steps: 10
|
147 |
-
-
|
148 |
|
149 |
### Training results
|
150 |
|
@@ -153,7 +145,7 @@ The following hyperparameters were used during training:
|
|
153 |
### Framework versions
|
154 |
|
155 |
- PEFT 0.11.1
|
156 |
-
- Transformers 4.
|
157 |
- Pytorch 2.1.2
|
158 |
- Datasets 2.19.1
|
159 |
- Tokenizers 0.19.1
|
|
|
1 |
---
|
2 |
+
base_model: meta-llama/Meta-Llama-3-8B
|
3 |
library_name: peft
|
4 |
+
license: llama3
|
5 |
tags:
|
6 |
- axolotl
|
7 |
- generated_from_trainer
|
|
|
8 |
model-index:
|
9 |
- name: Sanskrit-llama
|
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/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: `0.4.1`
|
|
|
22 |
base_model: meta-llama/Meta-Llama-3-8B
|
23 |
model_type: AutoModelForCausalLM
|
24 |
tokenizer_type: AutoTokenizer
|
25 |
+
max_steps:
|
26 |
bnb_config_kwargs:
|
27 |
llm_int8_has_fp16_weight: false
|
28 |
bnb_4bit_quant_type: nf4
|
|
|
33 |
strict: false
|
34 |
|
35 |
datasets:
|
36 |
+
- path: VinitT/Sanskrit-Llama_Base-Dataset
|
37 |
type: alpaca
|
38 |
dataset_prepared_path:
|
39 |
val_set_size: 0
|
|
|
44 |
adapter: qlora
|
45 |
lora_model_dir:
|
46 |
|
47 |
+
sequence_len: 512
|
48 |
sample_packing: true
|
49 |
eval_sample_packing: false
|
50 |
pad_to_sequence_len: true
|
|
|
54 |
lora_dropout: 0.05
|
55 |
lora_target_modules:
|
56 |
lora_target_linear: true
|
57 |
+
lora_fan_in_fan_out:
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
|
59 |
+
gradient_accumulation_steps: 8
|
60 |
micro_batch_size: 2
|
61 |
num_epochs: 1
|
62 |
optimizer: paged_adamw_8bit
|
63 |
lr_scheduler: cosine
|
64 |
cosine_min_lr_ratio: 0.2
|
65 |
+
learning_rate: 5e-5
|
66 |
|
67 |
train_on_inputs: false
|
68 |
group_by_length: false
|
|
|
124 |
### Training hyperparameters
|
125 |
|
126 |
The following hyperparameters were used during training:
|
127 |
+
- learning_rate: 5e-05
|
128 |
- train_batch_size: 2
|
129 |
- eval_batch_size: 2
|
130 |
- seed: 42
|
131 |
- distributed_type: multi-GPU
|
132 |
- num_devices: 2
|
133 |
+
- gradient_accumulation_steps: 8
|
134 |
+
- total_train_batch_size: 32
|
135 |
- total_eval_batch_size: 4
|
136 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
137 |
- lr_scheduler_type: cosine
|
138 |
- lr_scheduler_warmup_steps: 10
|
139 |
+
- num_epochs: 1
|
140 |
|
141 |
### Training results
|
142 |
|
|
|
145 |
### Framework versions
|
146 |
|
147 |
- PEFT 0.11.1
|
148 |
+
- Transformers 4.42.3
|
149 |
- Pytorch 2.1.2
|
150 |
- Datasets 2.19.1
|
151 |
- Tokenizers 0.19.1
|
adapter_model.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 167843194
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d59031ba061534bc251ce97171f3d11833bed818b498853a2f16aa29c16509f7
|
3 |
size 167843194
|