hugodk-sch
commited on
Commit
•
260ebb6
1
Parent(s):
4d3d718
Training in progress, step 400
Browse files- README.md +18 -30
- adapter_config.json +5 -5
- adapter_model.safetensors +1 -1
- all_results.json +17 -17
- config.json +1 -3
- eval_results.json +13 -13
- train_results.json +5 -5
- trainer_state.json +0 -0
- training_args.bin +1 -1
README.md
CHANGED
@@ -20,15 +20,15 @@ should probably proofread and complete it, then remove this comment. -->
|
|
20 |
|
21 |
This model is a fine-tuned version of [data/ap-gpt-j-6b-sft-qlora-04-08](https://huggingface.co/data/ap-gpt-j-6b-sft-qlora-04-08) on the hugodk-sch/aftonposten_title_prefs dataset.
|
22 |
It achieves the following results on the evaluation set:
|
23 |
-
- Loss: 0.
|
24 |
-
- Rewards/chosen: -0.
|
25 |
-
- Rewards/rejected: -0.
|
26 |
-
- Rewards/accuracies: 0.
|
27 |
-
- Rewards/margins: 0.
|
28 |
-
- Logps/rejected: -37.
|
29 |
-
- Logps/chosen: -34.
|
30 |
-
- Logits/rejected: -2.
|
31 |
-
- Logits/chosen: -2.
|
32 |
|
33 |
## Model description
|
34 |
|
@@ -57,33 +57,21 @@ The following hyperparameters were used during training:
|
|
57 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
58 |
- lr_scheduler_type: cosine
|
59 |
- lr_scheduler_warmup_ratio: 0.1
|
60 |
-
- num_epochs:
|
61 |
|
62 |
### Training results
|
63 |
|
64 |
-
| Training Loss | Epoch | Step |
|
65 |
-
|
66 |
-
| 0.
|
67 |
-
| 0.
|
68 |
-
| 0.
|
69 |
-
| 0.4664 | 1.04 | 400 | 0.9523 | -0.0303 | -0.0783 | 0.5806 | 0.0480 | -37.6732 | -34.0951 | -2.1923 | -2.1971 |
|
70 |
-
| 0.4189 | 1.3 | 500 | 0.9630 | -0.1072 | -0.1451 | 0.5631 | 0.0379 | -37.8069 | -34.2490 | -2.1893 | -2.1942 |
|
71 |
-
| 0.4171 | 1.56 | 600 | 0.9263 | -0.0581 | -0.1338 | 0.6009 | 0.0756 | -37.7842 | -34.1508 | -2.1730 | -2.1778 |
|
72 |
-
| 0.3491 | 1.82 | 700 | 0.9356 | -0.0720 | -0.1392 | 0.5627 | 0.0672 | -37.7950 | -34.1785 | -2.1634 | -2.1682 |
|
73 |
-
| 0.1693 | 2.08 | 800 | 0.9333 | -0.0958 | -0.1675 | 0.5831 | 0.0717 | -37.8517 | -34.2261 | -2.1557 | -2.1604 |
|
74 |
-
| 0.1819 | 2.34 | 900 | 0.9235 | -0.0902 | -0.1716 | 0.5947 | 0.0814 | -37.8598 | -34.2150 | -2.1563 | -2.1611 |
|
75 |
-
| 0.2378 | 2.6 | 1000 | 0.9275 | -0.1017 | -0.1825 | 0.6009 | 0.0808 | -37.8816 | -34.2380 | -2.1548 | -2.1595 |
|
76 |
-
| 0.1549 | 2.86 | 1100 | 0.9389 | -0.0965 | -0.1634 | 0.5631 | 0.0669 | -37.8434 | -34.2276 | -2.1554 | -2.1602 |
|
77 |
-
| 0.2405 | 3.12 | 1200 | 0.9305 | -0.0919 | -0.1673 | 0.5806 | 0.0754 | -37.8512 | -34.2183 | -2.1559 | -2.1606 |
|
78 |
-
| 0.1193 | 3.38 | 1300 | 0.9316 | -0.0960 | -0.1704 | 0.5922 | 0.0744 | -37.8575 | -34.2266 | -2.1563 | -2.1610 |
|
79 |
-
| 0.0703 | 3.64 | 1400 | 0.9304 | -0.0943 | -0.1698 | 0.5951 | 0.0755 | -37.8562 | -34.2232 | -2.1561 | -2.1608 |
|
80 |
-
| 0.1741 | 3.9 | 1500 | 0.9256 | -0.0884 | -0.1696 | 0.5860 | 0.0812 | -37.8558 | -34.2114 | -2.1560 | -2.1608 |
|
81 |
|
82 |
|
83 |
### Framework versions
|
84 |
|
85 |
-
- PEFT 0.
|
86 |
-
- Transformers 4.
|
87 |
- Pytorch 2.1.2+cu121
|
88 |
-
- Datasets 2.
|
89 |
- Tokenizers 0.15.1
|
|
|
20 |
|
21 |
This model is a fine-tuned version of [data/ap-gpt-j-6b-sft-qlora-04-08](https://huggingface.co/data/ap-gpt-j-6b-sft-qlora-04-08) on the hugodk-sch/aftonposten_title_prefs dataset.
|
22 |
It achieves the following results on the evaluation set:
|
23 |
+
- Loss: 0.9909
|
24 |
+
- Rewards/chosen: -0.0210
|
25 |
+
- Rewards/rejected: -0.0301
|
26 |
+
- Rewards/accuracies: 0.5129
|
27 |
+
- Rewards/margins: 0.0091
|
28 |
+
- Logps/rejected: -37.6169
|
29 |
+
- Logps/chosen: -34.1046
|
30 |
+
- Logits/rejected: -2.2201
|
31 |
+
- Logits/chosen: -2.2249
|
32 |
|
33 |
## Model description
|
34 |
|
|
|
57 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
58 |
- lr_scheduler_type: cosine
|
59 |
- lr_scheduler_warmup_ratio: 0.1
|
60 |
+
- num_epochs: 1
|
61 |
|
62 |
### Training results
|
63 |
|
64 |
+
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
65 |
+
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
66 |
+
| 0.956 | 0.26 | 100 | 0.9948 | -0.0045 | -0.0098 | 0.5631 | 0.0053 | -37.5492 | -34.0495 | -2.2276 | -2.2325 |
|
67 |
+
| 0.8922 | 0.52 | 200 | 0.9876 | -0.0094 | -0.0218 | 0.5303 | 0.0124 | -37.5894 | -34.0659 | -2.2231 | -2.2279 |
|
68 |
+
| 0.8033 | 0.78 | 300 | 0.9891 | -0.0205 | -0.0313 | 0.5183 | 0.0108 | -37.6211 | -34.1029 | -2.2202 | -2.2250 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
|
71 |
### Framework versions
|
72 |
|
73 |
+
- PEFT 0.8.2
|
74 |
+
- Transformers 4.37.2
|
75 |
- Pytorch 2.1.2+cu121
|
76 |
+
- Datasets 2.17.0
|
77 |
- Tokenizers 0.15.1
|
adapter_config.json
CHANGED
@@ -20,13 +20,13 @@
|
|
20 |
"rank_pattern": {},
|
21 |
"revision": null,
|
22 |
"target_modules": [
|
23 |
-
"down_proj",
|
24 |
-
"o_proj",
|
25 |
-
"gate_proj",
|
26 |
"up_proj",
|
27 |
"q_proj",
|
28 |
-
"
|
29 |
-
"
|
|
|
|
|
|
|
30 |
],
|
31 |
"task_type": "CAUSAL_LM",
|
32 |
"use_dora": false,
|
|
|
20 |
"rank_pattern": {},
|
21 |
"revision": null,
|
22 |
"target_modules": [
|
|
|
|
|
|
|
23 |
"up_proj",
|
24 |
"q_proj",
|
25 |
+
"k_proj",
|
26 |
+
"o_proj",
|
27 |
+
"down_proj",
|
28 |
+
"gate_proj",
|
29 |
+
"v_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:
|
3 |
size 176183216
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:5ce54f3458b304bfbc5be236eb4200b649350791bcd8613d0bde2973e3fd324c
|
3 |
size 176183216
|
all_results.json
CHANGED
@@ -1,21 +1,21 @@
|
|
1 |
{
|
2 |
-
"epoch":
|
3 |
-
"eval_logits/chosen": -2.
|
4 |
-
"eval_logits/rejected": -2.
|
5 |
-
"eval_logps/chosen": -34.
|
6 |
-
"eval_logps/rejected": -37.
|
7 |
-
"eval_loss": 0.
|
8 |
-
"eval_rewards/accuracies": 0.
|
9 |
-
"eval_rewards/chosen": -0.
|
10 |
-
"eval_rewards/margins": 0.
|
11 |
-
"eval_rewards/rejected": -0.
|
12 |
-
"eval_runtime": 145.
|
13 |
"eval_samples": 343,
|
14 |
-
"eval_samples_per_second": 2.
|
15 |
-
"eval_steps_per_second": 0.
|
16 |
-
"train_loss": 0.
|
17 |
-
"train_runtime":
|
18 |
"train_samples": 3079,
|
19 |
-
"train_samples_per_second":
|
20 |
-
"train_steps_per_second": 0.
|
21 |
}
|
|
|
1 |
{
|
2 |
+
"epoch": 1.0,
|
3 |
+
"eval_logits/chosen": -2.2249207496643066,
|
4 |
+
"eval_logits/rejected": -2.2200889587402344,
|
5 |
+
"eval_logps/chosen": -34.104576110839844,
|
6 |
+
"eval_logps/rejected": -37.61691665649414,
|
7 |
+
"eval_loss": 0.9908739328384399,
|
8 |
+
"eval_rewards/accuracies": 0.5128737688064575,
|
9 |
+
"eval_rewards/chosen": -0.02100813202559948,
|
10 |
+
"eval_rewards/margins": 0.009080832824110985,
|
11 |
+
"eval_rewards/rejected": -0.030088962987065315,
|
12 |
+
"eval_runtime": 145.588,
|
13 |
"eval_samples": 343,
|
14 |
+
"eval_samples_per_second": 2.356,
|
15 |
+
"eval_steps_per_second": 0.295,
|
16 |
+
"train_loss": 0.9217340717067967,
|
17 |
+
"train_runtime": 3251.2729,
|
18 |
"train_samples": 3079,
|
19 |
+
"train_samples_per_second": 0.947,
|
20 |
+
"train_steps_per_second": 0.118
|
21 |
}
|
config.json
CHANGED
@@ -18,8 +18,6 @@
|
|
18 |
"n_layer": 28,
|
19 |
"n_positions": 2048,
|
20 |
"quantization_config": {
|
21 |
-
"_load_in_4bit": true,
|
22 |
-
"_load_in_8bit": false,
|
23 |
"bnb_4bit_compute_dtype": "float16",
|
24 |
"bnb_4bit_quant_type": "nf4",
|
25 |
"bnb_4bit_use_double_quant": false,
|
@@ -47,7 +45,7 @@
|
|
47 |
},
|
48 |
"tie_word_embeddings": false,
|
49 |
"tokenizer_class": "GPT2Tokenizer",
|
50 |
-
"transformers_version": "4.
|
51 |
"use_cache": true,
|
52 |
"vocab_size": 50400
|
53 |
}
|
|
|
18 |
"n_layer": 28,
|
19 |
"n_positions": 2048,
|
20 |
"quantization_config": {
|
|
|
|
|
21 |
"bnb_4bit_compute_dtype": "float16",
|
22 |
"bnb_4bit_quant_type": "nf4",
|
23 |
"bnb_4bit_use_double_quant": false,
|
|
|
45 |
},
|
46 |
"tie_word_embeddings": false,
|
47 |
"tokenizer_class": "GPT2Tokenizer",
|
48 |
+
"transformers_version": "4.37.2",
|
49 |
"use_cache": true,
|
50 |
"vocab_size": 50400
|
51 |
}
|
eval_results.json
CHANGED
@@ -1,16 +1,16 @@
|
|
1 |
{
|
2 |
-
"epoch":
|
3 |
-
"eval_logits/chosen": -2.
|
4 |
-
"eval_logits/rejected": -2.
|
5 |
-
"eval_logps/chosen": -34.
|
6 |
-
"eval_logps/rejected": -37.
|
7 |
-
"eval_loss": 0.
|
8 |
-
"eval_rewards/accuracies": 0.
|
9 |
-
"eval_rewards/chosen": -0.
|
10 |
-
"eval_rewards/margins": 0.
|
11 |
-
"eval_rewards/rejected": -0.
|
12 |
-
"eval_runtime": 145.
|
13 |
"eval_samples": 343,
|
14 |
-
"eval_samples_per_second": 2.
|
15 |
-
"eval_steps_per_second": 0.
|
16 |
}
|
|
|
1 |
{
|
2 |
+
"epoch": 1.0,
|
3 |
+
"eval_logits/chosen": -2.2249207496643066,
|
4 |
+
"eval_logits/rejected": -2.2200889587402344,
|
5 |
+
"eval_logps/chosen": -34.104576110839844,
|
6 |
+
"eval_logps/rejected": -37.61691665649414,
|
7 |
+
"eval_loss": 0.9908739328384399,
|
8 |
+
"eval_rewards/accuracies": 0.5128737688064575,
|
9 |
+
"eval_rewards/chosen": -0.02100813202559948,
|
10 |
+
"eval_rewards/margins": 0.009080832824110985,
|
11 |
+
"eval_rewards/rejected": -0.030088962987065315,
|
12 |
+
"eval_runtime": 145.588,
|
13 |
"eval_samples": 343,
|
14 |
+
"eval_samples_per_second": 2.356,
|
15 |
+
"eval_steps_per_second": 0.295
|
16 |
}
|
train_results.json
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
{
|
2 |
-
"epoch":
|
3 |
-
"train_loss": 0.
|
4 |
-
"train_runtime":
|
5 |
"train_samples": 3079,
|
6 |
-
"train_samples_per_second":
|
7 |
-
"train_steps_per_second": 0.
|
8 |
}
|
|
|
1 |
{
|
2 |
+
"epoch": 1.0,
|
3 |
+
"train_loss": 0.9217340717067967,
|
4 |
+
"train_runtime": 3251.2729,
|
5 |
"train_samples": 3079,
|
6 |
+
"train_samples_per_second": 0.947,
|
7 |
+
"train_steps_per_second": 0.118
|
8 |
}
|
trainer_state.json
CHANGED
The diff for this file is too large to render.
See raw diff
|
|
training_args.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 5176
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:03f767c74f345697032eed65875a78939878c73ae7049f800d864f7bd6bbcfac
|
3 |
size 5176
|