Hugging Face's logo

jrahn
/
llama-3-8b-claudstruct-v2

PEFT
Safetensors
English
llama
Generated from Trainer
4-bit precision
bitsandbytes
Model card Files
xet
 Community
jrahn commited on
Commit
ee63a15
·
verified ·
1 Parent(s): ef6161c

Upload folder using huggingface_hub

Browse files
Files changed (4) hide show
  1. README.md +15 -19
  2. adapter_config.json +6 -6
  3. adapter_model.safetensors +1 -1
  4. training_args.bin +1 -1
README.md CHANGED
@@ -5,7 +5,7 @@ tags:
5
  - generated_from_trainer
6
  base_model: meta-llama/Meta-Llama-3-8B-Instruct
7
  model-index:
8
- - name: outputs/llama-3-8b-claudstruct-v3/
9
  results: []
10
  ---
11
 
@@ -31,7 +31,7 @@ datasets:
31
  type: alpaca
32
  dataset_prepared_path: last_run_prepared
33
  val_set_size: 0.05
34
- output_dir: ./outputs/llama-3-8b-claudstruct-v3/
35
 
36
  adapter: qlora
37
  lora_model_dir:
@@ -54,8 +54,8 @@ wandb_name:
54
  wandb_log_model:
55
 
56
  gradient_accumulation_steps: 1
57
- micro_batch_size: 8
58
- num_epochs: 2
59
  optimizer: adamw_torch
60
  lr_scheduler: cosine
61
  learning_rate: 0.00001
@@ -103,11 +103,11 @@ special_tokens:
103
 
104
  </details><br>
105
 
106
- # outputs/llama-3-8b-claudstruct-v3/
107
 
108
  This model is a fine-tuned version of [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) on the None dataset.
109
  It achieves the following results on the evaluation set:
110
- - Loss: 1.6226
111
 
112
  ## Model description
113
 
@@ -127,30 +127,26 @@ More information needed
127
 
128
  The following hyperparameters were used during training:
129
  - learning_rate: 1e-05
130
- - train_batch_size: 8
131
- - eval_batch_size: 8
132
  - seed: 42
133
  - distributed_type: multi-GPU
134
  - num_devices: 2
135
- - total_train_batch_size: 16
136
- - total_eval_batch_size: 16
137
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
138
  - lr_scheduler_type: cosine
139
  - lr_scheduler_warmup_steps: 10
140
- - num_epochs: 2
141
 
142
  ### Training results
143
 
144
  | Training Loss | Epoch | Step | Validation Loss |
145
  |:-------------:|:------:|:----:|:---------------:|
146
- | 2.2209 | 0.0007 | 1 | 2.0399 |
147
- | 1.7842 | 0.2502 | 341 | 1.6960 |
148
- | 1.6914 | 0.5004 | 682 | 1.6590 |
149
- | 1.6757 | 0.7506 | 1023 | 1.6414 |
150
- | 1.5182 | 1.0007 | 1364 | 1.6319 |
151
- | 1.8421 | 1.2509 | 1705 | 1.6264 |
152
- | 1.7271 | 1.5011 | 2046 | 1.6237 |
153
- | 1.4817 | 1.7513 | 2387 | 1.6226 |
154
 
155
 
156
  ### Framework versions
 
5
  - generated_from_trainer
6
  base_model: meta-llama/Meta-Llama-3-8B-Instruct
7
  model-index:
8
+ - name: outputs/llama-3-8b-claudstruct-v2/
9
  results: []
10
  ---
11
 
 
31
  type: alpaca
32
  dataset_prepared_path: last_run_prepared
33
  val_set_size: 0.05
34
+ output_dir: ./outputs/llama-3-8b-claudstruct-v2/
35
 
36
  adapter: qlora
37
  lora_model_dir:
 
54
  wandb_log_model:
55
 
56
  gradient_accumulation_steps: 1
57
+ micro_batch_size: 16
58
+ num_epochs: 1
59
  optimizer: adamw_torch
60
  lr_scheduler: cosine
61
  learning_rate: 0.00001
 
103
 
104
  </details><br>
105
 
106
+ # outputs/llama-3-8b-claudstruct-v2/
107
 
108
  This model is a fine-tuned version of [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) on the None dataset.
109
  It achieves the following results on the evaluation set:
110
+ - Loss: 1.6839
111
 
112
  ## Model description
113
 
 
127
 
128
  The following hyperparameters were used during training:
129
  - learning_rate: 1e-05
130
+ - train_batch_size: 16
131
+ - eval_batch_size: 16
132
  - seed: 42
133
  - distributed_type: multi-GPU
134
  - num_devices: 2
135
+ - total_train_batch_size: 32
136
+ - total_eval_batch_size: 32
137
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
138
  - lr_scheduler_type: cosine
139
  - lr_scheduler_warmup_steps: 10
140
+ - num_epochs: 1
141
 
142
  ### Training results
143
 
144
  | Training Loss | Epoch | Step | Validation Loss |
145
  |:-------------:|:------:|:----:|:---------------:|
146
+ | 2.0639 | 0.0015 | 1 | 2.0395 |
147
+ | 1.7905 | 0.2507 | 171 | 1.7402 |
148
+ | 1.8968 | 0.5015 | 342 | 1.6960 |
149
+ | 1.6319 | 0.7522 | 513 | 1.6839 |
 
 
 
 
150
 
151
 
152
  ### Framework versions
adapter_config.json CHANGED
@@ -4,7 +4,7 @@
4
  "base_model_name_or_path": "meta-llama/Meta-Llama-3-8B-Instruct",
5
  "bias": "none",
6
  "fan_in_fan_out": null,
7
- "inference_mode": true,
8
  "init_lora_weights": true,
9
  "layer_replication": null,
10
  "layers_pattern": null,
@@ -20,13 +20,13 @@
20
  "rank_pattern": {},
21
  "revision": null,
22
  "target_modules": [
23
- "v_proj",
24
- "up_proj",
25
- "gate_proj",
26
  "down_proj",
 
 
27
  "o_proj",
28
- "k_proj",
29
- "q_proj"
 
30
  ],
31
  "task_type": "CAUSAL_LM",
32
  "use_dora": false,
 
4
  "base_model_name_or_path": "meta-llama/Meta-Llama-3-8B-Instruct",
5
  "bias": "none",
6
  "fan_in_fan_out": null,
7
+ "inference_mode": false,
8
  "init_lora_weights": true,
9
  "layer_replication": null,
10
  "layers_pattern": null,
 
20
  "rank_pattern": {},
21
  "revision": null,
22
  "target_modules": [
 
 
 
23
  "down_proj",
24
+ "up_proj",
25
+ "v_proj",
26
  "o_proj",
27
+ "gate_proj",
28
+ "q_proj",
29
+ "k_proj"
30
  ],
31
  "task_type": "CAUSAL_LM",
32
  "use_dora": false,
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:87a0ecab10e267e31e2d676cb9219607001f5b9c2e4496c5c3ded514352d594d
3
  size 83945296
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b6e0677afc12f9bd6f0287c17c067ffcab7db91f52132f729145124c8e0e043
3
  size 83945296
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:899f1662c0af7238f2da34f9cfa528d22053871728445d51705cf917a4b82a05
3
  size 6456
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:313ecaaf2979fa4223bfe586f355a60818fb1fb4b9d7c3ccb2f966acc3ae428c
3
  size 6456