Commit
•
b9d6f4f
1
Parent(s):
a2b2ff0
Model save
Browse files- README.md +14 -5
- adapter_config.json +1 -1
- adapter_model.safetensors +1 -1
- all_results.json +19 -6
- eval_results.json +16 -0
- train_results.json +6 -6
- trainer_state.json +200 -16
- training_args.bin +1 -1
README.md
CHANGED
@@ -2,14 +2,10 @@
|
|
2 |
license: mit
|
3 |
library_name: peft
|
4 |
tags:
|
5 |
-
- alignment-handbook
|
6 |
-
- generated_from_trainer
|
7 |
- trl
|
8 |
- dpo
|
9 |
- generated_from_trainer
|
10 |
base_model: microsoft/phi-2
|
11 |
-
datasets:
|
12 |
-
- HuggingFaceH4/ultrafeedback_binarized
|
13 |
model-index:
|
14 |
- name: phi-2-gpo-test-longest-iter-1
|
15 |
results: []
|
@@ -20,7 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
|
|
20 |
|
21 |
# phi-2-gpo-test-longest-iter-1
|
22 |
|
23 |
-
This model is a fine-tuned version of [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
25 |
## Model description
|
26 |
|
@@ -53,6 +59,9 @@ The following hyperparameters were used during training:
|
|
53 |
|
54 |
### Training results
|
55 |
|
|
|
|
|
|
|
56 |
|
57 |
|
58 |
### Framework versions
|
|
|
2 |
license: mit
|
3 |
library_name: peft
|
4 |
tags:
|
|
|
|
|
5 |
- trl
|
6 |
- dpo
|
7 |
- generated_from_trainer
|
8 |
base_model: microsoft/phi-2
|
|
|
|
|
9 |
model-index:
|
10 |
- name: phi-2-gpo-test-longest-iter-1
|
11 |
results: []
|
|
|
16 |
|
17 |
# phi-2-gpo-test-longest-iter-1
|
18 |
|
19 |
+
This model is a fine-tuned version of [microsoft/phi-2](https://huggingface.co/microsoft/phi-2) on the None dataset.
|
20 |
+
It achieves the following results on the evaluation set:
|
21 |
+
- Loss: 0.0108
|
22 |
+
- Rewards/chosen: -0.0007
|
23 |
+
- Rewards/rejected: -0.0005
|
24 |
+
- Rewards/accuracies: 0.4995
|
25 |
+
- Rewards/margins: -0.0002
|
26 |
+
- Logps/rejected: -278.7133
|
27 |
+
- Logps/chosen: -306.4168
|
28 |
+
- Logits/rejected: 0.0867
|
29 |
+
- Logits/chosen: -0.0107
|
30 |
|
31 |
## Model description
|
32 |
|
|
|
59 |
|
60 |
### Training results
|
61 |
|
62 |
+
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
63 |
+
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
64 |
+
| 0.0103 | 1.6 | 100 | 0.0108 | -0.0015 | -0.0014 | 0.5010 | -0.0001 | -278.8064 | -306.4953 | 0.0771 | -0.0200 |
|
65 |
|
66 |
|
67 |
### Framework versions
|
adapter_config.json
CHANGED
@@ -19,9 +19,9 @@
|
|
19 |
"rank_pattern": {},
|
20 |
"revision": null,
|
21 |
"target_modules": [
|
|
|
22 |
"q_proj",
|
23 |
"dense",
|
24 |
-
"k_proj",
|
25 |
"v_proj"
|
26 |
],
|
27 |
"task_type": "CAUSAL_LM"
|
|
|
19 |
"rank_pattern": {},
|
20 |
"revision": null,
|
21 |
"target_modules": [
|
22 |
+
"k_proj",
|
23 |
"q_proj",
|
24 |
"dense",
|
|
|
25 |
"v_proj"
|
26 |
],
|
27 |
"task_type": "CAUSAL_LM"
|
adapter_model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 41977616
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b8680c357a8e3e01ad9a1e0ad14ff6f46858eb570d2f8d9a60a7311cb045637c
|
3 |
size 41977616
|
all_results.json
CHANGED
@@ -1,8 +1,21 @@
|
|
1 |
{
|
2 |
-
"epoch": 1.
|
3 |
-
"
|
4 |
-
"
|
5 |
-
"
|
6 |
-
"
|
7 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
}
|
|
|
1 |
{
|
2 |
+
"epoch": 1.98,
|
3 |
+
"eval_logits/chosen": -0.010728351771831512,
|
4 |
+
"eval_logits/rejected": 0.08673479408025742,
|
5 |
+
"eval_logps/chosen": -306.4167785644531,
|
6 |
+
"eval_logps/rejected": -278.7133483886719,
|
7 |
+
"eval_loss": 0.01081769447773695,
|
8 |
+
"eval_rewards/accuracies": 0.49950000643730164,
|
9 |
+
"eval_rewards/chosen": -0.0007103218231350183,
|
10 |
+
"eval_rewards/margins": -0.00023564710863865912,
|
11 |
+
"eval_rewards/rejected": -0.00047467477270402014,
|
12 |
+
"eval_runtime": 470.0123,
|
13 |
+
"eval_samples": 2000,
|
14 |
+
"eval_samples_per_second": 4.255,
|
15 |
+
"eval_steps_per_second": 1.064,
|
16 |
+
"train_loss": 0.010314414998696696,
|
17 |
+
"train_runtime": 1522.7247,
|
18 |
+
"train_samples": 61135,
|
19 |
+
"train_samples_per_second": 1.313,
|
20 |
+
"train_steps_per_second": 0.081
|
21 |
}
|
eval_results.json
ADDED
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 1.98,
|
3 |
+
"eval_logits/chosen": -0.010728351771831512,
|
4 |
+
"eval_logits/rejected": 0.08673479408025742,
|
5 |
+
"eval_logps/chosen": -306.4167785644531,
|
6 |
+
"eval_logps/rejected": -278.7133483886719,
|
7 |
+
"eval_loss": 0.01081769447773695,
|
8 |
+
"eval_rewards/accuracies": 0.49950000643730164,
|
9 |
+
"eval_rewards/chosen": -0.0007103218231350183,
|
10 |
+
"eval_rewards/margins": -0.00023564710863865912,
|
11 |
+
"eval_rewards/rejected": -0.00047467477270402014,
|
12 |
+
"eval_runtime": 470.0123,
|
13 |
+
"eval_samples": 2000,
|
14 |
+
"eval_samples_per_second": 4.255,
|
15 |
+
"eval_steps_per_second": 1.064
|
16 |
+
}
|
train_results.json
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
{
|
2 |
-
"epoch": 1.
|
3 |
-
"train_loss": 0.
|
4 |
-
"train_runtime":
|
5 |
-
"train_samples":
|
6 |
-
"train_samples_per_second":
|
7 |
-
"train_steps_per_second": 0.
|
8 |
}
|
|
|
1 |
{
|
2 |
+
"epoch": 1.98,
|
3 |
+
"train_loss": 0.010314414998696696,
|
4 |
+
"train_runtime": 1522.7247,
|
5 |
+
"train_samples": 61135,
|
6 |
+
"train_samples_per_second": 1.313,
|
7 |
+
"train_steps_per_second": 0.081
|
8 |
}
|
trainer_state.json
CHANGED
@@ -1,21 +1,21 @@
|
|
1 |
{
|
2 |
"best_metric": null,
|
3 |
"best_model_checkpoint": null,
|
4 |
-
"epoch": 1.
|
5 |
"eval_steps": 100,
|
6 |
-
"global_step":
|
7 |
"is_hyper_param_search": false,
|
8 |
"is_local_process_zero": true,
|
9 |
"is_world_process_zero": true,
|
10 |
"log_history": [
|
11 |
{
|
12 |
-
"epoch":
|
13 |
-
"learning_rate":
|
14 |
-
"logits/chosen": 0.
|
15 |
-
"logits/rejected": 0.
|
16 |
-
"logps/chosen": -
|
17 |
-
"logps/rejected": -
|
18 |
-
"loss": 0.
|
19 |
"rewards/accuracies": 0.0,
|
20 |
"rewards/chosen": 0.0,
|
21 |
"rewards/margins": 0.0,
|
@@ -23,17 +23,201 @@
|
|
23 |
"step": 1
|
24 |
},
|
25 |
{
|
26 |
-
"epoch":
|
27 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
"total_flos": 0.0,
|
29 |
-
"train_loss": 0.
|
30 |
-
"train_runtime":
|
31 |
-
"train_samples_per_second":
|
32 |
-
"train_steps_per_second": 0.
|
33 |
}
|
34 |
],
|
35 |
"logging_steps": 10,
|
36 |
-
"max_steps":
|
37 |
"num_input_tokens_seen": 0,
|
38 |
"num_train_epochs": 2,
|
39 |
"save_steps": 100,
|
|
|
1 |
{
|
2 |
"best_metric": null,
|
3 |
"best_model_checkpoint": null,
|
4 |
+
"epoch": 1.984,
|
5 |
"eval_steps": 100,
|
6 |
+
"global_step": 124,
|
7 |
"is_hyper_param_search": false,
|
8 |
"is_local_process_zero": true,
|
9 |
"is_world_process_zero": true,
|
10 |
"log_history": [
|
11 |
{
|
12 |
+
"epoch": 0.02,
|
13 |
+
"learning_rate": 3.846153846153847e-07,
|
14 |
+
"logits/chosen": 0.48606717586517334,
|
15 |
+
"logits/rejected": 0.39871031045913696,
|
16 |
+
"logps/chosen": -212.65087890625,
|
17 |
+
"logps/rejected": -202.6477508544922,
|
18 |
+
"loss": 0.0102,
|
19 |
"rewards/accuracies": 0.0,
|
20 |
"rewards/chosen": 0.0,
|
21 |
"rewards/margins": 0.0,
|
|
|
23 |
"step": 1
|
24 |
},
|
25 |
{
|
26 |
+
"epoch": 0.16,
|
27 |
+
"learning_rate": 3.846153846153847e-06,
|
28 |
+
"logits/chosen": 0.15301527082920074,
|
29 |
+
"logits/rejected": 0.20366798341274261,
|
30 |
+
"logps/chosen": -161.48782348632812,
|
31 |
+
"logps/rejected": -135.7443389892578,
|
32 |
+
"loss": 0.0105,
|
33 |
+
"rewards/accuracies": 0.3194444477558136,
|
34 |
+
"rewards/chosen": -0.002633917611092329,
|
35 |
+
"rewards/margins": -0.0009920704178512096,
|
36 |
+
"rewards/rejected": -0.0016418471932411194,
|
37 |
+
"step": 10
|
38 |
+
},
|
39 |
+
{
|
40 |
+
"epoch": 0.32,
|
41 |
+
"learning_rate": 4.951096619903317e-06,
|
42 |
+
"logits/chosen": 0.09449799358844757,
|
43 |
+
"logits/rejected": 0.0444490909576416,
|
44 |
+
"logps/chosen": -182.85354614257812,
|
45 |
+
"logps/rejected": -149.98532104492188,
|
46 |
+
"loss": 0.0104,
|
47 |
+
"rewards/accuracies": 0.40625,
|
48 |
+
"rewards/chosen": -0.0008511155610904098,
|
49 |
+
"rewards/margins": -0.0002212001709267497,
|
50 |
+
"rewards/rejected": -0.000629915448371321,
|
51 |
+
"step": 20
|
52 |
+
},
|
53 |
+
{
|
54 |
+
"epoch": 0.48,
|
55 |
+
"learning_rate": 4.716164218065246e-06,
|
56 |
+
"logits/chosen": 0.19659022986888885,
|
57 |
+
"logits/rejected": 0.24700064957141876,
|
58 |
+
"logps/chosen": -175.32278442382812,
|
59 |
+
"logps/rejected": -143.53839111328125,
|
60 |
+
"loss": 0.0104,
|
61 |
+
"rewards/accuracies": 0.38749998807907104,
|
62 |
+
"rewards/chosen": -0.0011144612217321992,
|
63 |
+
"rewards/margins": -0.000345538865076378,
|
64 |
+
"rewards/rejected": -0.000768922152929008,
|
65 |
+
"step": 30
|
66 |
+
},
|
67 |
+
{
|
68 |
+
"epoch": 0.64,
|
69 |
+
"learning_rate": 4.3048902348863116e-06,
|
70 |
+
"logits/chosen": 0.14261968433856964,
|
71 |
+
"logits/rejected": 0.19110862910747528,
|
72 |
+
"logps/chosen": -181.75755310058594,
|
73 |
+
"logps/rejected": -164.41835021972656,
|
74 |
+
"loss": 0.0103,
|
75 |
+
"rewards/accuracies": 0.39375001192092896,
|
76 |
+
"rewards/chosen": -0.0007939288625493646,
|
77 |
+
"rewards/margins": 0.00022867057123221457,
|
78 |
+
"rewards/rejected": -0.0010225994046777487,
|
79 |
+
"step": 40
|
80 |
+
},
|
81 |
+
{
|
82 |
+
"epoch": 0.8,
|
83 |
+
"learning_rate": 3.7500000000000005e-06,
|
84 |
+
"logits/chosen": 0.07993963360786438,
|
85 |
+
"logits/rejected": 0.1553264558315277,
|
86 |
+
"logps/chosen": -170.30532836914062,
|
87 |
+
"logps/rejected": -144.61050415039062,
|
88 |
+
"loss": 0.0101,
|
89 |
+
"rewards/accuracies": 0.42500001192092896,
|
90 |
+
"rewards/chosen": 0.00015746071585454047,
|
91 |
+
"rewards/margins": 0.0011130261700600386,
|
92 |
+
"rewards/rejected": -0.0009555654833093286,
|
93 |
+
"step": 50
|
94 |
+
},
|
95 |
+
{
|
96 |
+
"epoch": 0.96,
|
97 |
+
"learning_rate": 3.0956464785579125e-06,
|
98 |
+
"logits/chosen": 0.18755468726158142,
|
99 |
+
"logits/rejected": 0.17403154075145721,
|
100 |
+
"logps/chosen": -173.29977416992188,
|
101 |
+
"logps/rejected": -144.7869873046875,
|
102 |
+
"loss": 0.0105,
|
103 |
+
"rewards/accuracies": 0.4000000059604645,
|
104 |
+
"rewards/chosen": -0.0005446333670988679,
|
105 |
+
"rewards/margins": -0.000993421534076333,
|
106 |
+
"rewards/rejected": 0.00044878822518512607,
|
107 |
+
"step": 60
|
108 |
+
},
|
109 |
+
{
|
110 |
+
"epoch": 1.12,
|
111 |
+
"learning_rate": 2.39389699200963e-06,
|
112 |
+
"logits/chosen": 0.2639772593975067,
|
113 |
+
"logits/rejected": 0.2394874542951584,
|
114 |
+
"logps/chosen": -188.4938201904297,
|
115 |
+
"logps/rejected": -160.391357421875,
|
116 |
+
"loss": 0.0102,
|
117 |
+
"rewards/accuracies": 0.3812499940395355,
|
118 |
+
"rewards/chosen": -0.00019804532348643988,
|
119 |
+
"rewards/margins": 0.000605274923145771,
|
120 |
+
"rewards/rejected": -0.0008033200865611434,
|
121 |
+
"step": 70
|
122 |
+
},
|
123 |
+
{
|
124 |
+
"epoch": 1.28,
|
125 |
+
"learning_rate": 1.700590188571887e-06,
|
126 |
+
"logits/chosen": 0.1733393371105194,
|
127 |
+
"logits/rejected": 0.2405407875776291,
|
128 |
+
"logps/chosen": -166.1864471435547,
|
129 |
+
"logps/rejected": -136.4402618408203,
|
130 |
+
"loss": 0.0102,
|
131 |
+
"rewards/accuracies": 0.41874998807907104,
|
132 |
+
"rewards/chosen": 0.0004581899265758693,
|
133 |
+
"rewards/margins": 0.0007379798917099833,
|
134 |
+
"rewards/rejected": -0.0002797898487187922,
|
135 |
+
"step": 80
|
136 |
+
},
|
137 |
+
{
|
138 |
+
"epoch": 1.44,
|
139 |
+
"learning_rate": 1.0708929268538034e-06,
|
140 |
+
"logits/chosen": 0.16160213947296143,
|
141 |
+
"logits/rejected": 0.13504673540592194,
|
142 |
+
"logps/chosen": -175.88116455078125,
|
143 |
+
"logps/rejected": -145.18658447265625,
|
144 |
+
"loss": 0.01,
|
145 |
+
"rewards/accuracies": 0.4312500059604645,
|
146 |
+
"rewards/chosen": 0.0007930597057566047,
|
147 |
+
"rewards/margins": 0.001615246757864952,
|
148 |
+
"rewards/rejected": -0.0008221869356930256,
|
149 |
+
"step": 90
|
150 |
+
},
|
151 |
+
{
|
152 |
+
"epoch": 1.6,
|
153 |
+
"learning_rate": 5.549106142039018e-07,
|
154 |
+
"logits/chosen": 0.18513503670692444,
|
155 |
+
"logits/rejected": 0.14850696921348572,
|
156 |
+
"logps/chosen": -177.84588623046875,
|
157 |
+
"logps/rejected": -152.69290161132812,
|
158 |
+
"loss": 0.0103,
|
159 |
+
"rewards/accuracies": 0.4437499940395355,
|
160 |
+
"rewards/chosen": 0.0004619792161975056,
|
161 |
+
"rewards/margins": 0.00023423954553436488,
|
162 |
+
"rewards/rejected": 0.00022773972887080163,
|
163 |
+
"step": 100
|
164 |
+
},
|
165 |
+
{
|
166 |
+
"epoch": 1.6,
|
167 |
+
"eval_logits/chosen": -0.019983915612101555,
|
168 |
+
"eval_logits/rejected": 0.0771104097366333,
|
169 |
+
"eval_logps/chosen": -306.49530029296875,
|
170 |
+
"eval_logps/rejected": -278.806396484375,
|
171 |
+
"eval_loss": 0.010805144906044006,
|
172 |
+
"eval_rewards/accuracies": 0.5009999871253967,
|
173 |
+
"eval_rewards/chosen": -0.0014956285012885928,
|
174 |
+
"eval_rewards/margins": -9.011628571897745e-05,
|
175 |
+
"eval_rewards/rejected": -0.0014055122155696154,
|
176 |
+
"eval_runtime": 470.6682,
|
177 |
+
"eval_samples_per_second": 4.249,
|
178 |
+
"eval_steps_per_second": 1.062,
|
179 |
+
"step": 100
|
180 |
+
},
|
181 |
+
{
|
182 |
+
"epoch": 1.76,
|
183 |
+
"learning_rate": 1.937002879188285e-07,
|
184 |
+
"logits/chosen": 0.18757307529449463,
|
185 |
+
"logits/rejected": 0.19283342361450195,
|
186 |
+
"logps/chosen": -186.30393981933594,
|
187 |
+
"logps/rejected": -159.83816528320312,
|
188 |
+
"loss": 0.0104,
|
189 |
+
"rewards/accuracies": 0.375,
|
190 |
+
"rewards/chosen": 0.0009433077648282051,
|
191 |
+
"rewards/margins": -0.0004140757955610752,
|
192 |
+
"rewards/rejected": 0.0013573834439739585,
|
193 |
+
"step": 110
|
194 |
+
},
|
195 |
+
{
|
196 |
+
"epoch": 1.92,
|
197 |
+
"learning_rate": 1.6003680950742728e-08,
|
198 |
+
"logits/chosen": 0.07835674285888672,
|
199 |
+
"logits/rejected": 0.13737662136554718,
|
200 |
+
"logps/chosen": -174.80299377441406,
|
201 |
+
"logps/rejected": -154.0167694091797,
|
202 |
+
"loss": 0.0104,
|
203 |
+
"rewards/accuracies": 0.35624998807907104,
|
204 |
+
"rewards/chosen": -5.471259282785468e-05,
|
205 |
+
"rewards/margins": -0.0002534462546464056,
|
206 |
+
"rewards/rejected": 0.0001987336145248264,
|
207 |
+
"step": 120
|
208 |
+
},
|
209 |
+
{
|
210 |
+
"epoch": 1.98,
|
211 |
+
"step": 124,
|
212 |
"total_flos": 0.0,
|
213 |
+
"train_loss": 0.010314414998696696,
|
214 |
+
"train_runtime": 1522.7247,
|
215 |
+
"train_samples_per_second": 1.313,
|
216 |
+
"train_steps_per_second": 0.081
|
217 |
}
|
218 |
],
|
219 |
"logging_steps": 10,
|
220 |
+
"max_steps": 124,
|
221 |
"num_input_tokens_seen": 0,
|
222 |
"num_train_epochs": 2,
|
223 |
"save_steps": 100,
|
training_args.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 5880
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3eedcaca58d1acbf62f2e929c67a3b555a144c3870a46aa43ff58d4b0da18c70
|
3 |
size 5880
|