hugodk-sch
commited on
Commit
•
c6d1958
1
Parent(s):
2de4e50
Model save
Browse files- README.md +14 -16
- adapter_model.safetensors +1 -1
- all_results.json +2 -15
- train_results.json +2 -2
- trainer_state.json +414 -414
README.md
CHANGED
@@ -1,13 +1,11 @@
|
|
1 |
---
|
|
|
2 |
library_name: peft
|
3 |
tags:
|
4 |
-
- alignment-handbook
|
5 |
- trl
|
6 |
- dpo
|
7 |
- generated_from_trainer
|
8 |
base_model: norallm/normistral-7b-warm
|
9 |
-
datasets:
|
10 |
-
- hugodk-sch/aftonposten_title_prefs
|
11 |
model-index:
|
12 |
- name: ap-normistral-7b-align-scan
|
13 |
results: []
|
@@ -18,17 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
|
|
18 |
|
19 |
# ap-normistral-7b-align-scan
|
20 |
|
21 |
-
This model is a fine-tuned version of [
|
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:
|
28 |
-
- Logps/rejected: -35.
|
29 |
-
- Logps/chosen: -32.
|
30 |
-
- Logits/rejected: 98.
|
31 |
-
- Logits/chosen: 98.
|
32 |
|
33 |
## Model description
|
34 |
|
@@ -63,9 +61,9 @@ The following hyperparameters were used during training:
|
|
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.
|
67 |
-
| 0.
|
68 |
-
| 0.
|
69 |
|
70 |
|
71 |
### Framework versions
|
|
|
1 |
---
|
2 |
+
license: apache-2.0
|
3 |
library_name: peft
|
4 |
tags:
|
|
|
5 |
- trl
|
6 |
- dpo
|
7 |
- generated_from_trainer
|
8 |
base_model: norallm/normistral-7b-warm
|
|
|
|
|
9 |
model-index:
|
10 |
- name: ap-normistral-7b-align-scan
|
11 |
results: []
|
|
|
16 |
|
17 |
# ap-normistral-7b-align-scan
|
18 |
|
19 |
+
This model is a fine-tuned version of [norallm/normistral-7b-warm](https://huggingface.co/norallm/normistral-7b-warm) on the None dataset.
|
20 |
It achieves the following results on the evaluation set:
|
21 |
+
- Loss: 0.4962
|
22 |
+
- Rewards/chosen: 0.0856
|
23 |
+
- Rewards/rejected: 0.0600
|
24 |
+
- Rewards/accuracies: 0.5104
|
25 |
+
- Rewards/margins: 0.0256
|
26 |
+
- Logps/rejected: -35.8166
|
27 |
+
- Logps/chosen: -32.2293
|
28 |
+
- Logits/rejected: 98.3704
|
29 |
+
- Logits/chosen: 98.3655
|
30 |
|
31 |
## Model description
|
32 |
|
|
|
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 |
+
| 0.4749 | 0.26 | 100 | 0.5047 | 0.0131 | 0.0430 | 0.4747 | -0.0299 | -35.8591 | -32.4105 | 98.7377 | 98.7500 |
|
65 |
+
| 0.3843 | 0.52 | 200 | 0.4944 | 0.0238 | -0.0061 | 0.5307 | 0.0299 | -35.9817 | -32.3837 | 98.3784 | 98.3836 |
|
66 |
+
| 0.363 | 0.78 | 300 | 0.4962 | 0.0856 | 0.0600 | 0.5104 | 0.0256 | -35.8166 | -32.2293 | 98.3704 | 98.3655 |
|
67 |
|
68 |
|
69 |
### Framework versions
|
adapter_model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 671150064
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4e09aeea11f61f666c4168ca9571269efa984ea3dde291fbc770977b3ec68bbb
|
3 |
size 671150064
|
all_results.json
CHANGED
@@ -1,20 +1,7 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"
|
4 |
-
"
|
5 |
-
"eval_logps/chosen": -32.303466796875,
|
6 |
-
"eval_logps/rejected": -35.77931594848633,
|
7 |
-
"eval_loss": 0.5059725642204285,
|
8 |
-
"eval_rewards/accuracies": 0.5220099687576294,
|
9 |
-
"eval_rewards/chosen": 0.09779632091522217,
|
10 |
-
"eval_rewards/margins": -0.033251672983169556,
|
11 |
-
"eval_rewards/rejected": 0.13104799389839172,
|
12 |
-
"eval_runtime": 103.819,
|
13 |
-
"eval_samples": 343,
|
14 |
-
"eval_samples_per_second": 3.304,
|
15 |
-
"eval_steps_per_second": 0.414,
|
16 |
-
"train_loss": 0.38857739250381274,
|
17 |
-
"train_runtime": 2559.074,
|
18 |
"train_samples": 3079,
|
19 |
"train_samples_per_second": 1.203,
|
20 |
"train_steps_per_second": 0.15
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"train_loss": 0.4199758805237807,
|
4 |
+
"train_runtime": 2559.5376,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.203,
|
7 |
"train_steps_per_second": 0.15
|
train_results.json
CHANGED
@@ -1,7 +1,7 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"train_loss": 0.
|
4 |
-
"train_runtime": 2559.
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.203,
|
7 |
"train_steps_per_second": 0.15
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"train_loss": 0.4199758805237807,
|
4 |
+
"train_runtime": 2559.5376,
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.203,
|
7 |
"train_steps_per_second": 0.15
|
trainer_state.json
CHANGED
@@ -10,7 +10,7 @@
|
|
10 |
"log_history": [
|
11 |
{
|
12 |
"epoch": 0.0,
|
13 |
-
"grad_norm":
|
14 |
"learning_rate": 1.282051282051282e-07,
|
15 |
"logits/chosen": 88.18099975585938,
|
16 |
"logits/rejected": 88.25153350830078,
|
@@ -25,628 +25,628 @@
|
|
25 |
},
|
26 |
{
|
27 |
"epoch": 0.03,
|
28 |
-
"grad_norm":
|
29 |
"learning_rate": 1.282051282051282e-06,
|
30 |
-
"logits/chosen": 81.
|
31 |
-
"logits/rejected": 80.
|
32 |
-
"logps/chosen": -34.
|
33 |
-
"logps/rejected": -33.
|
34 |
-
"loss": 0.
|
35 |
-
"rewards/accuracies": 0.
|
36 |
-
"rewards/chosen": 0.
|
37 |
-
"rewards/margins": 0.
|
38 |
-
"rewards/rejected": -0.
|
39 |
"step": 10
|
40 |
},
|
41 |
{
|
42 |
"epoch": 0.05,
|
43 |
-
"grad_norm":
|
44 |
"learning_rate": 2.564102564102564e-06,
|
45 |
-
"logits/chosen": 80.
|
46 |
-
"logits/rejected": 80.
|
47 |
-
"logps/chosen": -33.
|
48 |
-
"logps/rejected": -30.
|
49 |
-
"loss": 0.
|
50 |
-
"rewards/accuracies": 0.
|
51 |
-
"rewards/chosen": 0.
|
52 |
-
"rewards/margins": 0.
|
53 |
-
"rewards/rejected": 0.
|
54 |
"step": 20
|
55 |
},
|
56 |
{
|
57 |
"epoch": 0.08,
|
58 |
-
"grad_norm":
|
59 |
"learning_rate": 3.846153846153847e-06,
|
60 |
-
"logits/chosen": 82.
|
61 |
-
"logits/rejected": 82.
|
62 |
-
"logps/chosen": -33.
|
63 |
-
"logps/rejected": -31.
|
64 |
-
"loss": 0.
|
65 |
-
"rewards/accuracies": 0.
|
66 |
-
"rewards/chosen": 0.
|
67 |
-
"rewards/margins": -0.
|
68 |
-
"rewards/rejected": 0.
|
69 |
"step": 30
|
70 |
},
|
71 |
{
|
72 |
"epoch": 0.1,
|
73 |
-
"grad_norm":
|
74 |
"learning_rate": 4.999896948438434e-06,
|
75 |
-
"logits/chosen": 81.
|
76 |
-
"logits/rejected": 81.
|
77 |
-
"logps/chosen": -32.
|
78 |
-
"logps/rejected": -33.
|
79 |
-
"loss": 0.
|
80 |
-
"rewards/accuracies": 0.
|
81 |
-
"rewards/chosen": 0.
|
82 |
-
"rewards/margins": 0.
|
83 |
-
"rewards/rejected": 0.
|
84 |
"step": 40
|
85 |
},
|
86 |
{
|
87 |
"epoch": 0.13,
|
88 |
-
"grad_norm":
|
89 |
"learning_rate": 4.987541037542187e-06,
|
90 |
-
"logits/chosen": 78.
|
91 |
-
"logits/rejected": 78.
|
92 |
-
"logps/chosen": -30.
|
93 |
-
"logps/rejected": -30.
|
94 |
-
"loss": 0.
|
95 |
-
"rewards/accuracies": 0.
|
96 |
-
"rewards/chosen": 0.
|
97 |
-
"rewards/margins": 0.
|
98 |
-
"rewards/rejected": 0.
|
99 |
"step": 50
|
100 |
},
|
101 |
{
|
102 |
"epoch": 0.16,
|
103 |
-
"grad_norm":
|
104 |
"learning_rate": 4.954691471941119e-06,
|
105 |
-
"logits/chosen": 83.
|
106 |
-
"logits/rejected": 83.
|
107 |
-
"logps/chosen": -30.
|
108 |
-
"logps/rejected": -29.
|
109 |
-
"loss": 0.
|
110 |
-
"rewards/accuracies": 0.
|
111 |
-
"rewards/chosen": 0.
|
112 |
-
"rewards/margins": 0.
|
113 |
-
"rewards/rejected": 0.
|
114 |
"step": 60
|
115 |
},
|
116 |
{
|
117 |
"epoch": 0.18,
|
118 |
-
"grad_norm":
|
119 |
"learning_rate": 4.901618883413549e-06,
|
120 |
-
"logits/chosen": 84.
|
121 |
-
"logits/rejected": 84.
|
122 |
-
"logps/chosen": -30.
|
123 |
-
"logps/rejected": -32.
|
124 |
-
"loss": 0.
|
125 |
-
"rewards/accuracies": 0.
|
126 |
-
"rewards/chosen": 0.
|
127 |
-
"rewards/margins": 0.
|
128 |
-
"rewards/rejected": 0.
|
129 |
"step": 70
|
130 |
},
|
131 |
{
|
132 |
"epoch": 0.21,
|
133 |
-
"grad_norm":
|
134 |
"learning_rate": 4.828760511501322e-06,
|
135 |
-
"logits/chosen": 81.
|
136 |
-
"logits/rejected": 81.
|
137 |
-
"logps/chosen": -31.
|
138 |
-
"logps/rejected": -30.
|
139 |
-
"loss": 0.
|
140 |
-
"rewards/accuracies": 0.
|
141 |
-
"rewards/chosen": 0.
|
142 |
-
"rewards/margins": 0.
|
143 |
-
"rewards/rejected": 0.
|
144 |
"step": 80
|
145 |
},
|
146 |
{
|
147 |
"epoch": 0.23,
|
148 |
-
"grad_norm":
|
149 |
"learning_rate": 4.7367166013034295e-06,
|
150 |
-
"logits/chosen": 78.
|
151 |
-
"logits/rejected": 78.
|
152 |
-
"logps/chosen": -32.
|
153 |
-
"logps/rejected": -30.
|
154 |
-
"loss": 0.
|
155 |
-
"rewards/accuracies": 0.
|
156 |
-
"rewards/chosen": 0.
|
157 |
-
"rewards/margins": 0.
|
158 |
-
"rewards/rejected": 0.
|
159 |
"step": 90
|
160 |
},
|
161 |
{
|
162 |
"epoch": 0.26,
|
163 |
-
"grad_norm":
|
164 |
"learning_rate": 4.626245458345211e-06,
|
165 |
-
"logits/chosen": 83.
|
166 |
-
"logits/rejected": 83.
|
167 |
-
"logps/chosen": -33.
|
168 |
-
"logps/rejected": -31.
|
169 |
-
"loss": 0.
|
170 |
-
"rewards/accuracies": 0.
|
171 |
-
"rewards/chosen": 0.
|
172 |
-
"rewards/margins": 0.
|
173 |
-
"rewards/rejected":
|
174 |
"step": 100
|
175 |
},
|
176 |
{
|
177 |
"epoch": 0.26,
|
178 |
-
"eval_logits/chosen": 98.
|
179 |
-
"eval_logits/rejected": 98.
|
180 |
-
"eval_logps/chosen": -32.
|
181 |
-
"eval_logps/rejected": -35.
|
182 |
-
"eval_loss": 0.
|
183 |
-
"eval_rewards/accuracies": 0.
|
184 |
-
"eval_rewards/chosen": 0.
|
185 |
-
"eval_rewards/margins": 0.
|
186 |
-
"eval_rewards/rejected": 0.
|
187 |
-
"eval_runtime": 104.
|
188 |
-
"eval_samples_per_second": 3.
|
189 |
-
"eval_steps_per_second": 0.
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
-
"grad_norm":
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
-
"logits/chosen":
|
197 |
-
"logits/rejected": 83.
|
198 |
-
"logps/chosen": -32.
|
199 |
-
"logps/rejected": -32.
|
200 |
-
"loss": 0.
|
201 |
-
"rewards/accuracies": 0.
|
202 |
-
"rewards/chosen": 0.
|
203 |
-
"rewards/margins": 0.
|
204 |
-
"rewards/rejected": 0.
|
205 |
"step": 110
|
206 |
},
|
207 |
{
|
208 |
"epoch": 0.31,
|
209 |
-
"grad_norm":
|
210 |
"learning_rate": 4.353806263777678e-06,
|
211 |
-
"logits/chosen":
|
212 |
-
"logits/rejected": 84.
|
213 |
-
"logps/chosen": -28.
|
214 |
-
"logps/rejected": -35.
|
215 |
-
"loss": 0.
|
216 |
-
"rewards/accuracies": 0.
|
217 |
-
"rewards/chosen": 0.
|
218 |
-
"rewards/margins": 0.
|
219 |
-
"rewards/rejected": 0.
|
220 |
"step": 120
|
221 |
},
|
222 |
{
|
223 |
"epoch": 0.34,
|
224 |
-
"grad_norm":
|
225 |
"learning_rate": 4.1940827077152755e-06,
|
226 |
-
"logits/chosen": 81.
|
227 |
-
"logits/rejected": 81.
|
228 |
-
"logps/chosen": -30.
|
229 |
-
"logps/rejected": -31.
|
230 |
-
"loss": 0.
|
231 |
-
"rewards/accuracies": 0.
|
232 |
-
"rewards/chosen": 0.
|
233 |
-
"rewards/margins": 0.
|
234 |
-
"rewards/rejected": 0.
|
235 |
"step": 130
|
236 |
},
|
237 |
{
|
238 |
"epoch": 0.36,
|
239 |
-
"grad_norm":
|
240 |
"learning_rate": 4.0204024186666215e-06,
|
241 |
-
"logits/chosen": 82.
|
242 |
-
"logits/rejected": 82.
|
243 |
-
"logps/chosen": -26.
|
244 |
-
"logps/rejected": -32.
|
245 |
-
"loss": 0.
|
246 |
-
"rewards/accuracies": 0.
|
247 |
-
"rewards/chosen": 0.
|
248 |
-
"rewards/margins": 0.
|
249 |
-
"rewards/rejected": -0.
|
250 |
"step": 140
|
251 |
},
|
252 |
{
|
253 |
"epoch": 0.39,
|
254 |
-
"grad_norm":
|
255 |
"learning_rate": 3.834196265035119e-06,
|
256 |
-
"logits/chosen": 80.
|
257 |
-
"logits/rejected": 80.
|
258 |
-
"logps/chosen": -28.
|
259 |
-
"logps/rejected": -32.
|
260 |
-
"loss": 0.
|
261 |
-
"rewards/accuracies": 0.
|
262 |
-
"rewards/chosen": 0.
|
263 |
-
"rewards/margins": 0.
|
264 |
-
"rewards/rejected": -0.
|
265 |
"step": 150
|
266 |
},
|
267 |
{
|
268 |
"epoch": 0.42,
|
269 |
-
"grad_norm":
|
270 |
"learning_rate": 3.636998309800573e-06,
|
271 |
-
"logits/chosen": 82.
|
272 |
-
"logits/rejected": 82.
|
273 |
-
"logps/chosen": -33.
|
274 |
-
"logps/rejected": -30.
|
275 |
-
"loss": 0.
|
276 |
-
"rewards/accuracies": 0.
|
277 |
-
"rewards/chosen": 0.
|
278 |
-
"rewards/margins": 0.
|
279 |
-
"rewards/rejected": -0.
|
280 |
"step": 160
|
281 |
},
|
282 |
{
|
283 |
"epoch": 0.44,
|
284 |
-
"grad_norm":
|
285 |
"learning_rate": 3.4304331721118078e-06,
|
286 |
-
"logits/chosen": 83.
|
287 |
-
"logits/rejected": 83.
|
288 |
-
"logps/chosen": -30.
|
289 |
-
"logps/rejected": -32.
|
290 |
-
"loss": 0.
|
291 |
"rewards/accuracies": 0.6499999761581421,
|
292 |
-
"rewards/chosen": 0.
|
293 |
-
"rewards/margins": 0.
|
294 |
-
"rewards/rejected": -0.
|
295 |
"step": 170
|
296 |
},
|
297 |
{
|
298 |
"epoch": 0.47,
|
299 |
-
"grad_norm":
|
300 |
"learning_rate": 3.2162026428305436e-06,
|
301 |
-
"logits/chosen": 80.
|
302 |
-
"logits/rejected": 80.
|
303 |
-
"logps/chosen": -30.
|
304 |
-
"logps/rejected": -31.
|
305 |
-
"loss": 0.
|
306 |
-
"rewards/accuracies": 0.
|
307 |
-
"rewards/chosen": 0.
|
308 |
-
"rewards/margins": 0.
|
309 |
-
"rewards/rejected": -0.
|
310 |
"step": 180
|
311 |
},
|
312 |
{
|
313 |
"epoch": 0.49,
|
314 |
-
"grad_norm":
|
315 |
"learning_rate": 2.996071664294641e-06,
|
316 |
-
"logits/chosen": 82.
|
317 |
-
"logits/rejected": 82.
|
318 |
-
"logps/chosen": -29.
|
319 |
-
"logps/rejected": -30.
|
320 |
-
"loss": 0.
|
321 |
"rewards/accuracies": 0.612500011920929,
|
322 |
-
"rewards/chosen": 0.
|
323 |
-
"rewards/margins": 0.
|
324 |
-
"rewards/rejected": 0.
|
325 |
"step": 190
|
326 |
},
|
327 |
{
|
328 |
"epoch": 0.52,
|
329 |
-
"grad_norm":
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
-
"logits/chosen": 77.
|
332 |
-
"logits/rejected": 77.
|
333 |
-
"logps/chosen": -33.
|
334 |
-
"logps/rejected": -32.
|
335 |
-
"loss": 0.
|
336 |
"rewards/accuracies": 0.699999988079071,
|
337 |
-
"rewards/chosen":
|
338 |
-
"rewards/margins":
|
339 |
-
"rewards/rejected": 0.
|
340 |
"step": 200
|
341 |
},
|
342 |
{
|
343 |
"epoch": 0.52,
|
344 |
-
"eval_logits/chosen": 98.
|
345 |
-
"eval_logits/rejected": 98.
|
346 |
-
"eval_logps/chosen": -32.
|
347 |
-
"eval_logps/rejected": -35.
|
348 |
-
"eval_loss": 0.
|
349 |
-
"eval_rewards/accuracies": 0.
|
350 |
-
"eval_rewards/chosen": 0.
|
351 |
-
"eval_rewards/margins": 0.
|
352 |
-
"eval_rewards/rejected": 0.
|
353 |
-
"eval_runtime": 104.
|
354 |
-
"eval_samples_per_second": 3.
|
355 |
"eval_steps_per_second": 0.413,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
-
"grad_norm":
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
-
"logits/chosen": 80.
|
363 |
-
"logits/rejected": 80.
|
364 |
-
"logps/chosen": -32.
|
365 |
-
"logps/rejected": -
|
366 |
-
"loss": 0.
|
367 |
-
"rewards/accuracies": 0.
|
368 |
-
"rewards/chosen": 0.
|
369 |
-
"rewards/margins": 0.
|
370 |
-
"rewards/rejected": 0.
|
371 |
"step": 210
|
372 |
},
|
373 |
{
|
374 |
"epoch": 0.57,
|
375 |
-
"grad_norm":
|
376 |
"learning_rate": 2.3185646976551794e-06,
|
377 |
-
"logits/chosen": 82.
|
378 |
-
"logits/rejected": 82.
|
379 |
-
"logps/chosen": -30.
|
380 |
-
"logps/rejected": -30.
|
381 |
-
"loss": 0.
|
382 |
-
"rewards/accuracies": 0.
|
383 |
-
"rewards/chosen":
|
384 |
-
"rewards/margins":
|
385 |
-
"rewards/rejected": 0.
|
386 |
"step": 220
|
387 |
},
|
388 |
{
|
389 |
"epoch": 0.6,
|
390 |
-
"grad_norm":
|
391 |
"learning_rate": 2.0932279108998323e-06,
|
392 |
-
"logits/chosen": 79.
|
393 |
-
"logits/rejected": 79.
|
394 |
-
"logps/chosen": -31.
|
395 |
-
"logps/rejected": -33.
|
396 |
-
"loss": 0.
|
397 |
-
"rewards/accuracies": 0.
|
398 |
-
"rewards/chosen": 0.
|
399 |
-
"rewards/margins": 0.
|
400 |
-
"rewards/rejected": 0.
|
401 |
"step": 230
|
402 |
},
|
403 |
{
|
404 |
"epoch": 0.62,
|
405 |
-
"grad_norm":
|
406 |
"learning_rate": 1.8712423238279358e-06,
|
407 |
-
"logits/chosen": 81.
|
408 |
-
"logits/rejected": 82.
|
409 |
-
"logps/chosen": -
|
410 |
-
"logps/rejected": -31.
|
411 |
-
"loss": 0.
|
412 |
-
"rewards/accuracies": 0.
|
413 |
-
"rewards/chosen":
|
414 |
-
"rewards/margins":
|
415 |
-
"rewards/rejected": 0.
|
416 |
"step": 240
|
417 |
},
|
418 |
{
|
419 |
"epoch": 0.65,
|
420 |
-
"grad_norm":
|
421 |
"learning_rate": 1.6544367689701824e-06,
|
422 |
-
"logits/chosen": 80.
|
423 |
-
"logits/rejected": 80.
|
424 |
-
"logps/chosen": -26.
|
425 |
-
"logps/rejected": -29.
|
426 |
-
"loss": 0.
|
427 |
-
"rewards/accuracies": 0.
|
428 |
-
"rewards/chosen": 0.
|
429 |
-
"rewards/margins": 0.
|
430 |
-
"rewards/rejected": 0.
|
431 |
"step": 250
|
432 |
},
|
433 |
{
|
434 |
"epoch": 0.68,
|
435 |
-
"grad_norm":
|
436 |
"learning_rate": 1.4445974030621963e-06,
|
437 |
-
"logits/chosen": 77.
|
438 |
-
"logits/rejected": 77.
|
439 |
-
"logps/chosen": -29.
|
440 |
-
"logps/rejected": -
|
441 |
-
"loss": 0.
|
442 |
-
"rewards/accuracies": 0.
|
443 |
-
"rewards/chosen":
|
444 |
-
"rewards/margins":
|
445 |
-
"rewards/rejected": 0.
|
446 |
"step": 260
|
447 |
},
|
448 |
{
|
449 |
"epoch": 0.7,
|
450 |
-
"grad_norm":
|
451 |
"learning_rate": 1.243452991757889e-06,
|
452 |
-
"logits/chosen": 77.
|
453 |
-
"logits/rejected": 77.
|
454 |
-
"logps/chosen": -30.
|
455 |
-
"logps/rejected": -31.
|
456 |
-
"loss": 0.
|
457 |
-
"rewards/accuracies": 0.
|
458 |
-
"rewards/chosen":
|
459 |
-
"rewards/margins":
|
460 |
-
"rewards/rejected": 0.
|
461 |
"step": 270
|
462 |
},
|
463 |
{
|
464 |
"epoch": 0.73,
|
465 |
-
"grad_norm":
|
466 |
"learning_rate": 1.0526606671603523e-06,
|
467 |
-
"logits/chosen": 79.
|
468 |
-
"logits/rejected": 79.
|
469 |
-
"logps/chosen": -30.
|
470 |
-
"logps/rejected": -29.
|
471 |
-
"loss": 0.
|
472 |
-
"rewards/accuracies": 0.
|
473 |
-
"rewards/chosen": 0.
|
474 |
-
"rewards/margins": 0.
|
475 |
-
"rewards/rejected": 0.
|
476 |
"step": 280
|
477 |
},
|
478 |
{
|
479 |
"epoch": 0.75,
|
480 |
-
"grad_norm":
|
481 |
"learning_rate": 8.737922755071455e-07,
|
482 |
-
"logits/chosen": 80.
|
483 |
-
"logits/rejected": 79.
|
484 |
-
"logps/chosen": -32.
|
485 |
-
"logps/rejected": -32.
|
486 |
-
"loss": 0.
|
487 |
-
"rewards/accuracies": 0.
|
488 |
-
"rewards/chosen":
|
489 |
-
"rewards/margins":
|
490 |
-
"rewards/rejected": -0.
|
491 |
"step": 290
|
492 |
},
|
493 |
{
|
494 |
"epoch": 0.78,
|
495 |
-
"grad_norm":
|
496 |
"learning_rate": 7.08321427484816e-07,
|
497 |
-
"logits/chosen": 75.
|
498 |
-
"logits/rejected": 75.
|
499 |
-
"logps/chosen": -31.
|
500 |
-
"logps/rejected": -
|
501 |
-
"loss": 0.
|
502 |
-
"rewards/accuracies": 0.
|
503 |
-
"rewards/chosen":
|
504 |
-
"rewards/margins":
|
505 |
-
"rewards/rejected": 0.
|
506 |
"step": 300
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.78,
|
510 |
-
"eval_logits/chosen": 98.
|
511 |
-
"eval_logits/rejected": 98.
|
512 |
-
"eval_logps/chosen": -32.
|
513 |
-
"eval_logps/rejected": -35.
|
514 |
-
"eval_loss": 0.
|
515 |
-
"eval_rewards/accuracies": 0.
|
516 |
-
"eval_rewards/chosen": 0.
|
517 |
-
"eval_rewards/margins":
|
518 |
-
"eval_rewards/rejected": 0.
|
519 |
-
"eval_runtime": 104.
|
520 |
-
"eval_samples_per_second": 3.
|
521 |
"eval_steps_per_second": 0.413,
|
522 |
"step": 300
|
523 |
},
|
524 |
{
|
525 |
"epoch": 0.81,
|
526 |
-
"grad_norm":
|
527 |
"learning_rate": 5.576113578589035e-07,
|
528 |
-
"logits/chosen": 82.
|
529 |
-
"logits/rejected": 82.
|
530 |
-
"logps/chosen": -29.
|
531 |
-
"logps/rejected": -32.
|
532 |
-
"loss": 0.
|
533 |
-
"rewards/accuracies": 0.
|
534 |
-
"rewards/chosen": 0.
|
535 |
-
"rewards/margins":
|
536 |
-
"rewards/rejected":
|
537 |
"step": 310
|
538 |
},
|
539 |
{
|
540 |
"epoch": 0.83,
|
541 |
-
"grad_norm":
|
542 |
"learning_rate": 4.229036944380913e-07,
|
543 |
-
"logits/chosen": 80.
|
544 |
-
"logits/rejected": 80.
|
545 |
-
"logps/chosen": -29.
|
546 |
-
"logps/rejected": -28.
|
547 |
-
"loss": 0.
|
548 |
-
"rewards/accuracies": 0.
|
549 |
-
"rewards/chosen":
|
550 |
-
"rewards/margins":
|
551 |
-
"rewards/rejected": 0.
|
552 |
"step": 320
|
553 |
},
|
554 |
{
|
555 |
"epoch": 0.86,
|
556 |
-
"grad_norm":
|
557 |
"learning_rate": 3.053082288996112e-07,
|
558 |
-
"logits/chosen": 77.
|
559 |
-
"logits/rejected": 77.
|
560 |
-
"logps/chosen": -28.
|
561 |
-
"logps/rejected": -32.
|
562 |
-
"loss": 0.
|
563 |
-
"rewards/accuracies": 0.
|
564 |
-
"rewards/chosen":
|
565 |
-
"rewards/margins":
|
566 |
-
"rewards/rejected":
|
567 |
"step": 330
|
568 |
},
|
569 |
{
|
570 |
"epoch": 0.88,
|
571 |
-
"grad_norm":
|
572 |
"learning_rate": 2.0579377374915805e-07,
|
573 |
-
"logits/chosen": 81.
|
574 |
-
"logits/rejected": 81.
|
575 |
-
"logps/chosen": -31.
|
576 |
-
"logps/rejected": -33.
|
577 |
-
"loss": 0.
|
578 |
-
"rewards/accuracies": 0.
|
579 |
-
"rewards/chosen":
|
580 |
-
"rewards/margins":
|
581 |
-
"rewards/rejected": 0.
|
582 |
"step": 340
|
583 |
},
|
584 |
{
|
585 |
"epoch": 0.91,
|
586 |
-
"grad_norm":
|
587 |
"learning_rate": 1.2518018074041684e-07,
|
588 |
-
"logits/chosen": 80.
|
589 |
-
"logits/rejected": 80.
|
590 |
-
"logps/chosen": -31.
|
591 |
-
"logps/rejected": -32.
|
592 |
-
"loss": 0.
|
593 |
-
"rewards/accuracies": 0.
|
594 |
-
"rewards/chosen":
|
595 |
-
"rewards/margins":
|
596 |
-
"rewards/rejected": 0.
|
597 |
"step": 350
|
598 |
},
|
599 |
{
|
600 |
"epoch": 0.94,
|
601 |
-
"grad_norm":
|
602 |
"learning_rate": 6.41315865106129e-08,
|
603 |
-
"logits/chosen": 82.
|
604 |
-
"logits/rejected": 82.
|
605 |
-
"logps/chosen": -27.
|
606 |
-
"logps/rejected": -31.
|
607 |
-
"loss": 0.
|
608 |
-
"rewards/accuracies": 0.
|
609 |
-
"rewards/chosen":
|
610 |
-
"rewards/margins":
|
611 |
-
"rewards/rejected": 0.
|
612 |
"step": 360
|
613 |
},
|
614 |
{
|
615 |
"epoch": 0.96,
|
616 |
-
"grad_norm":
|
617 |
"learning_rate": 2.3150941078050325e-08,
|
618 |
-
"logits/chosen": 81.
|
619 |
-
"logits/rejected": 81.
|
620 |
-
"logps/chosen": -31.
|
621 |
-
"logps/rejected": -34.
|
622 |
-
"loss": 0.
|
623 |
-
"rewards/accuracies": 0.
|
624 |
-
"rewards/chosen":
|
625 |
-
"rewards/margins": 0.
|
626 |
-
"rewards/rejected": 0.
|
627 |
"step": 370
|
628 |
},
|
629 |
{
|
630 |
"epoch": 0.99,
|
631 |
-
"grad_norm":
|
632 |
"learning_rate": 2.575864278703266e-09,
|
633 |
-
"logits/chosen": 75.
|
634 |
-
"logits/rejected": 75.
|
635 |
-
"logps/chosen": -29.
|
636 |
-
"logps/rejected": -27.
|
637 |
-
"loss": 0.
|
638 |
-
"rewards/accuracies": 0.
|
639 |
-
"rewards/chosen":
|
640 |
-
"rewards/margins": 0.
|
641 |
-
"rewards/rejected": 0.
|
642 |
"step": 380
|
643 |
},
|
644 |
{
|
645 |
"epoch": 1.0,
|
646 |
"step": 385,
|
647 |
"total_flos": 0.0,
|
648 |
-
"train_loss": 0.
|
649 |
-
"train_runtime": 2559.
|
650 |
"train_samples_per_second": 1.203,
|
651 |
"train_steps_per_second": 0.15
|
652 |
}
|
|
|
10 |
"log_history": [
|
11 |
{
|
12 |
"epoch": 0.0,
|
13 |
+
"grad_norm": 13.0625,
|
14 |
"learning_rate": 1.282051282051282e-07,
|
15 |
"logits/chosen": 88.18099975585938,
|
16 |
"logits/rejected": 88.25153350830078,
|
|
|
25 |
},
|
26 |
{
|
27 |
"epoch": 0.03,
|
28 |
+
"grad_norm": 5.375,
|
29 |
"learning_rate": 1.282051282051282e-06,
|
30 |
+
"logits/chosen": 81.08451080322266,
|
31 |
+
"logits/rejected": 80.78488159179688,
|
32 |
+
"logps/chosen": -34.28126525878906,
|
33 |
+
"logps/rejected": -33.13351058959961,
|
34 |
+
"loss": 0.492,
|
35 |
+
"rewards/accuracies": 0.4861111044883728,
|
36 |
+
"rewards/chosen": -0.015815330669283867,
|
37 |
+
"rewards/margins": 0.05422591418027878,
|
38 |
+
"rewards/rejected": -0.0700412467122078,
|
39 |
"step": 10
|
40 |
},
|
41 |
{
|
42 |
"epoch": 0.05,
|
43 |
+
"grad_norm": 9.6875,
|
44 |
"learning_rate": 2.564102564102564e-06,
|
45 |
+
"logits/chosen": 80.63957214355469,
|
46 |
+
"logits/rejected": 80.53150939941406,
|
47 |
+
"logps/chosen": -33.59695816040039,
|
48 |
+
"logps/rejected": -30.76776123046875,
|
49 |
+
"loss": 0.4944,
|
50 |
+
"rewards/accuracies": 0.5,
|
51 |
+
"rewards/chosen": 0.03718096390366554,
|
52 |
+
"rewards/margins": 0.03328876942396164,
|
53 |
+
"rewards/rejected": 0.003892195178195834,
|
54 |
"step": 20
|
55 |
},
|
56 |
{
|
57 |
"epoch": 0.08,
|
58 |
+
"grad_norm": 10.5,
|
59 |
"learning_rate": 3.846153846153847e-06,
|
60 |
+
"logits/chosen": 82.47281646728516,
|
61 |
+
"logits/rejected": 82.50285339355469,
|
62 |
+
"logps/chosen": -33.86055374145508,
|
63 |
+
"logps/rejected": -31.097183227539062,
|
64 |
+
"loss": 0.5134,
|
65 |
+
"rewards/accuracies": 0.4000000059604645,
|
66 |
+
"rewards/chosen": 0.06090690940618515,
|
67 |
+
"rewards/margins": -0.0697535052895546,
|
68 |
+
"rewards/rejected": 0.13066044449806213,
|
69 |
"step": 30
|
70 |
},
|
71 |
{
|
72 |
"epoch": 0.1,
|
73 |
+
"grad_norm": 10.875,
|
74 |
"learning_rate": 4.999896948438434e-06,
|
75 |
+
"logits/chosen": 81.05327606201172,
|
76 |
+
"logits/rejected": 81.04902648925781,
|
77 |
+
"logps/chosen": -32.787689208984375,
|
78 |
+
"logps/rejected": -33.05846405029297,
|
79 |
+
"loss": 0.4913,
|
80 |
+
"rewards/accuracies": 0.5,
|
81 |
+
"rewards/chosen": 0.15454962849617004,
|
82 |
+
"rewards/margins": 0.030526524409651756,
|
83 |
+
"rewards/rejected": 0.12402307987213135,
|
84 |
"step": 40
|
85 |
},
|
86 |
{
|
87 |
"epoch": 0.13,
|
88 |
+
"grad_norm": 8.125,
|
89 |
"learning_rate": 4.987541037542187e-06,
|
90 |
+
"logits/chosen": 78.79315185546875,
|
91 |
+
"logits/rejected": 78.80322265625,
|
92 |
+
"logps/chosen": -30.44384765625,
|
93 |
+
"logps/rejected": -30.80826759338379,
|
94 |
+
"loss": 0.4599,
|
95 |
+
"rewards/accuracies": 0.5874999761581421,
|
96 |
+
"rewards/chosen": 0.30455905199050903,
|
97 |
+
"rewards/margins": 0.19855086505413055,
|
98 |
+
"rewards/rejected": 0.10600819438695908,
|
99 |
"step": 50
|
100 |
},
|
101 |
{
|
102 |
"epoch": 0.16,
|
103 |
+
"grad_norm": 9.625,
|
104 |
"learning_rate": 4.954691471941119e-06,
|
105 |
+
"logits/chosen": 83.44609069824219,
|
106 |
+
"logits/rejected": 83.50300598144531,
|
107 |
+
"logps/chosen": -30.81852149963379,
|
108 |
+
"logps/rejected": -29.231618881225586,
|
109 |
+
"loss": 0.4994,
|
110 |
+
"rewards/accuracies": 0.4124999940395355,
|
111 |
+
"rewards/chosen": 0.1426558792591095,
|
112 |
+
"rewards/margins": -0.0009051367524079978,
|
113 |
+
"rewards/rejected": 0.1435610055923462,
|
114 |
"step": 60
|
115 |
},
|
116 |
{
|
117 |
"epoch": 0.18,
|
118 |
+
"grad_norm": 7.875,
|
119 |
"learning_rate": 4.901618883413549e-06,
|
120 |
+
"logits/chosen": 84.02252960205078,
|
121 |
+
"logits/rejected": 84.05018615722656,
|
122 |
+
"logps/chosen": -30.281408309936523,
|
123 |
+
"logps/rejected": -32.76702880859375,
|
124 |
+
"loss": 0.4981,
|
125 |
+
"rewards/accuracies": 0.4375,
|
126 |
+
"rewards/chosen": 0.13904651999473572,
|
127 |
+
"rewards/margins": 0.03077618218958378,
|
128 |
+
"rewards/rejected": 0.10827036201953888,
|
129 |
"step": 70
|
130 |
},
|
131 |
{
|
132 |
"epoch": 0.21,
|
133 |
+
"grad_norm": 9.5,
|
134 |
"learning_rate": 4.828760511501322e-06,
|
135 |
+
"logits/chosen": 81.84428405761719,
|
136 |
+
"logits/rejected": 81.82522583007812,
|
137 |
+
"logps/chosen": -31.032154083251953,
|
138 |
+
"logps/rejected": -30.712255477905273,
|
139 |
+
"loss": 0.4648,
|
140 |
+
"rewards/accuracies": 0.5874999761581421,
|
141 |
+
"rewards/chosen": 0.18911631405353546,
|
142 |
+
"rewards/margins": 0.1709602177143097,
|
143 |
+
"rewards/rejected": 0.018156107515096664,
|
144 |
"step": 80
|
145 |
},
|
146 |
{
|
147 |
"epoch": 0.23,
|
148 |
+
"grad_norm": 12.375,
|
149 |
"learning_rate": 4.7367166013034295e-06,
|
150 |
+
"logits/chosen": 78.68492126464844,
|
151 |
+
"logits/rejected": 78.66017150878906,
|
152 |
+
"logps/chosen": -32.12371063232422,
|
153 |
+
"logps/rejected": -30.837940216064453,
|
154 |
+
"loss": 0.4704,
|
155 |
+
"rewards/accuracies": 0.574999988079071,
|
156 |
+
"rewards/chosen": 0.20578794181346893,
|
157 |
+
"rewards/margins": 0.16037426888942719,
|
158 |
+
"rewards/rejected": 0.04541371017694473,
|
159 |
"step": 90
|
160 |
},
|
161 |
{
|
162 |
"epoch": 0.26,
|
163 |
+
"grad_norm": 9.9375,
|
164 |
"learning_rate": 4.626245458345211e-06,
|
165 |
+
"logits/chosen": 83.77175903320312,
|
166 |
+
"logits/rejected": 83.79080963134766,
|
167 |
+
"logps/chosen": -33.829063415527344,
|
168 |
+
"logps/rejected": -31.544397354125977,
|
169 |
+
"loss": 0.4749,
|
170 |
+
"rewards/accuracies": 0.5625,
|
171 |
+
"rewards/chosen": 0.1903691589832306,
|
172 |
+
"rewards/margins": 0.10790882259607315,
|
173 |
+
"rewards/rejected": 0.08246034383773804,
|
174 |
"step": 100
|
175 |
},
|
176 |
{
|
177 |
"epoch": 0.26,
|
178 |
+
"eval_logits/chosen": 98.75000762939453,
|
179 |
+
"eval_logits/rejected": 98.7376708984375,
|
180 |
+
"eval_logps/chosen": -32.41049575805664,
|
181 |
+
"eval_logps/rejected": -35.85907745361328,
|
182 |
+
"eval_loss": 0.504712700843811,
|
183 |
+
"eval_rewards/accuracies": 0.4746677577495575,
|
184 |
+
"eval_rewards/chosen": 0.013071590103209019,
|
185 |
+
"eval_rewards/margins": -0.02990747056901455,
|
186 |
+
"eval_rewards/rejected": 0.042979057878255844,
|
187 |
+
"eval_runtime": 104.3402,
|
188 |
+
"eval_samples_per_second": 3.287,
|
189 |
+
"eval_steps_per_second": 0.412,
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
+
"grad_norm": 11.875,
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
+
"logits/chosen": 84.02198028564453,
|
197 |
+
"logits/rejected": 83.90216064453125,
|
198 |
+
"logps/chosen": -32.067047119140625,
|
199 |
+
"logps/rejected": -32.60837936401367,
|
200 |
+
"loss": 0.4251,
|
201 |
+
"rewards/accuracies": 0.574999988079071,
|
202 |
+
"rewards/chosen": 0.382302463054657,
|
203 |
+
"rewards/margins": 0.36329856514930725,
|
204 |
+
"rewards/rejected": 0.01900387369096279,
|
205 |
"step": 110
|
206 |
},
|
207 |
{
|
208 |
"epoch": 0.31,
|
209 |
+
"grad_norm": 10.9375,
|
210 |
"learning_rate": 4.353806263777678e-06,
|
211 |
+
"logits/chosen": 84.06869506835938,
|
212 |
+
"logits/rejected": 84.17839050292969,
|
213 |
+
"logps/chosen": -28.029781341552734,
|
214 |
+
"logps/rejected": -35.15131378173828,
|
215 |
+
"loss": 0.4499,
|
216 |
+
"rewards/accuracies": 0.5874999761581421,
|
217 |
+
"rewards/chosen": 0.35991987586021423,
|
218 |
+
"rewards/margins": 0.23678629100322723,
|
219 |
+
"rewards/rejected": 0.123133584856987,
|
220 |
"step": 120
|
221 |
},
|
222 |
{
|
223 |
"epoch": 0.34,
|
224 |
+
"grad_norm": 9.125,
|
225 |
"learning_rate": 4.1940827077152755e-06,
|
226 |
+
"logits/chosen": 81.26686096191406,
|
227 |
+
"logits/rejected": 81.29707336425781,
|
228 |
+
"logps/chosen": -30.0821533203125,
|
229 |
+
"logps/rejected": -31.722143173217773,
|
230 |
+
"loss": 0.4515,
|
231 |
+
"rewards/accuracies": 0.637499988079071,
|
232 |
+
"rewards/chosen": 0.3300797641277313,
|
233 |
+
"rewards/margins": 0.24628591537475586,
|
234 |
+
"rewards/rejected": 0.08379384875297546,
|
235 |
"step": 130
|
236 |
},
|
237 |
{
|
238 |
"epoch": 0.36,
|
239 |
+
"grad_norm": 7.09375,
|
240 |
"learning_rate": 4.0204024186666215e-06,
|
241 |
+
"logits/chosen": 82.26895904541016,
|
242 |
+
"logits/rejected": 82.28730773925781,
|
243 |
+
"logps/chosen": -26.675317764282227,
|
244 |
+
"logps/rejected": -32.535396575927734,
|
245 |
+
"loss": 0.4328,
|
246 |
+
"rewards/accuracies": 0.5874999761581421,
|
247 |
+
"rewards/chosen": 0.30880242586135864,
|
248 |
+
"rewards/margins": 0.35360854864120483,
|
249 |
+
"rewards/rejected": -0.044806141406297684,
|
250 |
"step": 140
|
251 |
},
|
252 |
{
|
253 |
"epoch": 0.39,
|
254 |
+
"grad_norm": 8.75,
|
255 |
"learning_rate": 3.834196265035119e-06,
|
256 |
+
"logits/chosen": 80.73310089111328,
|
257 |
+
"logits/rejected": 80.70652770996094,
|
258 |
+
"logps/chosen": -28.468103408813477,
|
259 |
+
"logps/rejected": -32.77440643310547,
|
260 |
+
"loss": 0.4125,
|
261 |
+
"rewards/accuracies": 0.699999988079071,
|
262 |
+
"rewards/chosen": 0.36840274930000305,
|
263 |
+
"rewards/margins": 0.45156532526016235,
|
264 |
+
"rewards/rejected": -0.08316260576248169,
|
265 |
"step": 150
|
266 |
},
|
267 |
{
|
268 |
"epoch": 0.42,
|
269 |
+
"grad_norm": 8.5,
|
270 |
"learning_rate": 3.636998309800573e-06,
|
271 |
+
"logits/chosen": 82.41865539550781,
|
272 |
+
"logits/rejected": 82.44725799560547,
|
273 |
+
"logps/chosen": -33.112274169921875,
|
274 |
+
"logps/rejected": -30.1284122467041,
|
275 |
+
"loss": 0.4055,
|
276 |
+
"rewards/accuracies": 0.7250000238418579,
|
277 |
+
"rewards/chosen": 0.40979090332984924,
|
278 |
+
"rewards/margins": 0.4801415503025055,
|
279 |
+
"rewards/rejected": -0.07035063952207565,
|
280 |
"step": 160
|
281 |
},
|
282 |
{
|
283 |
"epoch": 0.44,
|
284 |
+
"grad_norm": 7.5625,
|
285 |
"learning_rate": 3.4304331721118078e-06,
|
286 |
+
"logits/chosen": 83.09947204589844,
|
287 |
+
"logits/rejected": 83.05732727050781,
|
288 |
+
"logps/chosen": -30.413660049438477,
|
289 |
+
"logps/rejected": -32.43395233154297,
|
290 |
+
"loss": 0.4144,
|
291 |
"rewards/accuracies": 0.6499999761581421,
|
292 |
+
"rewards/chosen": 0.33825185894966125,
|
293 |
+
"rewards/margins": 0.5038093328475952,
|
294 |
+
"rewards/rejected": -0.16555748879909515,
|
295 |
"step": 170
|
296 |
},
|
297 |
{
|
298 |
"epoch": 0.47,
|
299 |
+
"grad_norm": 6.46875,
|
300 |
"learning_rate": 3.2162026428305436e-06,
|
301 |
+
"logits/chosen": 80.59300231933594,
|
302 |
+
"logits/rejected": 80.56774139404297,
|
303 |
+
"logps/chosen": -30.52907371520996,
|
304 |
+
"logps/rejected": -31.434350967407227,
|
305 |
+
"loss": 0.433,
|
306 |
+
"rewards/accuracies": 0.637499988079071,
|
307 |
+
"rewards/chosen": 0.2672317624092102,
|
308 |
+
"rewards/margins": 0.3780880868434906,
|
309 |
+
"rewards/rejected": -0.1108563169836998,
|
310 |
"step": 180
|
311 |
},
|
312 |
{
|
313 |
"epoch": 0.49,
|
314 |
+
"grad_norm": 5.40625,
|
315 |
"learning_rate": 2.996071664294641e-06,
|
316 |
+
"logits/chosen": 82.23339080810547,
|
317 |
+
"logits/rejected": 82.1915283203125,
|
318 |
+
"logps/chosen": -29.8317813873291,
|
319 |
+
"logps/rejected": -30.190067291259766,
|
320 |
+
"loss": 0.4509,
|
321 |
"rewards/accuracies": 0.612500011920929,
|
322 |
+
"rewards/chosen": 0.37385472655296326,
|
323 |
+
"rewards/margins": 0.26267164945602417,
|
324 |
+
"rewards/rejected": 0.1111830323934555,
|
325 |
"step": 190
|
326 |
},
|
327 |
{
|
328 |
"epoch": 0.52,
|
329 |
+
"grad_norm": 6.625,
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
+
"logits/chosen": 77.5392074584961,
|
332 |
+
"logits/rejected": 77.48648834228516,
|
333 |
+
"logps/chosen": -33.0440788269043,
|
334 |
+
"logps/rejected": -32.247684478759766,
|
335 |
+
"loss": 0.3843,
|
336 |
"rewards/accuracies": 0.699999988079071,
|
337 |
+
"rewards/chosen": 0.6834636926651001,
|
338 |
+
"rewards/margins": 0.5851765871047974,
|
339 |
+
"rewards/rejected": 0.09828709810972214,
|
340 |
"step": 200
|
341 |
},
|
342 |
{
|
343 |
"epoch": 0.52,
|
344 |
+
"eval_logits/chosen": 98.38356018066406,
|
345 |
+
"eval_logits/rejected": 98.37836456298828,
|
346 |
+
"eval_logps/chosen": -32.3836784362793,
|
347 |
+
"eval_logps/rejected": -35.98166275024414,
|
348 |
+
"eval_loss": 0.4944417476654053,
|
349 |
+
"eval_rewards/accuracies": 0.5307309031486511,
|
350 |
+
"eval_rewards/chosen": 0.023798126727342606,
|
351 |
+
"eval_rewards/margins": 0.029852891340851784,
|
352 |
+
"eval_rewards/rejected": -0.00605476601049304,
|
353 |
+
"eval_runtime": 104.0926,
|
354 |
+
"eval_samples_per_second": 3.295,
|
355 |
"eval_steps_per_second": 0.413,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
+
"grad_norm": 12.3125,
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
+
"logits/chosen": 80.09881591796875,
|
363 |
+
"logits/rejected": 80.01982879638672,
|
364 |
+
"logps/chosen": -32.544151306152344,
|
365 |
+
"logps/rejected": -35.001224517822266,
|
366 |
+
"loss": 0.3882,
|
367 |
+
"rewards/accuracies": 0.612500011920929,
|
368 |
+
"rewards/chosen": 0.5606266260147095,
|
369 |
+
"rewards/margins": 0.5666781663894653,
|
370 |
+
"rewards/rejected": -0.006051521748304367,
|
371 |
"step": 210
|
372 |
},
|
373 |
{
|
374 |
"epoch": 0.57,
|
375 |
+
"grad_norm": 8.1875,
|
376 |
"learning_rate": 2.3185646976551794e-06,
|
377 |
+
"logits/chosen": 82.21588134765625,
|
378 |
+
"logits/rejected": 82.28758239746094,
|
379 |
+
"logps/chosen": -30.38724136352539,
|
380 |
+
"logps/rejected": -30.862102508544922,
|
381 |
+
"loss": 0.3707,
|
382 |
+
"rewards/accuracies": 0.762499988079071,
|
383 |
+
"rewards/chosen": 0.6246866583824158,
|
384 |
+
"rewards/margins": 0.6779459714889526,
|
385 |
+
"rewards/rejected": -0.05325937271118164,
|
386 |
"step": 220
|
387 |
},
|
388 |
{
|
389 |
"epoch": 0.6,
|
390 |
+
"grad_norm": 8.5,
|
391 |
"learning_rate": 2.0932279108998323e-06,
|
392 |
+
"logits/chosen": 79.33607482910156,
|
393 |
+
"logits/rejected": 79.39179992675781,
|
394 |
+
"logps/chosen": -31.715646743774414,
|
395 |
+
"logps/rejected": -33.92882537841797,
|
396 |
+
"loss": 0.4251,
|
397 |
+
"rewards/accuracies": 0.625,
|
398 |
+
"rewards/chosen": 0.4360387325286865,
|
399 |
+
"rewards/margins": 0.40270787477493286,
|
400 |
+
"rewards/rejected": 0.03333085775375366,
|
401 |
"step": 230
|
402 |
},
|
403 |
{
|
404 |
"epoch": 0.62,
|
405 |
+
"grad_norm": 10.625,
|
406 |
"learning_rate": 1.8712423238279358e-06,
|
407 |
+
"logits/chosen": 81.82881164550781,
|
408 |
+
"logits/rejected": 82.10590362548828,
|
409 |
+
"logps/chosen": -29.962848663330078,
|
410 |
+
"logps/rejected": -31.462871551513672,
|
411 |
+
"loss": 0.3667,
|
412 |
+
"rewards/accuracies": 0.7875000238418579,
|
413 |
+
"rewards/chosen": 0.6900869607925415,
|
414 |
+
"rewards/margins": 0.6624192595481873,
|
415 |
+
"rewards/rejected": 0.027667587623000145,
|
416 |
"step": 240
|
417 |
},
|
418 |
{
|
419 |
"epoch": 0.65,
|
420 |
+
"grad_norm": 11.25,
|
421 |
"learning_rate": 1.6544367689701824e-06,
|
422 |
+
"logits/chosen": 80.50505065917969,
|
423 |
+
"logits/rejected": 80.56990051269531,
|
424 |
+
"logps/chosen": -26.388418197631836,
|
425 |
+
"logps/rejected": -29.717565536499023,
|
426 |
+
"loss": 0.4245,
|
427 |
+
"rewards/accuracies": 0.625,
|
428 |
+
"rewards/chosen": 0.48299089074134827,
|
429 |
+
"rewards/margins": 0.40972191095352173,
|
430 |
+
"rewards/rejected": 0.07326899468898773,
|
431 |
"step": 250
|
432 |
},
|
433 |
{
|
434 |
"epoch": 0.68,
|
435 |
+
"grad_norm": 8.625,
|
436 |
"learning_rate": 1.4445974030621963e-06,
|
437 |
+
"logits/chosen": 77.76419830322266,
|
438 |
+
"logits/rejected": 77.94038391113281,
|
439 |
+
"logps/chosen": -29.758447647094727,
|
440 |
+
"logps/rejected": -36.150657653808594,
|
441 |
+
"loss": 0.3604,
|
442 |
+
"rewards/accuracies": 0.7124999761581421,
|
443 |
+
"rewards/chosen": 0.7210196256637573,
|
444 |
+
"rewards/margins": 0.7406320571899414,
|
445 |
+
"rewards/rejected": -0.0196123905479908,
|
446 |
"step": 260
|
447 |
},
|
448 |
{
|
449 |
"epoch": 0.7,
|
450 |
+
"grad_norm": 6.1875,
|
451 |
"learning_rate": 1.243452991757889e-06,
|
452 |
+
"logits/chosen": 77.1490707397461,
|
453 |
+
"logits/rejected": 77.18915557861328,
|
454 |
+
"logps/chosen": -30.186153411865234,
|
455 |
+
"logps/rejected": -31.373126983642578,
|
456 |
+
"loss": 0.3871,
|
457 |
+
"rewards/accuracies": 0.699999988079071,
|
458 |
+
"rewards/chosen": 0.6371081471443176,
|
459 |
+
"rewards/margins": 0.5841903686523438,
|
460 |
+
"rewards/rejected": 0.05291769653558731,
|
461 |
"step": 270
|
462 |
},
|
463 |
{
|
464 |
"epoch": 0.73,
|
465 |
+
"grad_norm": 12.5,
|
466 |
"learning_rate": 1.0526606671603523e-06,
|
467 |
+
"logits/chosen": 79.78591918945312,
|
468 |
+
"logits/rejected": 79.5605239868164,
|
469 |
+
"logps/chosen": -30.330188751220703,
|
470 |
+
"logps/rejected": -29.223669052124023,
|
471 |
+
"loss": 0.4182,
|
472 |
+
"rewards/accuracies": 0.675000011920929,
|
473 |
+
"rewards/chosen": 0.5918818712234497,
|
474 |
+
"rewards/margins": 0.41455134749412537,
|
475 |
+
"rewards/rejected": 0.17733044922351837,
|
476 |
"step": 280
|
477 |
},
|
478 |
{
|
479 |
"epoch": 0.75,
|
480 |
+
"grad_norm": 8.4375,
|
481 |
"learning_rate": 8.737922755071455e-07,
|
482 |
+
"logits/chosen": 80.03771209716797,
|
483 |
+
"logits/rejected": 79.94876861572266,
|
484 |
+
"logps/chosen": -32.170814514160156,
|
485 |
+
"logps/rejected": -32.12440872192383,
|
486 |
+
"loss": 0.3449,
|
487 |
+
"rewards/accuracies": 0.824999988079071,
|
488 |
+
"rewards/chosen": 0.7733729481697083,
|
489 |
+
"rewards/margins": 0.8163889646530151,
|
490 |
+
"rewards/rejected": -0.04301605746150017,
|
491 |
"step": 290
|
492 |
},
|
493 |
{
|
494 |
"epoch": 0.78,
|
495 |
+
"grad_norm": 5.5,
|
496 |
"learning_rate": 7.08321427484816e-07,
|
497 |
+
"logits/chosen": 75.5302505493164,
|
498 |
+
"logits/rejected": 75.59835052490234,
|
499 |
+
"logps/chosen": -31.593181610107422,
|
500 |
+
"logps/rejected": -29.076126098632812,
|
501 |
+
"loss": 0.363,
|
502 |
+
"rewards/accuracies": 0.7749999761581421,
|
503 |
+
"rewards/chosen": 0.7250088453292847,
|
504 |
+
"rewards/margins": 0.714650571346283,
|
505 |
+
"rewards/rejected": 0.010358264669775963,
|
506 |
"step": 300
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.78,
|
510 |
+
"eval_logits/chosen": 98.36553192138672,
|
511 |
+
"eval_logits/rejected": 98.37035369873047,
|
512 |
+
"eval_logps/chosen": -32.229251861572266,
|
513 |
+
"eval_logps/rejected": -35.81660079956055,
|
514 |
+
"eval_loss": 0.49616533517837524,
|
515 |
+
"eval_rewards/accuracies": 0.5103820562362671,
|
516 |
+
"eval_rewards/chosen": 0.0855708047747612,
|
517 |
+
"eval_rewards/margins": 0.025601176545023918,
|
518 |
+
"eval_rewards/rejected": 0.05996962636709213,
|
519 |
+
"eval_runtime": 104.1169,
|
520 |
+
"eval_samples_per_second": 3.294,
|
521 |
"eval_steps_per_second": 0.413,
|
522 |
"step": 300
|
523 |
},
|
524 |
{
|
525 |
"epoch": 0.81,
|
526 |
+
"grad_norm": 8.0625,
|
527 |
"learning_rate": 5.576113578589035e-07,
|
528 |
+
"logits/chosen": 82.7135238647461,
|
529 |
+
"logits/rejected": 82.74400329589844,
|
530 |
+
"logps/chosen": -29.19759178161621,
|
531 |
+
"logps/rejected": -32.07235336303711,
|
532 |
+
"loss": 0.3811,
|
533 |
+
"rewards/accuracies": 0.75,
|
534 |
+
"rewards/chosen": 0.666375994682312,
|
535 |
+
"rewards/margins": 0.6154125332832336,
|
536 |
+
"rewards/rejected": 0.05096355825662613,
|
537 |
"step": 310
|
538 |
},
|
539 |
{
|
540 |
"epoch": 0.83,
|
541 |
+
"grad_norm": 8.5,
|
542 |
"learning_rate": 4.229036944380913e-07,
|
543 |
+
"logits/chosen": 80.10868835449219,
|
544 |
+
"logits/rejected": 80.10973358154297,
|
545 |
+
"logps/chosen": -29.693145751953125,
|
546 |
+
"logps/rejected": -28.737625122070312,
|
547 |
+
"loss": 0.3658,
|
548 |
+
"rewards/accuracies": 0.7124999761581421,
|
549 |
+
"rewards/chosen": 0.7621666193008423,
|
550 |
+
"rewards/margins": 0.6894195675849915,
|
551 |
+
"rewards/rejected": 0.07274699211120605,
|
552 |
"step": 320
|
553 |
},
|
554 |
{
|
555 |
"epoch": 0.86,
|
556 |
+
"grad_norm": 7.8125,
|
557 |
"learning_rate": 3.053082288996112e-07,
|
558 |
+
"logits/chosen": 77.36323547363281,
|
559 |
+
"logits/rejected": 77.391357421875,
|
560 |
+
"logps/chosen": -28.293853759765625,
|
561 |
+
"logps/rejected": -32.6038703918457,
|
562 |
+
"loss": 0.3393,
|
563 |
+
"rewards/accuracies": 0.737500011920929,
|
564 |
+
"rewards/chosen": 0.8235009908676147,
|
565 |
+
"rewards/margins": 0.8111019134521484,
|
566 |
+
"rewards/rejected": 0.012399068102240562,
|
567 |
"step": 330
|
568 |
},
|
569 |
{
|
570 |
"epoch": 0.88,
|
571 |
+
"grad_norm": 10.625,
|
572 |
"learning_rate": 2.0579377374915805e-07,
|
573 |
+
"logits/chosen": 81.62095642089844,
|
574 |
+
"logits/rejected": 81.64241027832031,
|
575 |
+
"logps/chosen": -31.521636962890625,
|
576 |
+
"logps/rejected": -33.3758659362793,
|
577 |
+
"loss": 0.377,
|
578 |
+
"rewards/accuracies": 0.6625000238418579,
|
579 |
+
"rewards/chosen": 0.6762043833732605,
|
580 |
+
"rewards/margins": 0.6790729761123657,
|
581 |
+
"rewards/rejected": -0.002868417650461197,
|
582 |
"step": 340
|
583 |
},
|
584 |
{
|
585 |
"epoch": 0.91,
|
586 |
+
"grad_norm": 7.71875,
|
587 |
"learning_rate": 1.2518018074041684e-07,
|
588 |
+
"logits/chosen": 80.68013763427734,
|
589 |
+
"logits/rejected": 80.69649505615234,
|
590 |
+
"logps/chosen": -31.641170501708984,
|
591 |
+
"logps/rejected": -32.8623161315918,
|
592 |
+
"loss": 0.3712,
|
593 |
+
"rewards/accuracies": 0.737500011920929,
|
594 |
+
"rewards/chosen": 0.8010267019271851,
|
595 |
+
"rewards/margins": 0.7092341780662537,
|
596 |
+
"rewards/rejected": 0.0917925089597702,
|
597 |
"step": 350
|
598 |
},
|
599 |
{
|
600 |
"epoch": 0.94,
|
601 |
+
"grad_norm": 6.25,
|
602 |
"learning_rate": 6.41315865106129e-08,
|
603 |
+
"logits/chosen": 82.13801574707031,
|
604 |
+
"logits/rejected": 82.19556427001953,
|
605 |
+
"logps/chosen": -27.77614402770996,
|
606 |
+
"logps/rejected": -31.526113510131836,
|
607 |
+
"loss": 0.369,
|
608 |
+
"rewards/accuracies": 0.75,
|
609 |
+
"rewards/chosen": 0.7104269862174988,
|
610 |
+
"rewards/margins": 0.6609331965446472,
|
611 |
+
"rewards/rejected": 0.04949387162923813,
|
612 |
"step": 360
|
613 |
},
|
614 |
{
|
615 |
"epoch": 0.96,
|
616 |
+
"grad_norm": 7.5,
|
617 |
"learning_rate": 2.3150941078050325e-08,
|
618 |
+
"logits/chosen": 81.63580322265625,
|
619 |
+
"logits/rejected": 81.66941833496094,
|
620 |
+
"logps/chosen": -31.2968692779541,
|
621 |
+
"logps/rejected": -34.87516403198242,
|
622 |
+
"loss": 0.4021,
|
623 |
+
"rewards/accuracies": 0.6499999761581421,
|
624 |
+
"rewards/chosen": 0.6317180395126343,
|
625 |
+
"rewards/margins": 0.5380347967147827,
|
626 |
+
"rewards/rejected": 0.09368324279785156,
|
627 |
"step": 370
|
628 |
},
|
629 |
{
|
630 |
"epoch": 0.99,
|
631 |
+
"grad_norm": 8.625,
|
632 |
"learning_rate": 2.575864278703266e-09,
|
633 |
+
"logits/chosen": 75.4756088256836,
|
634 |
+
"logits/rejected": 75.35409545898438,
|
635 |
+
"logps/chosen": -29.19217300415039,
|
636 |
+
"logps/rejected": -27.851299285888672,
|
637 |
+
"loss": 0.4126,
|
638 |
+
"rewards/accuracies": 0.6875,
|
639 |
+
"rewards/chosen": 0.5535811185836792,
|
640 |
+
"rewards/margins": 0.4298717975616455,
|
641 |
+
"rewards/rejected": 0.1237092986702919,
|
642 |
"step": 380
|
643 |
},
|
644 |
{
|
645 |
"epoch": 1.0,
|
646 |
"step": 385,
|
647 |
"total_flos": 0.0,
|
648 |
+
"train_loss": 0.4199758805237807,
|
649 |
+
"train_runtime": 2559.5376,
|
650 |
"train_samples_per_second": 1.203,
|
651 |
"train_steps_per_second": 0.15
|
652 |
}
|