End of training
Browse files- README.md +15 -2
- all_results.json +22 -0
- eval_results.json +17 -0
- train_results.json +8 -0
- trainer_state.json +314 -0
- training_loss.png +0 -0
- training_rewards_accuracies.png +0 -0
README.md
CHANGED
@@ -3,9 +3,10 @@ base_model: mistralai/Mistral-Nemo-Instruct-2407
|
|
3 |
library_name: peft
|
4 |
license: apache-2.0
|
5 |
tags:
|
|
|
|
|
6 |
- trl
|
7 |
- dpo
|
8 |
-
- llama-factory
|
9 |
- generated_from_trainer
|
10 |
model-index:
|
11 |
- name: Mistral-Nemo-12B-Instruct-SAA
|
@@ -17,7 +18,19 @@ should probably proofread and complete it, then remove this comment. -->
|
|
17 |
|
18 |
# Mistral-Nemo-12B-Instruct-SAA
|
19 |
|
20 |
-
This model is a fine-tuned version of [mistralai/Mistral-Nemo-Instruct-2407](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407) on
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
## Model description
|
23 |
|
|
|
3 |
library_name: peft
|
4 |
license: apache-2.0
|
5 |
tags:
|
6 |
+
- llama-factory
|
7 |
+
- lora
|
8 |
- trl
|
9 |
- dpo
|
|
|
10 |
- generated_from_trainer
|
11 |
model-index:
|
12 |
- name: Mistral-Nemo-12B-Instruct-SAA
|
|
|
18 |
|
19 |
# Mistral-Nemo-12B-Instruct-SAA
|
20 |
|
21 |
+
This model is a fine-tuned version of [mistralai/Mistral-Nemo-Instruct-2407](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407) on the bct_non_cot_dpo_1000 dataset.
|
22 |
+
It achieves the following results on the evaluation set:
|
23 |
+
- Loss: 0.1609
|
24 |
+
- Rewards/chosen: -0.0115
|
25 |
+
- Rewards/rejected: -0.0450
|
26 |
+
- Rewards/accuracies: 0.7900
|
27 |
+
- Rewards/margins: 0.0335
|
28 |
+
- Logps/rejected: -0.4501
|
29 |
+
- Logps/chosen: -0.1152
|
30 |
+
- Logits/rejected: -2.2660
|
31 |
+
- Logits/chosen: -2.2422
|
32 |
+
- Sft Loss: 0.0143
|
33 |
+
- Odds Ratio Loss: 1.4658
|
34 |
|
35 |
## Model description
|
36 |
|
all_results.json
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 2.986666666666667,
|
3 |
+
"eval_logits/chosen": -2.242171049118042,
|
4 |
+
"eval_logits/rejected": -2.265955924987793,
|
5 |
+
"eval_logps/chosen": -0.11519017815589905,
|
6 |
+
"eval_logps/rejected": -0.4500538110733032,
|
7 |
+
"eval_loss": 0.16089552640914917,
|
8 |
+
"eval_odds_ratio_loss": 1.4658138751983643,
|
9 |
+
"eval_rewards/accuracies": 0.7900000214576721,
|
10 |
+
"eval_rewards/chosen": -0.011519019491970539,
|
11 |
+
"eval_rewards/margins": 0.033486369997262955,
|
12 |
+
"eval_rewards/rejected": -0.04500538855791092,
|
13 |
+
"eval_runtime": 5.4517,
|
14 |
+
"eval_samples_per_second": 18.343,
|
15 |
+
"eval_sft_loss": 0.01431415043771267,
|
16 |
+
"eval_steps_per_second": 9.171,
|
17 |
+
"total_flos": 4.410496642646016e+16,
|
18 |
+
"train_loss": 0.4940077399923688,
|
19 |
+
"train_runtime": 519.2548,
|
20 |
+
"train_samples_per_second": 5.2,
|
21 |
+
"train_steps_per_second": 0.324
|
22 |
+
}
|
eval_results.json
ADDED
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 2.986666666666667,
|
3 |
+
"eval_logits/chosen": -2.242171049118042,
|
4 |
+
"eval_logits/rejected": -2.265955924987793,
|
5 |
+
"eval_logps/chosen": -0.11519017815589905,
|
6 |
+
"eval_logps/rejected": -0.4500538110733032,
|
7 |
+
"eval_loss": 0.16089552640914917,
|
8 |
+
"eval_odds_ratio_loss": 1.4658138751983643,
|
9 |
+
"eval_rewards/accuracies": 0.7900000214576721,
|
10 |
+
"eval_rewards/chosen": -0.011519019491970539,
|
11 |
+
"eval_rewards/margins": 0.033486369997262955,
|
12 |
+
"eval_rewards/rejected": -0.04500538855791092,
|
13 |
+
"eval_runtime": 5.4517,
|
14 |
+
"eval_samples_per_second": 18.343,
|
15 |
+
"eval_sft_loss": 0.01431415043771267,
|
16 |
+
"eval_steps_per_second": 9.171
|
17 |
+
}
|
train_results.json
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 2.986666666666667,
|
3 |
+
"total_flos": 4.410496642646016e+16,
|
4 |
+
"train_loss": 0.4940077399923688,
|
5 |
+
"train_runtime": 519.2548,
|
6 |
+
"train_samples_per_second": 5.2,
|
7 |
+
"train_steps_per_second": 0.324
|
8 |
+
}
|
trainer_state.json
ADDED
@@ -0,0 +1,314 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"best_metric": null,
|
3 |
+
"best_model_checkpoint": null,
|
4 |
+
"epoch": 2.986666666666667,
|
5 |
+
"eval_steps": 500,
|
6 |
+
"global_step": 168,
|
7 |
+
"is_hyper_param_search": false,
|
8 |
+
"is_local_process_zero": true,
|
9 |
+
"is_world_process_zero": true,
|
10 |
+
"log_history": [
|
11 |
+
{
|
12 |
+
"epoch": 0.17777777777777778,
|
13 |
+
"grad_norm": 3.8474488258361816,
|
14 |
+
"learning_rate": 2.9411764705882355e-06,
|
15 |
+
"logits/chosen": -2.332357883453369,
|
16 |
+
"logits/rejected": -2.368460178375244,
|
17 |
+
"logps/chosen": -1.2429828643798828,
|
18 |
+
"logps/rejected": -1.659155249595642,
|
19 |
+
"loss": 1.2935,
|
20 |
+
"odds_ratio_loss": 11.457437515258789,
|
21 |
+
"rewards/accuracies": 0.800000011920929,
|
22 |
+
"rewards/chosen": -0.12429828941822052,
|
23 |
+
"rewards/margins": 0.04161724075675011,
|
24 |
+
"rewards/rejected": -0.16591551899909973,
|
25 |
+
"sft_loss": 0.14774402976036072,
|
26 |
+
"step": 10
|
27 |
+
},
|
28 |
+
{
|
29 |
+
"epoch": 0.35555555555555557,
|
30 |
+
"grad_norm": 5.139791965484619,
|
31 |
+
"learning_rate": 4.995131923687488e-06,
|
32 |
+
"logits/chosen": -2.292804002761841,
|
33 |
+
"logits/rejected": -2.327223300933838,
|
34 |
+
"logps/chosen": -1.2883718013763428,
|
35 |
+
"logps/rejected": -1.7239799499511719,
|
36 |
+
"loss": 1.337,
|
37 |
+
"odds_ratio_loss": 11.885274887084961,
|
38 |
+
"rewards/accuracies": 0.831250011920929,
|
39 |
+
"rewards/chosen": -0.12883718311786652,
|
40 |
+
"rewards/margins": 0.043560806661844254,
|
41 |
+
"rewards/rejected": -0.17239800095558167,
|
42 |
+
"sft_loss": 0.14851602911949158,
|
43 |
+
"step": 20
|
44 |
+
},
|
45 |
+
{
|
46 |
+
"epoch": 0.5333333333333333,
|
47 |
+
"grad_norm": 5.0462751388549805,
|
48 |
+
"learning_rate": 4.90911473983908e-06,
|
49 |
+
"logits/chosen": -2.3472771644592285,
|
50 |
+
"logits/rejected": -2.3845486640930176,
|
51 |
+
"logps/chosen": -1.186187982559204,
|
52 |
+
"logps/rejected": -1.59175705909729,
|
53 |
+
"loss": 1.2375,
|
54 |
+
"odds_ratio_loss": 10.995233535766602,
|
55 |
+
"rewards/accuracies": 0.8187500238418579,
|
56 |
+
"rewards/chosen": -0.11861880123615265,
|
57 |
+
"rewards/margins": 0.04055692255496979,
|
58 |
+
"rewards/rejected": -0.15917572379112244,
|
59 |
+
"sft_loss": 0.13796505331993103,
|
60 |
+
"step": 30
|
61 |
+
},
|
62 |
+
{
|
63 |
+
"epoch": 0.7111111111111111,
|
64 |
+
"grad_norm": 3.613401412963867,
|
65 |
+
"learning_rate": 4.71919261421297e-06,
|
66 |
+
"logits/chosen": -2.370697021484375,
|
67 |
+
"logits/rejected": -2.4233319759368896,
|
68 |
+
"logps/chosen": -0.9008461236953735,
|
69 |
+
"logps/rejected": -1.4053981304168701,
|
70 |
+
"loss": 0.9426,
|
71 |
+
"odds_ratio_loss": 8.488945007324219,
|
72 |
+
"rewards/accuracies": 0.8687499761581421,
|
73 |
+
"rewards/chosen": -0.09008461236953735,
|
74 |
+
"rewards/margins": 0.05045522004365921,
|
75 |
+
"rewards/rejected": -0.14053983986377716,
|
76 |
+
"sft_loss": 0.09369887411594391,
|
77 |
+
"step": 40
|
78 |
+
},
|
79 |
+
{
|
80 |
+
"epoch": 0.8888888888888888,
|
81 |
+
"grad_norm": 3.07717227935791,
|
82 |
+
"learning_rate": 4.43355687413747e-06,
|
83 |
+
"logits/chosen": -2.4286742210388184,
|
84 |
+
"logits/rejected": -2.4549202919006348,
|
85 |
+
"logps/chosen": -0.6461768746376038,
|
86 |
+
"logps/rejected": -1.0805357694625854,
|
87 |
+
"loss": 0.6945,
|
88 |
+
"odds_ratio_loss": 6.29970645904541,
|
89 |
+
"rewards/accuracies": 0.824999988079071,
|
90 |
+
"rewards/chosen": -0.06461768597364426,
|
91 |
+
"rewards/margins": 0.04343589395284653,
|
92 |
+
"rewards/rejected": -0.10805357992649078,
|
93 |
+
"sft_loss": 0.06456876546144485,
|
94 |
+
"step": 50
|
95 |
+
},
|
96 |
+
{
|
97 |
+
"epoch": 1.0666666666666667,
|
98 |
+
"grad_norm": 2.27549409866333,
|
99 |
+
"learning_rate": 4.064526968101844e-06,
|
100 |
+
"logits/chosen": -2.317702531814575,
|
101 |
+
"logits/rejected": -2.337463855743408,
|
102 |
+
"logps/chosen": -0.5583322048187256,
|
103 |
+
"logps/rejected": -1.0102574825286865,
|
104 |
+
"loss": 0.6081,
|
105 |
+
"odds_ratio_loss": 5.508663177490234,
|
106 |
+
"rewards/accuracies": 0.7562500238418579,
|
107 |
+
"rewards/chosen": -0.05583322048187256,
|
108 |
+
"rewards/margins": 0.04519252851605415,
|
109 |
+
"rewards/rejected": -0.10102574527263641,
|
110 |
+
"sft_loss": 0.05719046667218208,
|
111 |
+
"step": 60
|
112 |
+
},
|
113 |
+
{
|
114 |
+
"epoch": 1.2444444444444445,
|
115 |
+
"grad_norm": 2.98760724067688,
|
116 |
+
"learning_rate": 3.6280191288478437e-06,
|
117 |
+
"logits/chosen": -2.4034271240234375,
|
118 |
+
"logits/rejected": -2.4294583797454834,
|
119 |
+
"logps/chosen": -0.35430365800857544,
|
120 |
+
"logps/rejected": -0.7878178358078003,
|
121 |
+
"loss": 0.3968,
|
122 |
+
"odds_ratio_loss": 3.5909438133239746,
|
123 |
+
"rewards/accuracies": 0.824999988079071,
|
124 |
+
"rewards/chosen": -0.035430364310741425,
|
125 |
+
"rewards/margins": 0.0433514229953289,
|
126 |
+
"rewards/rejected": -0.07878179103136063,
|
127 |
+
"sft_loss": 0.037743426859378815,
|
128 |
+
"step": 70
|
129 |
+
},
|
130 |
+
{
|
131 |
+
"epoch": 1.4222222222222223,
|
132 |
+
"grad_norm": 2.200446605682373,
|
133 |
+
"learning_rate": 3.142859907420615e-06,
|
134 |
+
"logits/chosen": -2.336760997772217,
|
135 |
+
"logits/rejected": -2.365061044692993,
|
136 |
+
"logps/chosen": -0.2458254098892212,
|
137 |
+
"logps/rejected": -0.6315831542015076,
|
138 |
+
"loss": 0.2882,
|
139 |
+
"odds_ratio_loss": 2.5942039489746094,
|
140 |
+
"rewards/accuracies": 0.831250011920929,
|
141 |
+
"rewards/chosen": -0.02458254061639309,
|
142 |
+
"rewards/margins": 0.03857577592134476,
|
143 |
+
"rewards/rejected": -0.063158318400383,
|
144 |
+
"sft_loss": 0.028748363256454468,
|
145 |
+
"step": 80
|
146 |
+
},
|
147 |
+
{
|
148 |
+
"epoch": 1.6,
|
149 |
+
"grad_norm": 1.6536734104156494,
|
150 |
+
"learning_rate": 2.629974185404951e-06,
|
151 |
+
"logits/chosen": -2.313284397125244,
|
152 |
+
"logits/rejected": -2.3488173484802246,
|
153 |
+
"logps/chosen": -0.1942504495382309,
|
154 |
+
"logps/rejected": -0.6369145512580872,
|
155 |
+
"loss": 0.2406,
|
156 |
+
"odds_ratio_loss": 2.1658334732055664,
|
157 |
+
"rewards/accuracies": 0.793749988079071,
|
158 |
+
"rewards/chosen": -0.01942504197359085,
|
159 |
+
"rewards/margins": 0.04426640644669533,
|
160 |
+
"rewards/rejected": -0.06369145214557648,
|
161 |
+
"sft_loss": 0.023997236043214798,
|
162 |
+
"step": 90
|
163 |
+
},
|
164 |
+
{
|
165 |
+
"epoch": 1.7777777777777777,
|
166 |
+
"grad_norm": 1.2765487432479858,
|
167 |
+
"learning_rate": 2.1114826863194882e-06,
|
168 |
+
"logits/chosen": -2.2804689407348633,
|
169 |
+
"logits/rejected": -2.3144404888153076,
|
170 |
+
"logps/chosen": -0.13879844546318054,
|
171 |
+
"logps/rejected": -0.6243221163749695,
|
172 |
+
"loss": 0.1759,
|
173 |
+
"odds_ratio_loss": 1.5965832471847534,
|
174 |
+
"rewards/accuracies": 0.862500011920929,
|
175 |
+
"rewards/chosen": -0.013879844918847084,
|
176 |
+
"rewards/margins": 0.048552367836236954,
|
177 |
+
"rewards/rejected": -0.06243220716714859,
|
178 |
+
"sft_loss": 0.01626196689903736,
|
179 |
+
"step": 100
|
180 |
+
},
|
181 |
+
{
|
182 |
+
"epoch": 1.9555555555555557,
|
183 |
+
"grad_norm": 1.2211233377456665,
|
184 |
+
"learning_rate": 1.6097479104361328e-06,
|
185 |
+
"logits/chosen": -2.317593812942505,
|
186 |
+
"logits/rejected": -2.336158037185669,
|
187 |
+
"logps/chosen": -0.1410079300403595,
|
188 |
+
"logps/rejected": -0.5155030488967896,
|
189 |
+
"loss": 0.1836,
|
190 |
+
"odds_ratio_loss": 1.6868371963500977,
|
191 |
+
"rewards/accuracies": 0.793749988079071,
|
192 |
+
"rewards/chosen": -0.014100792817771435,
|
193 |
+
"rewards/margins": 0.037449512630701065,
|
194 |
+
"rewards/rejected": -0.051550306379795074,
|
195 |
+
"sft_loss": 0.01492035947740078,
|
196 |
+
"step": 110
|
197 |
+
},
|
198 |
+
{
|
199 |
+
"epoch": 2.1333333333333333,
|
200 |
+
"grad_norm": 1.3183128833770752,
|
201 |
+
"learning_rate": 1.1464096417858821e-06,
|
202 |
+
"logits/chosen": -2.290233850479126,
|
203 |
+
"logits/rejected": -2.3088765144348145,
|
204 |
+
"logps/chosen": -0.10447756201028824,
|
205 |
+
"logps/rejected": -0.6317979097366333,
|
206 |
+
"loss": 0.1326,
|
207 |
+
"odds_ratio_loss": 1.1886184215545654,
|
208 |
+
"rewards/accuracies": 0.887499988079071,
|
209 |
+
"rewards/chosen": -0.010447757318615913,
|
210 |
+
"rewards/margins": 0.052732039242982864,
|
211 |
+
"rewards/rejected": -0.06317979097366333,
|
212 |
+
"sft_loss": 0.013712344691157341,
|
213 |
+
"step": 120
|
214 |
+
},
|
215 |
+
{
|
216 |
+
"epoch": 2.311111111111111,
|
217 |
+
"grad_norm": 1.7163466215133667,
|
218 |
+
"learning_rate": 7.414516258630245e-07,
|
219 |
+
"logits/chosen": -2.262594223022461,
|
220 |
+
"logits/rejected": -2.284545421600342,
|
221 |
+
"logps/chosen": -0.12122899293899536,
|
222 |
+
"logps/rejected": -0.5315740704536438,
|
223 |
+
"loss": 0.1618,
|
224 |
+
"odds_ratio_loss": 1.4604320526123047,
|
225 |
+
"rewards/accuracies": 0.8062499761581421,
|
226 |
+
"rewards/chosen": -0.012122899293899536,
|
227 |
+
"rewards/margins": 0.0410345084965229,
|
228 |
+
"rewards/rejected": -0.05315741151571274,
|
229 |
+
"sft_loss": 0.015804503113031387,
|
230 |
+
"step": 130
|
231 |
+
},
|
232 |
+
{
|
233 |
+
"epoch": 2.488888888888889,
|
234 |
+
"grad_norm": 3.0588433742523193,
|
235 |
+
"learning_rate": 4.123396721497977e-07,
|
236 |
+
"logits/chosen": -2.3231940269470215,
|
237 |
+
"logits/rejected": -2.34096622467041,
|
238 |
+
"logps/chosen": -0.12840591371059418,
|
239 |
+
"logps/rejected": -0.5461179614067078,
|
240 |
+
"loss": 0.1688,
|
241 |
+
"odds_ratio_loss": 1.521639108657837,
|
242 |
+
"rewards/accuracies": 0.831250011920929,
|
243 |
+
"rewards/chosen": -0.012840591371059418,
|
244 |
+
"rewards/margins": 0.04177120327949524,
|
245 |
+
"rewards/rejected": -0.05461179465055466,
|
246 |
+
"sft_loss": 0.016592377796769142,
|
247 |
+
"step": 140
|
248 |
+
},
|
249 |
+
{
|
250 |
+
"epoch": 2.6666666666666665,
|
251 |
+
"grad_norm": 2.731790065765381,
|
252 |
+
"learning_rate": 1.7326835503629542e-07,
|
253 |
+
"logits/chosen": -2.2994039058685303,
|
254 |
+
"logits/rejected": -2.324022054672241,
|
255 |
+
"logps/chosen": -0.12199757248163223,
|
256 |
+
"logps/rejected": -0.5101950168609619,
|
257 |
+
"loss": 0.1648,
|
258 |
+
"odds_ratio_loss": 1.4926129579544067,
|
259 |
+
"rewards/accuracies": 0.8062499761581421,
|
260 |
+
"rewards/chosen": -0.012199757620692253,
|
261 |
+
"rewards/margins": 0.03881974145770073,
|
262 |
+
"rewards/rejected": -0.05101950094103813,
|
263 |
+
"sft_loss": 0.015529977157711983,
|
264 |
+
"step": 150
|
265 |
+
},
|
266 |
+
{
|
267 |
+
"epoch": 2.8444444444444446,
|
268 |
+
"grad_norm": 1.6209518909454346,
|
269 |
+
"learning_rate": 3.4548802869627806e-08,
|
270 |
+
"logits/chosen": -2.19964599609375,
|
271 |
+
"logits/rejected": -2.222959041595459,
|
272 |
+
"logps/chosen": -0.11084076017141342,
|
273 |
+
"logps/rejected": -0.5317128300666809,
|
274 |
+
"loss": 0.1456,
|
275 |
+
"odds_ratio_loss": 1.3233401775360107,
|
276 |
+
"rewards/accuracies": 0.824999988079071,
|
277 |
+
"rewards/chosen": -0.011084076017141342,
|
278 |
+
"rewards/margins": 0.042087212204933167,
|
279 |
+
"rewards/rejected": -0.05317128449678421,
|
280 |
+
"sft_loss": 0.013311339542269707,
|
281 |
+
"step": 160
|
282 |
+
},
|
283 |
+
{
|
284 |
+
"epoch": 2.986666666666667,
|
285 |
+
"step": 168,
|
286 |
+
"total_flos": 4.410496642646016e+16,
|
287 |
+
"train_loss": 0.4940077399923688,
|
288 |
+
"train_runtime": 519.2548,
|
289 |
+
"train_samples_per_second": 5.2,
|
290 |
+
"train_steps_per_second": 0.324
|
291 |
+
}
|
292 |
+
],
|
293 |
+
"logging_steps": 10,
|
294 |
+
"max_steps": 168,
|
295 |
+
"num_input_tokens_seen": 0,
|
296 |
+
"num_train_epochs": 3,
|
297 |
+
"save_steps": 500,
|
298 |
+
"stateful_callbacks": {
|
299 |
+
"TrainerControl": {
|
300 |
+
"args": {
|
301 |
+
"should_epoch_stop": false,
|
302 |
+
"should_evaluate": false,
|
303 |
+
"should_log": false,
|
304 |
+
"should_save": true,
|
305 |
+
"should_training_stop": true
|
306 |
+
},
|
307 |
+
"attributes": {}
|
308 |
+
}
|
309 |
+
},
|
310 |
+
"total_flos": 4.410496642646016e+16,
|
311 |
+
"train_batch_size": 2,
|
312 |
+
"trial_name": null,
|
313 |
+
"trial_params": null
|
314 |
+
}
|
training_loss.png
ADDED
training_rewards_accuracies.png
ADDED