tuanna08go commited on
Commit
f104b64
·
verified ·
1 Parent(s): 414d3ab

End of training

Browse files
Files changed (2) hide show
  1. README.md +11 -4
  2. adapter_model.bin +1 -1
README.md CHANGED
@@ -65,7 +65,7 @@ lora_model_dir: null
65
  lora_r: 8
66
  lora_target_linear: true
67
  lr_scheduler: cosine
68
- max_steps: 1
69
  micro_batch_size: 2
70
  mlflow_experiment_name: /tmp/c004c75e3599110f_train_data.json
71
  model_type: AutoModelForCausalLM
@@ -92,7 +92,7 @@ wandb_name: ebfa63ec-b835-4714-bd4e-ef9e45a9563e
92
  wandb_project: Gradients-On-Demand
93
  wandb_run: your_name
94
  wandb_runid: ebfa63ec-b835-4714-bd4e-ef9e45a9563e
95
- warmup_steps: 1
96
  weight_decay: 0.0
97
  xformers_attention: null
98
 
@@ -103,6 +103,8 @@ xformers_attention: null
103
  # f7f023be-e950-0d39-6318-3f875dd8bee5
104
 
105
  This model is a fine-tuned version of [JackFram/llama-160m](https://huggingface.co/JackFram/llama-160m) on the None dataset.
 
 
106
 
107
  ## Model description
108
 
@@ -129,14 +131,19 @@ The following hyperparameters were used during training:
129
  - total_train_batch_size: 8
130
  - optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
131
  - lr_scheduler_type: cosine
132
- - lr_scheduler_warmup_steps: 2
133
- - training_steps: 1
134
 
135
  ### Training results
136
 
137
  | Training Loss | Epoch | Step | Validation Loss |
138
  |:-------------:|:------:|:----:|:---------------:|
139
  | No log | 0.0004 | 1 | 3.4804 |
 
 
 
 
 
140
 
141
 
142
  ### Framework versions
 
65
  lora_r: 8
66
  lora_target_linear: true
67
  lr_scheduler: cosine
68
+ max_steps: 50
69
  micro_batch_size: 2
70
  mlflow_experiment_name: /tmp/c004c75e3599110f_train_data.json
71
  model_type: AutoModelForCausalLM
 
92
  wandb_project: Gradients-On-Demand
93
  wandb_run: your_name
94
  wandb_runid: ebfa63ec-b835-4714-bd4e-ef9e45a9563e
95
+ warmup_steps: 10
96
  weight_decay: 0.0
97
  xformers_attention: null
98
 
 
103
  # f7f023be-e950-0d39-6318-3f875dd8bee5
104
 
105
  This model is a fine-tuned version of [JackFram/llama-160m](https://huggingface.co/JackFram/llama-160m) on the None dataset.
106
+ It achieves the following results on the evaluation set:
107
+ - Loss: 3.1211
108
 
109
  ## Model description
110
 
 
131
  - total_train_batch_size: 8
132
  - optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
133
  - lr_scheduler_type: cosine
134
+ - lr_scheduler_warmup_steps: 10
135
+ - training_steps: 50
136
 
137
  ### Training results
138
 
139
  | Training Loss | Epoch | Step | Validation Loss |
140
  |:-------------:|:------:|:----:|:---------------:|
141
  | No log | 0.0004 | 1 | 3.4804 |
142
+ | 3.4641 | 0.0043 | 10 | 3.4382 |
143
+ | 3.2662 | 0.0086 | 20 | 3.3009 |
144
+ | 3.2848 | 0.0128 | 30 | 3.1940 |
145
+ | 3.1459 | 0.0171 | 40 | 3.1329 |
146
+ | 3.195 | 0.0214 | 50 | 3.1211 |
147
 
148
 
149
  ### Framework versions
adapter_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:545f9e3befcbefedb715699554941c9416a8df10fd9024d38e09da98a181ce91
3
  size 6843306
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:521bfe348d54d66ad6231ad8338395652c3397ad2a29d7d56e403e942c4b7e87
3
  size 6843306