Update cfg.yaml
Browse files
cfg.yaml
CHANGED
|
@@ -28,7 +28,7 @@ dataset:
|
|
| 28 |
text_answer_separator: <|answer|>
|
| 29 |
text_prompt_start: <|prompt|>
|
| 30 |
train_dataframe: data/user/oasst/train_full_allrank.pq
|
| 31 |
-
validation_dataframe: data/user/oasst/
|
| 32 |
validation_size: 0.01
|
| 33 |
validation_strategy: custom
|
| 34 |
environment:
|
|
@@ -39,7 +39,7 @@ environment:
|
|
| 39 |
- '1'
|
| 40 |
- '2'
|
| 41 |
- '3'
|
| 42 |
-
huggingface_branch:
|
| 43 |
mixed_precision: true
|
| 44 |
number_of_workers: 8
|
| 45 |
seed: -1
|
|
@@ -47,10 +47,6 @@ environment:
|
|
| 47 |
use_fsdp: false
|
| 48 |
experiment_name: h2ogpt-gm-oasst1-en-xgen-7b-8k
|
| 49 |
llm_backbone: Salesforce/xgen-7b-8k-base
|
| 50 |
-
logging:
|
| 51 |
-
logger: Neptune
|
| 52 |
-
neptune_project: Zoo/h2o-llm
|
| 53 |
-
number_of_texts: 10
|
| 54 |
output_directory: output/user/h2ogpt-gm-oasst1-en-xgen-7b-8k/
|
| 55 |
prediction:
|
| 56 |
batch_size_inference: 0
|
|
@@ -105,7 +101,6 @@ training:
|
|
| 105 |
ppo_clip_value: 0.2
|
| 106 |
ppo_epochs: 4
|
| 107 |
ppo_generate_temperature: 1.0
|
| 108 |
-
reward_model: OpenAssistant/reward-model-deberta-v3-large-v2
|
| 109 |
save_best_checkpoint: false
|
| 110 |
scaling_factor_value_loss: 0.1
|
| 111 |
schedule: Cosine
|
|
|
|
| 28 |
text_answer_separator: <|answer|>
|
| 29 |
text_prompt_start: <|prompt|>
|
| 30 |
train_dataframe: data/user/oasst/train_full_allrank.pq
|
| 31 |
+
validation_dataframe: data/user/oasst/val.csv
|
| 32 |
validation_size: 0.01
|
| 33 |
validation_strategy: custom
|
| 34 |
environment:
|
|
|
|
| 39 |
- '1'
|
| 40 |
- '2'
|
| 41 |
- '3'
|
| 42 |
+
huggingface_branch: main
|
| 43 |
mixed_precision: true
|
| 44 |
number_of_workers: 8
|
| 45 |
seed: -1
|
|
|
|
| 47 |
use_fsdp: false
|
| 48 |
experiment_name: h2ogpt-gm-oasst1-en-xgen-7b-8k
|
| 49 |
llm_backbone: Salesforce/xgen-7b-8k-base
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
output_directory: output/user/h2ogpt-gm-oasst1-en-xgen-7b-8k/
|
| 51 |
prediction:
|
| 52 |
batch_size_inference: 0
|
|
|
|
| 101 |
ppo_clip_value: 0.2
|
| 102 |
ppo_epochs: 4
|
| 103 |
ppo_generate_temperature: 1.0
|
|
|
|
| 104 |
save_best_checkpoint: false
|
| 105 |
scaling_factor_value_loss: 0.1
|
| 106 |
schedule: Cosine
|