hugodk-sch
commited on
Commit
•
40d6fc5
1
Parent(s):
7ea9a6c
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: 0.
|
28 |
-
- Logps/rejected: -36.
|
29 |
-
- Logps/chosen: -
|
30 |
-
- Logits/rejected:
|
31 |
-
- Logits/chosen:
|
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: 0.7744
|
22 |
+
- Rewards/chosen: -0.0855
|
23 |
+
- Rewards/rejected: -0.2061
|
24 |
+
- Rewards/accuracies: 0.5428
|
25 |
+
- Rewards/margins: 0.1206
|
26 |
+
- Logps/rejected: -36.2242
|
27 |
+
- Logps/chosen: -32.5501
|
28 |
+
- Logits/rejected: 98.7517
|
29 |
+
- Logits/chosen: 98.7722
|
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 |
+
| 0.7017 | 0.26 | 100 | 0.8030 | -0.0408 | -0.0544 | 0.5137 | 0.0136 | -36.0346 | -32.4942 | 98.7637 | 98.7765 |
|
65 |
+
| 0.6236 | 0.52 | 200 | 0.7719 | -0.1720 | -0.2931 | 0.5216 | 0.1211 | -36.3329 | -32.6582 | 98.7039 | 98.7271 |
|
66 |
+
| 0.5655 | 0.78 | 300 | 0.7744 | -0.0855 | -0.2061 | 0.5428 | 0.1206 | -36.2242 | -32.5501 | 98.7517 | 98.7722 |
|
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:fd5aae0188b2a0ac26a7deeeb9849d866486b8ff0df32512de4839b557ca80f2
|
3 |
size 671150064
|
all_results.json
CHANGED
@@ -1,20 +1,7 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"
|
4 |
-
"
|
5 |
-
"eval_logps/chosen": -33.003868103027344,
|
6 |
-
"eval_logps/rejected": -36.95783615112305,
|
7 |
-
"eval_loss": 0.6761856079101562,
|
8 |
-
"eval_rewards/accuracies": 0.5888704061508179,
|
9 |
-
"eval_rewards/chosen": -0.05606912076473236,
|
10 |
-
"eval_rewards/margins": 0.04306147247552872,
|
11 |
-
"eval_rewards/rejected": -0.09913058578968048,
|
12 |
-
"eval_runtime": 103.6924,
|
13 |
-
"eval_samples": 343,
|
14 |
-
"eval_samples_per_second": 3.308,
|
15 |
-
"eval_steps_per_second": 0.415,
|
16 |
-
"train_loss": 0.6461187455561254,
|
17 |
-
"train_runtime": 2556.0398,
|
18 |
"train_samples": 3079,
|
19 |
"train_samples_per_second": 1.205,
|
20 |
"train_steps_per_second": 0.151
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"train_loss": 0.6050039254225694,
|
4 |
+
"train_runtime": 2556.17,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.205,
|
7 |
"train_steps_per_second": 0.151
|
train_results.json
CHANGED
@@ -1,7 +1,7 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"train_loss": 0.
|
4 |
-
"train_runtime": 2556.
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.205,
|
7 |
"train_steps_per_second": 0.151
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"train_loss": 0.6050039254225694,
|
4 |
+
"train_runtime": 2556.17,
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.205,
|
7 |
"train_steps_per_second": 0.151
|
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":
|
121 |
-
"logits/rejected":
|
122 |
-
"logps/chosen": -30.
|
123 |
-
"logps/rejected": -
|
124 |
-
"loss": 0.
|
125 |
-
"rewards/accuracies": 0.
|
126 |
-
"rewards/chosen": 0.
|
127 |
-
"rewards/margins":
|
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": -
|
154 |
-
"loss": 0.
|
155 |
"rewards/accuracies": 0.5625,
|
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": -
|
168 |
-
"logps/rejected": -31.
|
169 |
-
"loss": 0.
|
170 |
-
"rewards/accuracies": 0.
|
171 |
-
"rewards/chosen": 0.
|
172 |
-
"rewards/margins": 0.
|
173 |
-
"rewards/rejected": 0.
|
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": -
|
182 |
-
"eval_loss": 0.
|
183 |
-
"eval_rewards/accuracies": 0.
|
184 |
-
"eval_rewards/chosen": 0.
|
185 |
-
"eval_rewards/margins":
|
186 |
-
"eval_rewards/rejected": 0.
|
187 |
-
"eval_runtime": 104.
|
188 |
-
"eval_samples_per_second": 3.
|
189 |
"eval_steps_per_second": 0.413,
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
-
"grad_norm":
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
-
"logits/chosen": 83.
|
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": 83.
|
212 |
-
"logits/rejected": 83.
|
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":
|
227 |
-
"logits/rejected":
|
228 |
-
"logps/chosen": -30.
|
229 |
-
"logps/rejected": -32.
|
230 |
"loss": 0.6608,
|
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":
|
242 |
-
"logits/rejected":
|
243 |
-
"logps/chosen": -26.
|
244 |
-
"logps/rejected": -
|
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": -
|
259 |
-
"logps/rejected": -33.
|
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":
|
272 |
-
"logits/rejected":
|
273 |
-
"logps/chosen": -
|
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":
|
287 |
-
"logits/rejected":
|
288 |
-
"logps/chosen": -30.
|
289 |
-
"logps/rejected": -
|
290 |
-
"loss": 0.
|
291 |
-
"rewards/accuracies": 0.
|
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":
|
302 |
-
"logits/rejected":
|
303 |
-
"logps/chosen": -30.
|
304 |
-
"logps/rejected": -
|
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":
|
317 |
-
"logits/rejected":
|
318 |
-
"logps/chosen": -30.
|
319 |
-
"logps/rejected": -
|
320 |
-
"loss": 0.
|
321 |
-
"rewards/accuracies": 0.
|
322 |
-
"rewards/chosen": 0.
|
323 |
-
"rewards/margins": 0.
|
324 |
-
"rewards/rejected": -0.
|
325 |
"step": 190
|
326 |
},
|
327 |
{
|
328 |
"epoch": 0.52,
|
329 |
-
"grad_norm": 5
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
-
"logits/chosen":
|
332 |
-
"logits/rejected":
|
333 |
-
"logps/chosen": -
|
334 |
-
"logps/rejected": -
|
335 |
-
"loss": 0.
|
336 |
-
"rewards/accuracies": 0.
|
337 |
-
"rewards/chosen": 0.
|
338 |
-
"rewards/margins": 0.
|
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": -36.
|
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": 103.
|
354 |
-
"eval_samples_per_second": 3.
|
355 |
"eval_steps_per_second": 0.414,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
-
"grad_norm":
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
-
"logits/chosen":
|
363 |
-
"logits/rejected":
|
364 |
-
"logps/chosen": -33.
|
365 |
-
"logps/rejected": -35.
|
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":
|
378 |
-
"logits/rejected":
|
379 |
-
"logps/chosen": -31.
|
380 |
-
"logps/rejected": -
|
381 |
-
"loss": 0.
|
382 |
-
"rewards/accuracies": 0.
|
383 |
-
"rewards/chosen": 0.
|
384 |
-
"rewards/margins": 0.
|
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":
|
393 |
-
"logits/rejected":
|
394 |
-
"logps/chosen": -32.
|
395 |
-
"logps/rejected": -34.
|
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":
|
408 |
-
"logits/rejected":
|
409 |
-
"logps/chosen": -
|
410 |
-
"logps/rejected": -
|
411 |
-
"loss": 0.
|
412 |
-
"rewards/accuracies": 0.
|
413 |
-
"rewards/chosen": 0.
|
414 |
-
"rewards/margins": 0.
|
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":
|
423 |
-
"logits/rejected":
|
424 |
-
"logps/chosen": -
|
425 |
-
"logps/rejected": -30.
|
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":
|
438 |
-
"logits/rejected":
|
439 |
-
"logps/chosen": -30.
|
440 |
-
"logps/rejected": -
|
441 |
-
"loss": 0.
|
442 |
"rewards/accuracies": 0.800000011920929,
|
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":
|
453 |
-
"logits/rejected":
|
454 |
-
"logps/chosen": -
|
455 |
-
"logps/rejected": -
|
456 |
-
"loss": 0.
|
457 |
-
"rewards/accuracies": 0.
|
458 |
-
"rewards/chosen": 0.
|
459 |
-
"rewards/margins": 0.
|
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":
|
468 |
-
"logits/rejected":
|
469 |
-
"logps/chosen": -31.
|
470 |
-
"logps/rejected": -
|
471 |
-
"loss": 0.
|
472 |
-
"rewards/accuracies": 0.
|
473 |
-
"rewards/chosen":
|
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":
|
483 |
-
"logits/rejected":
|
484 |
-
"logps/chosen": -33.
|
485 |
-
"logps/rejected": -
|
486 |
-
"loss": 0.
|
487 |
-
"rewards/accuracies": 0.
|
488 |
-
"rewards/chosen": 0.
|
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":
|
498 |
-
"logits/rejected":
|
499 |
-
"logps/chosen": -32.
|
500 |
-
"logps/rejected": -
|
501 |
-
"loss": 0.
|
502 |
-
"rewards/accuracies": 0.
|
503 |
-
"rewards/chosen": 0.
|
504 |
-
"rewards/margins":
|
505 |
-
"rewards/rejected": -0.
|
506 |
"step": 300
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.78,
|
510 |
-
"eval_logits/chosen":
|
511 |
-
"eval_logits/rejected":
|
512 |
-
"eval_logps/chosen": -
|
513 |
-
"eval_logps/rejected": -
|
514 |
-
"eval_loss": 0.
|
515 |
-
"eval_rewards/accuracies": 0.
|
516 |
-
"eval_rewards/chosen": -0.
|
517 |
-
"eval_rewards/margins": 0.
|
518 |
-
"eval_rewards/rejected": -0.
|
519 |
-
"eval_runtime":
|
520 |
-
"eval_samples_per_second": 3.
|
521 |
-
"eval_steps_per_second": 0.
|
522 |
"step": 300
|
523 |
},
|
524 |
{
|
525 |
"epoch": 0.81,
|
526 |
-
"grad_norm":
|
527 |
"learning_rate": 5.576113578589035e-07,
|
528 |
-
"logits/chosen":
|
529 |
-
"logits/rejected":
|
530 |
-
"logps/chosen": -30.
|
531 |
-
"logps/rejected": -
|
532 |
-
"loss": 0.
|
533 |
-
"rewards/accuracies": 0.
|
534 |
-
"rewards/chosen": 0.
|
535 |
-
"rewards/margins": 0.
|
536 |
-
"rewards/rejected": -0.
|
537 |
"step": 310
|
538 |
},
|
539 |
{
|
540 |
"epoch": 0.83,
|
541 |
-
"grad_norm":
|
542 |
"learning_rate": 4.229036944380913e-07,
|
543 |
-
"logits/chosen":
|
544 |
-
"logits/rejected":
|
545 |
-
"logps/chosen": -30.
|
546 |
-
"logps/rejected": -
|
547 |
-
"loss": 0.
|
548 |
-
"rewards/accuracies": 0.
|
549 |
-
"rewards/chosen": 0.
|
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":
|
559 |
-
"logits/rejected":
|
560 |
-
"logps/chosen": -29.
|
561 |
-
"logps/rejected": -
|
562 |
-
"loss": 0.
|
563 |
-
"rewards/accuracies": 0.
|
564 |
-
"rewards/chosen": 0.
|
565 |
-
"rewards/margins":
|
566 |
-
"rewards/rejected": -0.
|
567 |
"step": 330
|
568 |
},
|
569 |
{
|
570 |
"epoch": 0.88,
|
571 |
-
"grad_norm":
|
572 |
"learning_rate": 2.0579377374915805e-07,
|
573 |
-
"logits/chosen":
|
574 |
-
"logits/rejected":
|
575 |
-
"logps/chosen": -32.
|
576 |
-
"logps/rejected": -
|
577 |
-
"loss": 0.
|
578 |
-
"rewards/accuracies": 0.
|
579 |
-
"rewards/chosen": 0.
|
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":
|
589 |
-
"logits/rejected":
|
590 |
-
"logps/chosen": -32.
|
591 |
-
"logps/rejected": -
|
592 |
-
"loss": 0.
|
593 |
-
"rewards/accuracies": 0.
|
594 |
-
"rewards/chosen": 0.
|
595 |
-
"rewards/margins": 0.
|
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":
|
604 |
-
"logits/rejected":
|
605 |
-
"logps/chosen": -28.
|
606 |
-
"logps/rejected": -
|
607 |
-
"loss": 0.
|
608 |
-
"rewards/accuracies": 0.
|
609 |
-
"rewards/chosen": 0.
|
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":
|
619 |
-
"logits/rejected":
|
620 |
-
"logps/chosen": -
|
621 |
-
"logps/rejected": -
|
622 |
-
"loss": 0.
|
623 |
-
"rewards/accuracies": 0.
|
624 |
-
"rewards/chosen": 0.
|
625 |
-
"rewards/margins":
|
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":
|
634 |
-
"logits/rejected":
|
635 |
-
"logps/chosen": -
|
636 |
-
"logps/rejected": -
|
637 |
-
"loss": 0.
|
638 |
-
"rewards/accuracies": 0.
|
639 |
-
"rewards/chosen": 0.
|
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": 2556.
|
650 |
"train_samples_per_second": 1.205,
|
651 |
"train_steps_per_second": 0.151
|
652 |
}
|
|
|
10 |
"log_history": [
|
11 |
{
|
12 |
"epoch": 0.0,
|
13 |
+
"grad_norm": 52.25,
|
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": 40.75,
|
29 |
"learning_rate": 1.282051282051282e-06,
|
30 |
+
"logits/chosen": 81.06388092041016,
|
31 |
+
"logits/rejected": 80.76617431640625,
|
32 |
+
"logps/chosen": -34.315635681152344,
|
33 |
+
"logps/rejected": -33.083072662353516,
|
34 |
+
"loss": 0.7277,
|
35 |
+
"rewards/accuracies": 0.4305555522441864,
|
36 |
+
"rewards/chosen": -0.05912955477833748,
|
37 |
+
"rewards/margins": 0.04060414060950279,
|
38 |
+
"rewards/rejected": -0.09973368048667908,
|
39 |
"step": 10
|
40 |
},
|
41 |
{
|
42 |
"epoch": 0.05,
|
43 |
+
"grad_norm": 36.25,
|
44 |
"learning_rate": 2.564102564102564e-06,
|
45 |
+
"logits/chosen": 80.6710433959961,
|
46 |
+
"logits/rejected": 80.55816650390625,
|
47 |
+
"logps/chosen": -33.6015625,
|
48 |
+
"logps/rejected": -30.793548583984375,
|
49 |
+
"loss": 0.724,
|
50 |
+
"rewards/accuracies": 0.4749999940395355,
|
51 |
+
"rewards/chosen": 0.07067807018756866,
|
52 |
+
"rewards/margins": 0.08352533727884293,
|
53 |
+
"rewards/rejected": -0.012847280129790306,
|
54 |
"step": 20
|
55 |
},
|
56 |
{
|
57 |
"epoch": 0.08,
|
58 |
+
"grad_norm": 48.0,
|
59 |
"learning_rate": 3.846153846153847e-06,
|
60 |
+
"logits/chosen": 82.53105163574219,
|
61 |
+
"logits/rejected": 82.56202697753906,
|
62 |
+
"logps/chosen": -33.846534729003906,
|
63 |
+
"logps/rejected": -31.217113494873047,
|
64 |
+
"loss": 0.823,
|
65 |
+
"rewards/accuracies": 0.44999998807907104,
|
66 |
+
"rewards/chosen": 0.13303251564502716,
|
67 |
+
"rewards/margins": -0.03234311193227768,
|
68 |
+
"rewards/rejected": 0.16537563502788544,
|
69 |
"step": 30
|
70 |
},
|
71 |
{
|
72 |
"epoch": 0.1,
|
73 |
+
"grad_norm": 49.0,
|
74 |
"learning_rate": 4.999896948438434e-06,
|
75 |
+
"logits/chosen": 81.03451538085938,
|
76 |
+
"logits/rejected": 81.03263092041016,
|
77 |
+
"logps/chosen": -32.87987518310547,
|
78 |
+
"logps/rejected": -33.08030700683594,
|
79 |
+
"loss": 0.7898,
|
80 |
+
"rewards/accuracies": 0.5,
|
81 |
+
"rewards/chosen": 0.2353484332561493,
|
82 |
+
"rewards/margins": 0.004774211905896664,
|
83 |
+
"rewards/rejected": 0.23057425022125244,
|
84 |
"step": 40
|
85 |
},
|
86 |
{
|
87 |
"epoch": 0.13,
|
88 |
+
"grad_norm": 37.0,
|
89 |
"learning_rate": 4.987541037542187e-06,
|
90 |
+
"logits/chosen": 78.62713623046875,
|
91 |
+
"logits/rejected": 78.64068603515625,
|
92 |
+
"logps/chosen": -30.693645477294922,
|
93 |
+
"logps/rejected": -30.871618270874023,
|
94 |
+
"loss": 0.7362,
|
95 |
+
"rewards/accuracies": 0.550000011920929,
|
96 |
+
"rewards/chosen": 0.40928253531455994,
|
97 |
+
"rewards/margins": 0.24794897437095642,
|
98 |
+
"rewards/rejected": 0.1613335758447647,
|
99 |
"step": 50
|
100 |
},
|
101 |
{
|
102 |
"epoch": 0.16,
|
103 |
+
"grad_norm": 35.75,
|
104 |
"learning_rate": 4.954691471941119e-06,
|
105 |
+
"logits/chosen": 83.1414566040039,
|
106 |
+
"logits/rejected": 83.20195007324219,
|
107 |
+
"logps/chosen": -30.973648071289062,
|
108 |
+
"logps/rejected": -29.62088394165039,
|
109 |
+
"loss": 0.7214,
|
110 |
+
"rewards/accuracies": 0.5,
|
111 |
+
"rewards/chosen": 0.16120928525924683,
|
112 |
+
"rewards/margins": 0.18549844622612,
|
113 |
+
"rewards/rejected": -0.02428916096687317,
|
114 |
"step": 60
|
115 |
},
|
116 |
{
|
117 |
"epoch": 0.18,
|
118 |
+
"grad_norm": 72.5,
|
119 |
"learning_rate": 4.901618883413549e-06,
|
120 |
+
"logits/chosen": 83.78003692626953,
|
121 |
+
"logits/rejected": 83.8111343383789,
|
122 |
+
"logps/chosen": -30.537906646728516,
|
123 |
+
"logps/rejected": -33.126197814941406,
|
124 |
+
"loss": 0.7291,
|
125 |
+
"rewards/accuracies": 0.612500011920929,
|
126 |
+
"rewards/chosen": 0.07289580255746841,
|
127 |
+
"rewards/margins": 0.14368507266044617,
|
128 |
+
"rewards/rejected": -0.07078926265239716,
|
129 |
"step": 70
|
130 |
},
|
131 |
{
|
132 |
"epoch": 0.21,
|
133 |
+
"grad_norm": 46.25,
|
134 |
"learning_rate": 4.828760511501322e-06,
|
135 |
+
"logits/chosen": 81.44255065917969,
|
136 |
+
"logits/rejected": 81.430908203125,
|
137 |
+
"logps/chosen": -31.439010620117188,
|
138 |
+
"logps/rejected": -30.995372772216797,
|
139 |
+
"loss": 0.7039,
|
140 |
+
"rewards/accuracies": 0.48750001192092896,
|
141 |
+
"rewards/chosen": 0.0527455098927021,
|
142 |
+
"rewards/margins": 0.24292418360710144,
|
143 |
+
"rewards/rejected": -0.19017870724201202,
|
144 |
"step": 80
|
145 |
},
|
146 |
{
|
147 |
"epoch": 0.23,
|
148 |
+
"grad_norm": 61.75,
|
149 |
"learning_rate": 4.7367166013034295e-06,
|
150 |
+
"logits/chosen": 78.15830993652344,
|
151 |
+
"logits/rejected": 78.127197265625,
|
152 |
+
"logps/chosen": -32.702423095703125,
|
153 |
+
"logps/rejected": -31.252826690673828,
|
154 |
+
"loss": 0.7773,
|
155 |
"rewards/accuracies": 0.5625,
|
156 |
+
"rewards/chosen": -0.0513942614197731,
|
157 |
+
"rewards/margins": 0.18969133496284485,
|
158 |
+
"rewards/rejected": -0.24108560383319855,
|
159 |
"step": 90
|
160 |
},
|
161 |
{
|
162 |
"epoch": 0.26,
|
163 |
+
"grad_norm": 39.0,
|
164 |
"learning_rate": 4.626245458345211e-06,
|
165 |
+
"logits/chosen": 83.44832611083984,
|
166 |
+
"logits/rejected": 83.47621154785156,
|
167 |
+
"logps/chosen": -34.1397819519043,
|
168 |
+
"logps/rejected": -31.86690330505371,
|
169 |
+
"loss": 0.7017,
|
170 |
+
"rewards/accuracies": 0.574999988079071,
|
171 |
+
"rewards/chosen": 0.1321648806333542,
|
172 |
+
"rewards/margins": 0.225250244140625,
|
173 |
+
"rewards/rejected": -0.09308534860610962,
|
174 |
"step": 100
|
175 |
},
|
176 |
{
|
177 |
"epoch": 0.26,
|
178 |
+
"eval_logits/chosen": 98.7764892578125,
|
179 |
+
"eval_logits/rejected": 98.76370239257812,
|
180 |
+
"eval_logps/chosen": -32.49420166015625,
|
181 |
+
"eval_logps/rejected": -36.03456497192383,
|
182 |
+
"eval_loss": 0.8030149340629578,
|
183 |
+
"eval_rewards/accuracies": 0.5137043595314026,
|
184 |
+
"eval_rewards/chosen": -0.04081834852695465,
|
185 |
+
"eval_rewards/margins": 0.013612199574708939,
|
186 |
+
"eval_rewards/rejected": -0.05443055182695389,
|
187 |
+
"eval_runtime": 104.1164,
|
188 |
+
"eval_samples_per_second": 3.294,
|
189 |
"eval_steps_per_second": 0.413,
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
+
"grad_norm": 48.25,
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
+
"logits/chosen": 83.67167663574219,
|
197 |
+
"logits/rejected": 83.56147766113281,
|
198 |
+
"logps/chosen": -32.48063659667969,
|
199 |
+
"logps/rejected": -32.763511657714844,
|
200 |
+
"loss": 0.6258,
|
201 |
+
"rewards/accuracies": 0.6000000238418579,
|
202 |
+
"rewards/chosen": 0.43373650312423706,
|
203 |
+
"rewards/margins": 0.5198318362236023,
|
204 |
+
"rewards/rejected": -0.08609537780284882,
|
205 |
"step": 110
|
206 |
},
|
207 |
{
|
208 |
"epoch": 0.31,
|
209 |
+
"grad_norm": 53.75,
|
210 |
"learning_rate": 4.353806263777678e-06,
|
211 |
+
"logits/chosen": 83.80461120605469,
|
212 |
+
"logits/rejected": 83.91490173339844,
|
213 |
+
"logps/chosen": -28.273183822631836,
|
214 |
+
"logps/rejected": -35.56282043457031,
|
215 |
+
"loss": 0.5439,
|
216 |
+
"rewards/accuracies": 0.7250000238418579,
|
217 |
+
"rewards/chosen": 0.5251210927963257,
|
218 |
+
"rewards/margins": 0.608059287071228,
|
219 |
+
"rewards/rejected": -0.08293820172548294,
|
220 |
"step": 120
|
221 |
},
|
222 |
{
|
223 |
"epoch": 0.34,
|
224 |
+
"grad_norm": 28.75,
|
225 |
"learning_rate": 4.1940827077152755e-06,
|
226 |
+
"logits/chosen": 81.06788635253906,
|
227 |
+
"logits/rejected": 81.09136962890625,
|
228 |
+
"logps/chosen": -30.509033203125,
|
229 |
+
"logps/rejected": -32.07707595825195,
|
230 |
"loss": 0.6608,
|
231 |
+
"rewards/accuracies": 0.5874999761581421,
|
232 |
+
"rewards/chosen": 0.3186565339565277,
|
233 |
+
"rewards/margins": 0.4350167214870453,
|
234 |
+
"rewards/rejected": -0.11636020243167877,
|
235 |
"step": 130
|
236 |
},
|
237 |
{
|
238 |
"epoch": 0.36,
|
239 |
+
"grad_norm": 24.875,
|
240 |
"learning_rate": 4.0204024186666215e-06,
|
241 |
+
"logits/chosen": 82.33757019042969,
|
242 |
+
"logits/rejected": 82.34171295166016,
|
243 |
+
"logps/chosen": -26.971057891845703,
|
244 |
+
"logps/rejected": -32.84387969970703,
|
245 |
+
"loss": 0.5554,
|
246 |
+
"rewards/accuracies": 0.7250000238418579,
|
247 |
+
"rewards/chosen": 0.38101473450660706,
|
248 |
+
"rewards/margins": 0.7174161076545715,
|
249 |
+
"rewards/rejected": -0.33640143275260925,
|
250 |
"step": 140
|
251 |
},
|
252 |
{
|
253 |
"epoch": 0.39,
|
254 |
+
"grad_norm": 30.125,
|
255 |
"learning_rate": 3.834196265035119e-06,
|
256 |
+
"logits/chosen": 80.85295104980469,
|
257 |
+
"logits/rejected": 80.82408142089844,
|
258 |
+
"logps/chosen": -28.91988754272461,
|
259 |
+
"logps/rejected": -33.04113006591797,
|
260 |
+
"loss": 0.5578,
|
261 |
+
"rewards/accuracies": 0.699999988079071,
|
262 |
+
"rewards/chosen": 0.375379741191864,
|
263 |
+
"rewards/margins": 0.7550806403160095,
|
264 |
+
"rewards/rejected": -0.37970098853111267,
|
265 |
"step": 150
|
266 |
},
|
267 |
{
|
268 |
"epoch": 0.42,
|
269 |
+
"grad_norm": 60.5,
|
270 |
"learning_rate": 3.636998309800573e-06,
|
271 |
+
"logits/chosen": 82.69292449951172,
|
272 |
+
"logits/rejected": 82.70225524902344,
|
273 |
+
"logps/chosen": -33.54145050048828,
|
274 |
+
"logps/rejected": -30.415197372436523,
|
275 |
+
"loss": 0.6193,
|
276 |
+
"rewards/accuracies": 0.699999988079071,
|
277 |
+
"rewards/chosen": 0.4762394428253174,
|
278 |
+
"rewards/margins": 0.8463689684867859,
|
279 |
+
"rewards/rejected": -0.37012940645217896,
|
280 |
"step": 160
|
281 |
},
|
282 |
{
|
283 |
"epoch": 0.44,
|
284 |
+
"grad_norm": 30.375,
|
285 |
"learning_rate": 3.4304331721118078e-06,
|
286 |
+
"logits/chosen": 83.42604064941406,
|
287 |
+
"logits/rejected": 83.37831115722656,
|
288 |
+
"logps/chosen": -30.8799991607666,
|
289 |
+
"logps/rejected": -32.62963104248047,
|
290 |
+
"loss": 0.5673,
|
291 |
+
"rewards/accuracies": 0.6625000238418579,
|
292 |
+
"rewards/chosen": 0.3034316599369049,
|
293 |
+
"rewards/margins": 0.7910871505737305,
|
294 |
+
"rewards/rejected": -0.4876554012298584,
|
295 |
"step": 170
|
296 |
},
|
297 |
{
|
298 |
"epoch": 0.47,
|
299 |
+
"grad_norm": 32.0,
|
300 |
"learning_rate": 3.2162026428305436e-06,
|
301 |
+
"logits/chosen": 80.95600891113281,
|
302 |
+
"logits/rejected": 80.93622589111328,
|
303 |
+
"logps/chosen": -30.429906845092773,
|
304 |
+
"logps/rejected": -31.591838836669922,
|
305 |
+
"loss": 0.4992,
|
306 |
+
"rewards/accuracies": 0.737500011920929,
|
307 |
+
"rewards/chosen": 0.6137940287590027,
|
308 |
+
"rewards/margins": 0.9614995718002319,
|
309 |
+
"rewards/rejected": -0.34770551323890686,
|
310 |
"step": 180
|
311 |
},
|
312 |
{
|
313 |
"epoch": 0.49,
|
314 |
+
"grad_norm": 17.25,
|
315 |
"learning_rate": 2.996071664294641e-06,
|
316 |
+
"logits/chosen": 82.7037124633789,
|
317 |
+
"logits/rejected": 82.69169616699219,
|
318 |
+
"logps/chosen": -30.465667724609375,
|
319 |
+
"logps/rejected": -30.747249603271484,
|
320 |
+
"loss": 0.7493,
|
321 |
+
"rewards/accuracies": 0.574999988079071,
|
322 |
+
"rewards/chosen": 0.24060054123401642,
|
323 |
+
"rewards/margins": 0.463981568813324,
|
324 |
+
"rewards/rejected": -0.22338099777698517,
|
325 |
"step": 190
|
326 |
},
|
327 |
{
|
328 |
"epoch": 0.52,
|
329 |
+
"grad_norm": 21.5,
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
+
"logits/chosen": 78.2643814086914,
|
332 |
+
"logits/rejected": 78.21286010742188,
|
333 |
+
"logps/chosen": -33.79939270019531,
|
334 |
+
"logps/rejected": -32.67485046386719,
|
335 |
+
"loss": 0.6236,
|
336 |
+
"rewards/accuracies": 0.675000011920929,
|
337 |
+
"rewards/chosen": 0.7626792192459106,
|
338 |
+
"rewards/margins": 0.9078337550163269,
|
339 |
+
"rewards/rejected": -0.14515459537506104,
|
340 |
"step": 200
|
341 |
},
|
342 |
{
|
343 |
"epoch": 0.52,
|
344 |
+
"eval_logits/chosen": 98.72710418701172,
|
345 |
+
"eval_logits/rejected": 98.70391082763672,
|
346 |
+
"eval_logps/chosen": -32.6581916809082,
|
347 |
+
"eval_logps/rejected": -36.33292770385742,
|
348 |
+
"eval_loss": 0.7719493508338928,
|
349 |
+
"eval_rewards/accuracies": 0.5215947031974792,
|
350 |
+
"eval_rewards/chosen": -0.17201441526412964,
|
351 |
+
"eval_rewards/margins": 0.12110573798418045,
|
352 |
+
"eval_rewards/rejected": -0.2931201457977295,
|
353 |
+
"eval_runtime": 103.8817,
|
354 |
+
"eval_samples_per_second": 3.302,
|
355 |
"eval_steps_per_second": 0.414,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
+
"grad_norm": 78.5,
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
+
"logits/chosen": 80.8266830444336,
|
363 |
+
"logits/rejected": 80.73197174072266,
|
364 |
+
"logps/chosen": -33.31422805786133,
|
365 |
+
"logps/rejected": -35.28189468383789,
|
366 |
+
"loss": 0.6059,
|
367 |
+
"rewards/accuracies": 0.6625000238418579,
|
368 |
+
"rewards/chosen": 0.5051938891410828,
|
369 |
+
"rewards/margins": 0.7418341636657715,
|
370 |
+
"rewards/rejected": -0.23664025962352753,
|
371 |
"step": 210
|
372 |
},
|
373 |
{
|
374 |
"epoch": 0.57,
|
375 |
+
"grad_norm": 21.0,
|
376 |
"learning_rate": 2.3185646976551794e-06,
|
377 |
+
"logits/chosen": 83.00508117675781,
|
378 |
+
"logits/rejected": 83.09242248535156,
|
379 |
+
"logps/chosen": -31.03920555114746,
|
380 |
+
"logps/rejected": -30.981882095336914,
|
381 |
+
"loss": 0.5212,
|
382 |
+
"rewards/accuracies": 0.800000011920929,
|
383 |
+
"rewards/chosen": 0.7277995347976685,
|
384 |
+
"rewards/margins": 0.930141270160675,
|
385 |
+
"rewards/rejected": -0.20234176516532898,
|
386 |
"step": 220
|
387 |
},
|
388 |
{
|
389 |
"epoch": 0.6,
|
390 |
+
"grad_norm": 49.25,
|
391 |
"learning_rate": 2.0932279108998323e-06,
|
392 |
+
"logits/chosen": 80.0581283569336,
|
393 |
+
"logits/rejected": 80.11595153808594,
|
394 |
+
"logps/chosen": -32.08351516723633,
|
395 |
+
"logps/rejected": -34.33815002441406,
|
396 |
+
"loss": 0.594,
|
397 |
+
"rewards/accuracies": 0.637499988079071,
|
398 |
+
"rewards/chosen": 0.5777846574783325,
|
399 |
+
"rewards/margins": 0.8385850787162781,
|
400 |
+
"rewards/rejected": -0.26080039143562317,
|
401 |
"step": 230
|
402 |
},
|
403 |
{
|
404 |
"epoch": 0.62,
|
405 |
+
"grad_norm": 28.0,
|
406 |
"learning_rate": 1.8712423238279358e-06,
|
407 |
+
"logits/chosen": 82.56427001953125,
|
408 |
+
"logits/rejected": 82.85057830810547,
|
409 |
+
"logps/chosen": -30.730464935302734,
|
410 |
+
"logps/rejected": -31.813983917236328,
|
411 |
+
"loss": 0.4701,
|
412 |
+
"rewards/accuracies": 0.75,
|
413 |
+
"rewards/chosen": 0.7660826444625854,
|
414 |
+
"rewards/margins": 0.9916342496871948,
|
415 |
+
"rewards/rejected": -0.22555160522460938,
|
416 |
"step": 240
|
417 |
},
|
418 |
{
|
419 |
"epoch": 0.65,
|
420 |
+
"grad_norm": 50.5,
|
421 |
"learning_rate": 1.6544367689701824e-06,
|
422 |
+
"logits/chosen": 81.2629623413086,
|
423 |
+
"logits/rejected": 81.32804870605469,
|
424 |
+
"logps/chosen": -26.856103897094727,
|
425 |
+
"logps/rejected": -30.107614517211914,
|
426 |
+
"loss": 0.6072,
|
427 |
+
"rewards/accuracies": 0.6499999761581421,
|
428 |
+
"rewards/chosen": 0.5918346047401428,
|
429 |
+
"rewards/margins": 0.7573351263999939,
|
430 |
+
"rewards/rejected": -0.1655004471540451,
|
431 |
"step": 250
|
432 |
},
|
433 |
{
|
434 |
"epoch": 0.68,
|
435 |
+
"grad_norm": 36.25,
|
436 |
"learning_rate": 1.4445974030621963e-06,
|
437 |
+
"logits/chosen": 78.54646301269531,
|
438 |
+
"logits/rejected": 78.68824768066406,
|
439 |
+
"logps/chosen": -30.280277252197266,
|
440 |
+
"logps/rejected": -36.395103454589844,
|
441 |
+
"loss": 0.4323,
|
442 |
"rewards/accuracies": 0.800000011920929,
|
443 |
+
"rewards/chosen": 1.0245726108551025,
|
444 |
+
"rewards/margins": 1.259353756904602,
|
445 |
+
"rewards/rejected": -0.23478105664253235,
|
446 |
"step": 260
|
447 |
},
|
448 |
{
|
449 |
"epoch": 0.7,
|
450 |
+
"grad_norm": 22.625,
|
451 |
"learning_rate": 1.243452991757889e-06,
|
452 |
+
"logits/chosen": 77.9116439819336,
|
453 |
+
"logits/rejected": 77.93907165527344,
|
454 |
+
"logps/chosen": -30.94207763671875,
|
455 |
+
"logps/rejected": -31.825298309326172,
|
456 |
+
"loss": 0.5298,
|
457 |
+
"rewards/accuracies": 0.7124999761581421,
|
458 |
+
"rewards/chosen": 0.6694773435592651,
|
459 |
+
"rewards/margins": 0.9253811836242676,
|
460 |
+
"rewards/rejected": -0.2559038996696472,
|
461 |
"step": 270
|
462 |
},
|
463 |
{
|
464 |
"epoch": 0.73,
|
465 |
+
"grad_norm": 48.0,
|
466 |
"learning_rate": 1.0526606671603523e-06,
|
467 |
+
"logits/chosen": 80.6185531616211,
|
468 |
+
"logits/rejected": 80.40409851074219,
|
469 |
+
"logps/chosen": -31.1258487701416,
|
470 |
+
"logps/rejected": -29.83230972290039,
|
471 |
+
"loss": 0.6589,
|
472 |
+
"rewards/accuracies": 0.6625000238418579,
|
473 |
+
"rewards/chosen": 0.5472350716590881,
|
474 |
+
"rewards/margins": 0.6794873476028442,
|
475 |
+
"rewards/rejected": -0.1322522908449173,
|
476 |
"step": 280
|
477 |
},
|
478 |
{
|
479 |
"epoch": 0.75,
|
480 |
+
"grad_norm": 17.25,
|
481 |
"learning_rate": 8.737922755071455e-07,
|
482 |
+
"logits/chosen": 80.76725006103516,
|
483 |
+
"logits/rejected": 80.67757415771484,
|
484 |
+
"logps/chosen": -33.04716491699219,
|
485 |
+
"logps/rejected": -32.507728576660156,
|
486 |
+
"loss": 0.4727,
|
487 |
+
"rewards/accuracies": 0.7250000238418579,
|
488 |
+
"rewards/chosen": 0.8456665277481079,
|
489 |
+
"rewards/margins": 1.238355040550232,
|
490 |
+
"rewards/rejected": -0.39268869161605835,
|
491 |
"step": 290
|
492 |
},
|
493 |
{
|
494 |
"epoch": 0.78,
|
495 |
+
"grad_norm": 44.75,
|
496 |
"learning_rate": 7.08321427484816e-07,
|
497 |
+
"logits/chosen": 76.41008758544922,
|
498 |
+
"logits/rejected": 76.48558044433594,
|
499 |
+
"logps/chosen": -32.2567253112793,
|
500 |
+
"logps/rejected": -29.211252212524414,
|
501 |
+
"loss": 0.5655,
|
502 |
+
"rewards/accuracies": 0.75,
|
503 |
+
"rewards/chosen": 0.9191843271255493,
|
504 |
+
"rewards/margins": 1.0065667629241943,
|
505 |
+
"rewards/rejected": -0.08738242089748383,
|
506 |
"step": 300
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.78,
|
510 |
+
"eval_logits/chosen": 98.77224731445312,
|
511 |
+
"eval_logits/rejected": 98.75171661376953,
|
512 |
+
"eval_logps/chosen": -32.550052642822266,
|
513 |
+
"eval_logps/rejected": -36.224205017089844,
|
514 |
+
"eval_loss": 0.7744476795196533,
|
515 |
+
"eval_rewards/accuracies": 0.5427741408348083,
|
516 |
+
"eval_rewards/chosen": -0.08550228923559189,
|
517 |
+
"eval_rewards/margins": 0.1206393763422966,
|
518 |
+
"eval_rewards/rejected": -0.2061416506767273,
|
519 |
+
"eval_runtime": 104.0332,
|
520 |
+
"eval_samples_per_second": 3.297,
|
521 |
+
"eval_steps_per_second": 0.413,
|
522 |
"step": 300
|
523 |
},
|
524 |
{
|
525 |
"epoch": 0.81,
|
526 |
+
"grad_norm": 31.875,
|
527 |
"learning_rate": 5.576113578589035e-07,
|
528 |
+
"logits/chosen": 83.43260192871094,
|
529 |
+
"logits/rejected": 83.46700286865234,
|
530 |
+
"logps/chosen": -30.02802085876465,
|
531 |
+
"logps/rejected": -32.44657897949219,
|
532 |
+
"loss": 0.5905,
|
533 |
+
"rewards/accuracies": 0.699999988079071,
|
534 |
+
"rewards/chosen": 0.6684112548828125,
|
535 |
+
"rewards/margins": 0.8658668398857117,
|
536 |
+
"rewards/rejected": -0.19745555520057678,
|
537 |
"step": 310
|
538 |
},
|
539 |
{
|
540 |
"epoch": 0.83,
|
541 |
+
"grad_norm": 24.875,
|
542 |
"learning_rate": 4.229036944380913e-07,
|
543 |
+
"logits/chosen": 80.97013092041016,
|
544 |
+
"logits/rejected": 80.96881103515625,
|
545 |
+
"logps/chosen": -30.49709701538086,
|
546 |
+
"logps/rejected": -29.160675048828125,
|
547 |
+
"loss": 0.4662,
|
548 |
+
"rewards/accuracies": 0.7250000238418579,
|
549 |
+
"rewards/chosen": 0.8811699748039246,
|
550 |
+
"rewards/margins": 1.0741162300109863,
|
551 |
+
"rewards/rejected": -0.19294631481170654,
|
552 |
"step": 320
|
553 |
},
|
554 |
{
|
555 |
"epoch": 0.86,
|
556 |
+
"grad_norm": 19.625,
|
557 |
"learning_rate": 3.053082288996112e-07,
|
558 |
+
"logits/chosen": 78.14985656738281,
|
559 |
+
"logits/rejected": 78.19586181640625,
|
560 |
+
"logps/chosen": -29.189855575561523,
|
561 |
+
"logps/rejected": -32.966094970703125,
|
562 |
+
"loss": 0.4527,
|
563 |
+
"rewards/accuracies": 0.7749999761581421,
|
564 |
+
"rewards/chosen": 0.9302013516426086,
|
565 |
+
"rewards/margins": 1.195185661315918,
|
566 |
+
"rewards/rejected": -0.26498422026634216,
|
567 |
"step": 330
|
568 |
},
|
569 |
{
|
570 |
"epoch": 0.88,
|
571 |
+
"grad_norm": 61.5,
|
572 |
"learning_rate": 2.0579377374915805e-07,
|
573 |
+
"logits/chosen": 82.46419525146484,
|
574 |
+
"logits/rejected": 82.48377990722656,
|
575 |
+
"logps/chosen": -32.1870002746582,
|
576 |
+
"logps/rejected": -33.69792175292969,
|
577 |
+
"loss": 0.5537,
|
578 |
+
"rewards/accuracies": 0.737500011920929,
|
579 |
+
"rewards/chosen": 0.8201173543930054,
|
580 |
+
"rewards/margins": 1.0834996700286865,
|
581 |
+
"rewards/rejected": -0.26338234543800354,
|
582 |
"step": 340
|
583 |
},
|
584 |
{
|
585 |
"epoch": 0.91,
|
586 |
+
"grad_norm": 16.75,
|
587 |
"learning_rate": 1.2518018074041684e-07,
|
588 |
+
"logits/chosen": 81.44327545166016,
|
589 |
+
"logits/rejected": 81.4502944946289,
|
590 |
+
"logps/chosen": -32.606056213378906,
|
591 |
+
"logps/rejected": -33.298622131347656,
|
592 |
+
"loss": 0.5676,
|
593 |
+
"rewards/accuracies": 0.6875,
|
594 |
+
"rewards/chosen": 0.8301448822021484,
|
595 |
+
"rewards/margins": 0.9956042170524597,
|
596 |
+
"rewards/rejected": -0.16545933485031128,
|
597 |
"step": 350
|
598 |
},
|
599 |
{
|
600 |
"epoch": 0.94,
|
601 |
+
"grad_norm": 33.0,
|
602 |
"learning_rate": 6.41315865106129e-08,
|
603 |
+
"logits/chosen": 82.95188903808594,
|
604 |
+
"logits/rejected": 82.9841079711914,
|
605 |
+
"logps/chosen": -28.371601104736328,
|
606 |
+
"logps/rejected": -31.830394744873047,
|
607 |
+
"loss": 0.4872,
|
608 |
+
"rewards/accuracies": 0.7124999761581421,
|
609 |
+
"rewards/chosen": 0.9444905519485474,
|
610 |
+
"rewards/margins": 1.0889289379119873,
|
611 |
+
"rewards/rejected": -0.14443838596343994,
|
612 |
"step": 360
|
613 |
},
|
614 |
{
|
615 |
"epoch": 0.96,
|
616 |
+
"grad_norm": 34.25,
|
617 |
"learning_rate": 2.3150941078050325e-08,
|
618 |
+
"logits/chosen": 82.37725067138672,
|
619 |
+
"logits/rejected": 82.40049743652344,
|
620 |
+
"logps/chosen": -31.749374389648438,
|
621 |
+
"logps/rejected": -35.2952766418457,
|
622 |
+
"loss": 0.508,
|
623 |
+
"rewards/accuracies": 0.7124999761581421,
|
624 |
+
"rewards/chosen": 0.9014309644699097,
|
625 |
+
"rewards/margins": 1.0501554012298584,
|
626 |
+
"rewards/rejected": -0.14872442185878754,
|
627 |
"step": 370
|
628 |
},
|
629 |
{
|
630 |
"epoch": 0.99,
|
631 |
+
"grad_norm": 38.0,
|
632 |
"learning_rate": 2.575864278703266e-09,
|
633 |
+
"logits/chosen": 76.2959976196289,
|
634 |
+
"logits/rejected": 76.17599487304688,
|
635 |
+
"logps/chosen": -29.714282989501953,
|
636 |
+
"logps/rejected": -28.307018280029297,
|
637 |
+
"loss": 0.57,
|
638 |
+
"rewards/accuracies": 0.6625000238418579,
|
639 |
+
"rewards/chosen": 0.6894746422767639,
|
640 |
+
"rewards/margins": 0.806629478931427,
|
641 |
+
"rewards/rejected": -0.11715485900640488,
|
642 |
"step": 380
|
643 |
},
|
644 |
{
|
645 |
"epoch": 1.0,
|
646 |
"step": 385,
|
647 |
"total_flos": 0.0,
|
648 |
+
"train_loss": 0.6050039254225694,
|
649 |
+
"train_runtime": 2556.17,
|
650 |
"train_samples_per_second": 1.205,
|
651 |
"train_steps_per_second": 0.151
|
652 |
}
|