hugodk-sch
commited on
Commit
•
1aff0c5
1
Parent(s):
1ef5ee4
Model save
Browse files- README.md +14 -16
- adapter_model.safetensors +1 -1
- all_results.json +4 -17
- train_results.json +4 -4
- trainer_state.json +418 -418
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: -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: 0.7290
|
22 |
+
- Rewards/chosen: -0.1187
|
23 |
+
- Rewards/rejected: -0.1591
|
24 |
+
- Rewards/accuracies: 0.5453
|
25 |
+
- Rewards/margins: 0.0404
|
26 |
+
- Logps/rejected: -36.2848
|
27 |
+
- Logps/chosen: -32.6805
|
28 |
+
- Logits/rejected: 98.5351
|
29 |
+
- Logits/chosen: 98.5572
|
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 |
+
| 0.6859 | 0.26 | 100 | 0.7229 | -0.0559 | -0.0850 | 0.5282 | 0.0290 | -36.1364 | -32.5550 | 98.6727 | 98.6826 |
|
65 |
+
| 0.5453 | 0.52 | 200 | 0.7068 | -0.1291 | -0.2159 | 0.5332 | 0.0868 | -36.3984 | -32.7014 | 98.4732 | 98.4956 |
|
66 |
+
| 0.5511 | 0.78 | 300 | 0.7290 | -0.1187 | -0.1591 | 0.5453 | 0.0404 | -36.2848 | -32.6805 | 98.5351 | 98.5572 |
|
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 sha256:5c0ea97376da5afb222e6a83a4557406c60f92c0d7ecff04ae353fe63ef5dae6
|
3 |
size 671150064
|
all_results.json
CHANGED
@@ -1,21 +1,8 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"
|
4 |
-
"
|
5 |
-
"eval_logps/chosen": -32.56877517700195,
|
6 |
-
"eval_logps/rejected": -36.22236251831055,
|
7 |
-
"eval_loss": 0.7821906208992004,
|
8 |
-
"eval_rewards/accuracies": 0.52491694688797,
|
9 |
-
"eval_rewards/chosen": -0.10047715157270432,
|
10 |
-
"eval_rewards/margins": 0.1041894406080246,
|
11 |
-
"eval_rewards/rejected": -0.20466658473014832,
|
12 |
-
"eval_runtime": 103.6205,
|
13 |
-
"eval_samples": 343,
|
14 |
-
"eval_samples_per_second": 3.31,
|
15 |
-
"eval_steps_per_second": 0.415,
|
16 |
-
"train_loss": 0.6050039254225694,
|
17 |
-
"train_runtime": 2556.17,
|
18 |
"train_samples": 3079,
|
19 |
-
"train_samples_per_second": 1.
|
20 |
-
"train_steps_per_second": 0.
|
21 |
}
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"train_loss": 0.5860598966672823,
|
4 |
+
"train_runtime": 2558.949,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
"train_samples": 3079,
|
6 |
+
"train_samples_per_second": 1.203,
|
7 |
+
"train_steps_per_second": 0.15
|
8 |
}
|
train_results.json
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"train_loss": 0.
|
4 |
-
"train_runtime":
|
5 |
"train_samples": 3079,
|
6 |
-
"train_samples_per_second": 1.
|
7 |
-
"train_steps_per_second": 0.
|
8 |
}
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"train_loss": 0.5860598966672823,
|
4 |
+
"train_runtime": 2558.949,
|
5 |
"train_samples": 3079,
|
6 |
+
"train_samples_per_second": 1.203,
|
7 |
+
"train_steps_per_second": 0.15
|
8 |
}
|
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,
|
@@ -25,630 +25,630 @@
|
|
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":
|
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":
|
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":
|
114 |
"step": 60
|
115 |
},
|
116 |
{
|
117 |
"epoch": 0.18,
|
118 |
-
"grad_norm":
|
119 |
"learning_rate": 4.901618883413549e-06,
|
120 |
-
"logits/chosen": 83.
|
121 |
-
"logits/rejected": 83.
|
122 |
-
"logps/chosen": -30.
|
123 |
-
"logps/rejected": -33.
|
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": -
|
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": -31.
|
154 |
-
"loss": 0.
|
155 |
-
"rewards/accuracies": 0.
|
156 |
-
"rewards/chosen":
|
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": -34.
|
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": -36.
|
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": 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.
|
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": -
|
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": -28.
|
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": 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.
|
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": -30.
|
319 |
-
"logps/rejected": -30.
|
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":
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
-
"logits/chosen":
|
332 |
-
"logits/rejected":
|
333 |
-
"logps/chosen": -33.
|
334 |
-
"logps/rejected": -32.
|
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": 80.
|
363 |
-
"logits/rejected": 80.
|
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": 0.637499988079071,
|
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": 82.
|
409 |
-
"logps/chosen": -30.
|
410 |
-
"logps/rejected": -31.
|
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": -26.
|
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": 78.
|
439 |
-
"logps/chosen": -30.
|
440 |
-
"logps/rejected": -36.
|
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": 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": 80.
|
468 |
-
"logits/rejected":
|
469 |
-
"logps/chosen": -31.
|
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": 80.
|
484 |
-
"logps/chosen": -33.
|
485 |
-
"logps/rejected": -32.
|
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": -29.
|
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": 98.
|
511 |
-
"eval_logits/rejected": 98.
|
512 |
-
"eval_logps/chosen": -32.
|
513 |
-
"eval_logps/rejected": -36.
|
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": 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":
|
529 |
-
"logits/rejected":
|
530 |
-
"logps/chosen": -30.
|
531 |
-
"logps/rejected": -32.
|
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": 80.
|
544 |
-
"logits/rejected": 80.
|
545 |
-
"logps/chosen": -30.
|
546 |
-
"logps/rejected": -29.
|
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": -33.
|
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": -33.
|
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": 82.
|
604 |
-
"logits/rejected": 82.
|
605 |
-
"logps/chosen": -28.
|
606 |
-
"logps/rejected": -31.
|
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": -31.
|
621 |
-
"logps/rejected": -35.
|
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": -29.
|
636 |
-
"logps/rejected": -28.
|
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":
|
650 |
-
"train_samples_per_second": 1.
|
651 |
-
"train_steps_per_second": 0.
|
652 |
}
|
653 |
],
|
654 |
"logging_steps": 10,
|
|
|
10 |
"log_history": [
|
11 |
{
|
12 |
"epoch": 0.0,
|
13 |
+
"grad_norm": 32.75,
|
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": 25.5,
|
29 |
"learning_rate": 1.282051282051282e-06,
|
30 |
+
"logits/chosen": 81.08353424072266,
|
31 |
+
"logits/rejected": 80.7865219116211,
|
32 |
+
"logps/chosen": -34.20063781738281,
|
33 |
+
"logps/rejected": -33.02577590942383,
|
34 |
+
"loss": 0.684,
|
35 |
+
"rewards/accuracies": 0.5,
|
36 |
+
"rewards/chosen": 0.020544879138469696,
|
37 |
+
"rewards/margins": 0.05422975495457649,
|
38 |
+
"rewards/rejected": -0.033684875816106796,
|
39 |
"step": 10
|
40 |
},
|
41 |
{
|
42 |
"epoch": 0.05,
|
43 |
+
"grad_norm": 25.0,
|
44 |
"learning_rate": 2.564102564102564e-06,
|
45 |
+
"logits/chosen": 80.67228698730469,
|
46 |
+
"logits/rejected": 80.55956268310547,
|
47 |
+
"logps/chosen": -33.523399353027344,
|
48 |
+
"logps/rejected": -30.766698837280273,
|
49 |
+
"loss": 0.6867,
|
50 |
+
"rewards/accuracies": 0.512499988079071,
|
51 |
+
"rewards/chosen": 0.0832558423280716,
|
52 |
+
"rewards/margins": 0.0778605192899704,
|
53 |
+
"rewards/rejected": 0.00539533281698823,
|
54 |
"step": 20
|
55 |
},
|
56 |
{
|
57 |
"epoch": 0.08,
|
58 |
+
"grad_norm": 29.375,
|
59 |
"learning_rate": 3.846153846153847e-06,
|
60 |
+
"logits/chosen": 82.52102661132812,
|
61 |
+
"logits/rejected": 82.54798889160156,
|
62 |
+
"logps/chosen": -33.94279861450195,
|
63 |
+
"logps/rejected": -31.214736938476562,
|
64 |
+
"loss": 0.7575,
|
65 |
+
"rewards/accuracies": 0.4375,
|
66 |
+
"rewards/chosen": 0.035010773688554764,
|
67 |
+
"rewards/margins": -0.06953731179237366,
|
68 |
+
"rewards/rejected": 0.10454808175563812,
|
69 |
"step": 30
|
70 |
},
|
71 |
{
|
72 |
"epoch": 0.1,
|
73 |
+
"grad_norm": 24.25,
|
74 |
"learning_rate": 4.999896948438434e-06,
|
75 |
+
"logits/chosen": 81.06597137451172,
|
76 |
+
"logits/rejected": 81.06147766113281,
|
77 |
+
"logps/chosen": -32.76929473876953,
|
78 |
+
"logps/rejected": -33.212310791015625,
|
79 |
+
"loss": 0.6836,
|
80 |
+
"rewards/accuracies": 0.6000000238418579,
|
81 |
+
"rewards/chosen": 0.20238244533538818,
|
82 |
+
"rewards/margins": 0.12427721172571182,
|
83 |
+
"rewards/rejected": 0.07810524851083755,
|
84 |
"step": 40
|
85 |
},
|
86 |
{
|
87 |
"epoch": 0.13,
|
88 |
+
"grad_norm": 18.75,
|
89 |
"learning_rate": 4.987541037542187e-06,
|
90 |
+
"logits/chosen": 78.68827819824219,
|
91 |
+
"logits/rejected": 78.70021057128906,
|
92 |
+
"logps/chosen": -30.579212188720703,
|
93 |
+
"logps/rejected": -30.904521942138672,
|
94 |
+
"loss": 0.6716,
|
95 |
+
"rewards/accuracies": 0.637499988079071,
|
96 |
+
"rewards/chosen": 0.3130193054676056,
|
97 |
+
"rewards/margins": 0.22863590717315674,
|
98 |
+
"rewards/rejected": 0.08438339829444885,
|
99 |
"step": 50
|
100 |
},
|
101 |
{
|
102 |
"epoch": 0.16,
|
103 |
+
"grad_norm": 25.375,
|
104 |
"learning_rate": 4.954691471941119e-06,
|
105 |
+
"logits/chosen": 83.18531799316406,
|
106 |
+
"logits/rejected": 83.2416763305664,
|
107 |
+
"logps/chosen": -30.874561309814453,
|
108 |
+
"logps/rejected": -29.523483276367188,
|
109 |
+
"loss": 0.6904,
|
110 |
+
"rewards/accuracies": 0.5625,
|
111 |
+
"rewards/chosen": 0.15030017495155334,
|
112 |
+
"rewards/margins": 0.11678256839513779,
|
113 |
+
"rewards/rejected": 0.03351757675409317,
|
114 |
"step": 60
|
115 |
},
|
116 |
{
|
117 |
"epoch": 0.18,
|
118 |
+
"grad_norm": 40.75,
|
119 |
"learning_rate": 4.901618883413549e-06,
|
120 |
+
"logits/chosen": 83.8118896484375,
|
121 |
+
"logits/rejected": 83.84835815429688,
|
122 |
+
"logps/chosen": -30.521596908569336,
|
123 |
+
"logps/rejected": -33.058387756347656,
|
124 |
+
"loss": 0.7083,
|
125 |
+
"rewards/accuracies": 0.512499988079071,
|
126 |
+
"rewards/chosen": 0.05371224880218506,
|
127 |
+
"rewards/margins": 0.06404958665370941,
|
128 |
+
"rewards/rejected": -0.010337340645492077,
|
129 |
"step": 70
|
130 |
},
|
131 |
{
|
132 |
"epoch": 0.21,
|
133 |
+
"grad_norm": 25.625,
|
134 |
"learning_rate": 4.828760511501322e-06,
|
135 |
+
"logits/chosen": 81.4435043334961,
|
136 |
+
"logits/rejected": 81.42585754394531,
|
137 |
+
"logps/chosen": -31.408594131469727,
|
138 |
+
"logps/rejected": -31.023513793945312,
|
139 |
+
"loss": 0.6594,
|
140 |
+
"rewards/accuracies": 0.6000000238418579,
|
141 |
+
"rewards/chosen": 0.04817419499158859,
|
142 |
+
"rewards/margins": 0.18110866844654083,
|
143 |
+
"rewards/rejected": -0.13293446600437164,
|
144 |
"step": 80
|
145 |
},
|
146 |
{
|
147 |
"epoch": 0.23,
|
148 |
+
"grad_norm": 31.75,
|
149 |
"learning_rate": 4.7367166013034295e-06,
|
150 |
+
"logits/chosen": 78.11305236816406,
|
151 |
+
"logits/rejected": 78.08197784423828,
|
152 |
+
"logps/chosen": -32.596641540527344,
|
153 |
+
"logps/rejected": -31.254547119140625,
|
154 |
+
"loss": 0.677,
|
155 |
+
"rewards/accuracies": 0.6000000238418579,
|
156 |
+
"rewards/chosen": 0.02076822891831398,
|
157 |
+
"rewards/margins": 0.17230577766895294,
|
158 |
+
"rewards/rejected": -0.15153756737709045,
|
159 |
"step": 90
|
160 |
},
|
161 |
{
|
162 |
"epoch": 0.26,
|
163 |
+
"grad_norm": 24.25,
|
164 |
"learning_rate": 4.626245458345211e-06,
|
165 |
+
"logits/chosen": 83.37003326416016,
|
166 |
+
"logits/rejected": 83.38694763183594,
|
167 |
+
"logps/chosen": -34.16094207763672,
|
168 |
+
"logps/rejected": -31.8507137298584,
|
169 |
+
"loss": 0.6859,
|
170 |
+
"rewards/accuracies": 0.5874999761581421,
|
171 |
+
"rewards/chosen": 0.0720243752002716,
|
172 |
+
"rewards/margins": 0.12210811674594879,
|
173 |
+
"rewards/rejected": -0.05008373409509659,
|
174 |
"step": 100
|
175 |
},
|
176 |
{
|
177 |
"epoch": 0.26,
|
178 |
+
"eval_logits/chosen": 98.68260955810547,
|
179 |
+
"eval_logits/rejected": 98.67269134521484,
|
180 |
+
"eval_logps/chosen": -32.555023193359375,
|
181 |
+
"eval_logps/rejected": -36.136436462402344,
|
182 |
+
"eval_loss": 0.7229137420654297,
|
183 |
+
"eval_rewards/accuracies": 0.5282392501831055,
|
184 |
+
"eval_rewards/chosen": -0.05592246353626251,
|
185 |
+
"eval_rewards/margins": 0.029031021520495415,
|
186 |
+
"eval_rewards/rejected": -0.08495348691940308,
|
187 |
+
"eval_runtime": 104.2675,
|
188 |
+
"eval_samples_per_second": 3.29,
|
189 |
+
"eval_steps_per_second": 0.412,
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
+
"grad_norm": 35.25,
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
+
"logits/chosen": 83.4867172241211,
|
197 |
+
"logits/rejected": 83.372802734375,
|
198 |
+
"logps/chosen": -32.426979064941406,
|
199 |
+
"logps/rejected": -32.882118225097656,
|
200 |
+
"loss": 0.5808,
|
201 |
+
"rewards/accuracies": 0.75,
|
202 |
+
"rewards/chosen": 0.2979138493537903,
|
203 |
+
"rewards/margins": 0.4110265374183655,
|
204 |
+
"rewards/rejected": -0.11311274766921997,
|
205 |
"step": 110
|
206 |
},
|
207 |
{
|
208 |
"epoch": 0.31,
|
209 |
+
"grad_norm": 38.5,
|
210 |
"learning_rate": 4.353806263777678e-06,
|
211 |
+
"logits/chosen": 83.65151977539062,
|
212 |
+
"logits/rejected": 83.7606201171875,
|
213 |
+
"logps/chosen": -28.169925689697266,
|
214 |
+
"logps/rejected": -35.475196838378906,
|
215 |
+
"loss": 0.6054,
|
216 |
+
"rewards/accuracies": 0.612500011920929,
|
217 |
+
"rewards/chosen": 0.37982815504074097,
|
218 |
+
"rewards/margins": 0.3878515660762787,
|
219 |
+
"rewards/rejected": -0.008023401722311974,
|
220 |
"step": 120
|
221 |
},
|
222 |
{
|
223 |
"epoch": 0.34,
|
224 |
+
"grad_norm": 24.0,
|
225 |
"learning_rate": 4.1940827077152755e-06,
|
226 |
+
"logits/chosen": 80.93061065673828,
|
227 |
+
"logits/rejected": 80.95037841796875,
|
228 |
+
"logps/chosen": -30.428030014038086,
|
229 |
+
"logps/rejected": -32.12748336791992,
|
230 |
+
"loss": 0.6058,
|
231 |
+
"rewards/accuracies": 0.6625000238418579,
|
232 |
+
"rewards/chosen": 0.23966185748577118,
|
233 |
+
"rewards/margins": 0.33758825063705444,
|
234 |
+
"rewards/rejected": -0.09792639315128326,
|
235 |
"step": 130
|
236 |
},
|
237 |
{
|
238 |
"epoch": 0.36,
|
239 |
+
"grad_norm": 17.75,
|
240 |
"learning_rate": 4.0204024186666215e-06,
|
241 |
+
"logits/chosen": 82.01722717285156,
|
242 |
+
"logits/rejected": 82.02776336669922,
|
243 |
+
"logps/chosen": -27.013805389404297,
|
244 |
+
"logps/rejected": -33.108734130859375,
|
245 |
+
"loss": 0.5348,
|
246 |
+
"rewards/accuracies": 0.737500011920929,
|
247 |
+
"rewards/chosen": 0.2167574167251587,
|
248 |
+
"rewards/margins": 0.5594369173049927,
|
249 |
+
"rewards/rejected": -0.3426794409751892,
|
250 |
"step": 140
|
251 |
},
|
252 |
{
|
253 |
"epoch": 0.39,
|
254 |
+
"grad_norm": 26.125,
|
255 |
"learning_rate": 3.834196265035119e-06,
|
256 |
+
"logits/chosen": 80.49574279785156,
|
257 |
+
"logits/rejected": 80.46543884277344,
|
258 |
+
"logps/chosen": -28.96371078491211,
|
259 |
+
"logps/rejected": -33.28276443481445,
|
260 |
+
"loss": 0.5393,
|
261 |
+
"rewards/accuracies": 0.75,
|
262 |
+
"rewards/chosen": 0.21269837021827698,
|
263 |
+
"rewards/margins": 0.5708296895027161,
|
264 |
+
"rewards/rejected": -0.3581313192844391,
|
265 |
"step": 150
|
266 |
},
|
267 |
{
|
268 |
"epoch": 0.42,
|
269 |
+
"grad_norm": 36.75,
|
270 |
"learning_rate": 3.636998309800573e-06,
|
271 |
+
"logits/chosen": 82.2945556640625,
|
272 |
+
"logits/rejected": 82.3069839477539,
|
273 |
+
"logps/chosen": -33.60367202758789,
|
274 |
+
"logps/rejected": -30.597631454467773,
|
275 |
+
"loss": 0.5835,
|
276 |
+
"rewards/accuracies": 0.7124999761581421,
|
277 |
+
"rewards/chosen": 0.266539990901947,
|
278 |
+
"rewards/margins": 0.5890872478485107,
|
279 |
+
"rewards/rejected": -0.3225473165512085,
|
280 |
"step": 160
|
281 |
},
|
282 |
{
|
283 |
"epoch": 0.44,
|
284 |
+
"grad_norm": 26.25,
|
285 |
"learning_rate": 3.4304331721118078e-06,
|
286 |
+
"logits/chosen": 83.09800720214844,
|
287 |
+
"logits/rejected": 83.04072570800781,
|
288 |
+
"logps/chosen": -30.741390228271484,
|
289 |
+
"logps/rejected": -32.57554626464844,
|
290 |
+
"loss": 0.5698,
|
291 |
+
"rewards/accuracies": 0.675000011920929,
|
292 |
+
"rewards/chosen": 0.25894853472709656,
|
293 |
+
"rewards/margins": 0.5366927981376648,
|
294 |
+
"rewards/rejected": -0.27774426341056824,
|
295 |
"step": 170
|
296 |
},
|
297 |
{
|
298 |
"epoch": 0.47,
|
299 |
+
"grad_norm": 23.625,
|
300 |
"learning_rate": 3.2162026428305436e-06,
|
301 |
+
"logits/chosen": 80.62547302246094,
|
302 |
+
"logits/rejected": 80.6036605834961,
|
303 |
+
"logps/chosen": -30.658870697021484,
|
304 |
+
"logps/rejected": -31.662927627563477,
|
305 |
+
"loss": 0.55,
|
306 |
+
"rewards/accuracies": 0.7250000238418579,
|
307 |
+
"rewards/chosen": 0.2691388726234436,
|
308 |
+
"rewards/margins": 0.5219991207122803,
|
309 |
+
"rewards/rejected": -0.25286024808883667,
|
310 |
"step": 180
|
311 |
},
|
312 |
{
|
313 |
"epoch": 0.49,
|
314 |
+
"grad_norm": 12.4375,
|
315 |
"learning_rate": 2.996071664294641e-06,
|
316 |
+
"logits/chosen": 82.23011779785156,
|
317 |
+
"logits/rejected": 82.23200225830078,
|
318 |
+
"logps/chosen": -30.246551513671875,
|
319 |
+
"logps/rejected": -30.743694305419922,
|
320 |
+
"loss": 0.6343,
|
321 |
+
"rewards/accuracies": 0.637499988079071,
|
322 |
+
"rewards/chosen": 0.25993233919143677,
|
323 |
+
"rewards/margins": 0.3977679908275604,
|
324 |
+
"rewards/rejected": -0.13783565163612366,
|
325 |
"step": 190
|
326 |
},
|
327 |
{
|
328 |
"epoch": 0.52,
|
329 |
+
"grad_norm": 14.625,
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
+
"logits/chosen": 77.68287658691406,
|
332 |
+
"logits/rejected": 77.6307144165039,
|
333 |
+
"logps/chosen": -33.85746383666992,
|
334 |
+
"logps/rejected": -32.90498352050781,
|
335 |
+
"loss": 0.5453,
|
336 |
+
"rewards/accuracies": 0.7250000238418579,
|
337 |
+
"rewards/chosen": 0.44763606786727905,
|
338 |
+
"rewards/margins": 0.6534249186515808,
|
339 |
+
"rewards/rejected": -0.20578882098197937,
|
340 |
"step": 200
|
341 |
},
|
342 |
{
|
343 |
"epoch": 0.52,
|
344 |
+
"eval_logits/chosen": 98.49555206298828,
|
345 |
+
"eval_logits/rejected": 98.47319030761719,
|
346 |
+
"eval_logps/chosen": -32.7014274597168,
|
347 |
+
"eval_logps/rejected": -36.39837646484375,
|
348 |
+
"eval_loss": 0.7067598700523376,
|
349 |
+
"eval_rewards/accuracies": 0.5332226157188416,
|
350 |
+
"eval_rewards/chosen": -0.12912562489509583,
|
351 |
+
"eval_rewards/margins": 0.0867983028292656,
|
352 |
+
"eval_rewards/rejected": -0.21592393517494202,
|
353 |
+
"eval_runtime": 103.9124,
|
354 |
+
"eval_samples_per_second": 3.301,
|
355 |
"eval_steps_per_second": 0.414,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
+
"grad_norm": 46.5,
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
+
"logits/chosen": 80.28144836425781,
|
363 |
+
"logits/rejected": 80.18412780761719,
|
364 |
+
"logps/chosen": -33.1484489440918,
|
365 |
+
"logps/rejected": -35.34803009033203,
|
366 |
+
"loss": 0.5401,
|
367 |
+
"rewards/accuracies": 0.762499988079071,
|
368 |
+
"rewards/chosen": 0.39863476157188416,
|
369 |
+
"rewards/margins": 0.5796052813529968,
|
370 |
+
"rewards/rejected": -0.18097054958343506,
|
371 |
"step": 210
|
372 |
},
|
373 |
{
|
374 |
"epoch": 0.57,
|
375 |
+
"grad_norm": 18.875,
|
376 |
"learning_rate": 2.3185646976551794e-06,
|
377 |
+
"logits/chosen": 82.44231414794922,
|
378 |
+
"logits/rejected": 82.5237808227539,
|
379 |
+
"logps/chosen": -31.035247802734375,
|
380 |
+
"logps/rejected": -31.218652725219727,
|
381 |
+
"loss": 0.5206,
|
382 |
+
"rewards/accuracies": 0.762499988079071,
|
383 |
+
"rewards/chosen": 0.45685210824012756,
|
384 |
+
"rewards/margins": 0.7017010450363159,
|
385 |
+
"rewards/rejected": -0.24484892189502716,
|
386 |
"step": 220
|
387 |
},
|
388 |
{
|
389 |
"epoch": 0.6,
|
390 |
+
"grad_norm": 27.5,
|
391 |
"learning_rate": 2.0932279108998323e-06,
|
392 |
+
"logits/chosen": 79.57551574707031,
|
393 |
+
"logits/rejected": 79.63117218017578,
|
394 |
+
"logps/chosen": -32.342308044433594,
|
395 |
+
"logps/rejected": -34.377655029296875,
|
396 |
+
"loss": 0.6337,
|
397 |
"rewards/accuracies": 0.637499988079071,
|
398 |
+
"rewards/chosen": 0.23171910643577576,
|
399 |
+
"rewards/margins": 0.4144717752933502,
|
400 |
+
"rewards/rejected": -0.18275271356105804,
|
401 |
"step": 230
|
402 |
},
|
403 |
{
|
404 |
"epoch": 0.62,
|
405 |
+
"grad_norm": 29.375,
|
406 |
"learning_rate": 1.8712423238279358e-06,
|
407 |
+
"logits/chosen": 81.96827697753906,
|
408 |
+
"logits/rejected": 82.25470733642578,
|
409 |
+
"logps/chosen": -30.665319442749023,
|
410 |
+
"logps/rejected": -31.84256362915039,
|
411 |
+
"loss": 0.5035,
|
412 |
+
"rewards/accuracies": 0.737500011920929,
|
413 |
+
"rewards/chosen": 0.5113734006881714,
|
414 |
+
"rewards/margins": 0.6666342616081238,
|
415 |
+
"rewards/rejected": -0.1552608758211136,
|
416 |
"step": 240
|
417 |
},
|
418 |
{
|
419 |
"epoch": 0.65,
|
420 |
+
"grad_norm": 27.125,
|
421 |
"learning_rate": 1.6544367689701824e-06,
|
422 |
+
"logits/chosen": 80.69163513183594,
|
423 |
+
"logits/rejected": 80.74885559082031,
|
424 |
+
"logps/chosen": -26.948383331298828,
|
425 |
+
"logps/rejected": -30.213348388671875,
|
426 |
+
"loss": 0.6037,
|
427 |
+
"rewards/accuracies": 0.6625000238418579,
|
428 |
+
"rewards/chosen": 0.32375770807266235,
|
429 |
+
"rewards/margins": 0.48006218671798706,
|
430 |
+
"rewards/rejected": -0.1563045233488083,
|
431 |
"step": 250
|
432 |
},
|
433 |
{
|
434 |
"epoch": 0.68,
|
435 |
+
"grad_norm": 27.75,
|
436 |
"learning_rate": 1.4445974030621963e-06,
|
437 |
+
"logits/chosen": 77.91614532470703,
|
438 |
+
"logits/rejected": 78.04708099365234,
|
439 |
+
"logps/chosen": -30.49196434020996,
|
440 |
+
"logps/rejected": -36.51909255981445,
|
441 |
+
"loss": 0.4936,
|
442 |
+
"rewards/accuracies": 0.7250000238418579,
|
443 |
+
"rewards/chosen": 0.5345155000686646,
|
444 |
+
"rewards/margins": 0.7432474493980408,
|
445 |
+
"rewards/rejected": -0.2087319791316986,
|
446 |
"step": 260
|
447 |
},
|
448 |
{
|
449 |
"epoch": 0.7,
|
450 |
+
"grad_norm": 16.75,
|
451 |
"learning_rate": 1.243452991757889e-06,
|
452 |
+
"logits/chosen": 77.25732421875,
|
453 |
+
"logits/rejected": 77.28456115722656,
|
454 |
+
"logps/chosen": -30.877111434936523,
|
455 |
+
"logps/rejected": -31.975589752197266,
|
456 |
+
"loss": 0.5167,
|
457 |
+
"rewards/accuracies": 0.737500011920929,
|
458 |
+
"rewards/chosen": 0.4509071707725525,
|
459 |
+
"rewards/margins": 0.685990035533905,
|
460 |
+
"rewards/rejected": -0.23508290946483612,
|
461 |
"step": 270
|
462 |
},
|
463 |
{
|
464 |
"epoch": 0.73,
|
465 |
+
"grad_norm": 25.125,
|
466 |
"learning_rate": 1.0526606671603523e-06,
|
467 |
+
"logits/chosen": 80.03535461425781,
|
468 |
+
"logits/rejected": 79.8197250366211,
|
469 |
+
"logps/chosen": -31.083843231201172,
|
470 |
+
"logps/rejected": -29.844961166381836,
|
471 |
+
"loss": 0.6198,
|
472 |
+
"rewards/accuracies": 0.675000011920929,
|
473 |
+
"rewards/chosen": 0.3630252182483673,
|
474 |
+
"rewards/margins": 0.45200806856155396,
|
475 |
+
"rewards/rejected": -0.08898283541202545,
|
476 |
"step": 280
|
477 |
},
|
478 |
{
|
479 |
"epoch": 0.75,
|
480 |
+
"grad_norm": 16.625,
|
481 |
"learning_rate": 8.737922755071455e-07,
|
482 |
+
"logits/chosen": 80.13519287109375,
|
483 |
+
"logits/rejected": 80.05864715576172,
|
484 |
+
"logps/chosen": -33.047523498535156,
|
485 |
+
"logps/rejected": -32.68352127075195,
|
486 |
+
"loss": 0.4787,
|
487 |
+
"rewards/accuracies": 0.75,
|
488 |
+
"rewards/chosen": 0.5283600091934204,
|
489 |
+
"rewards/margins": 0.8616873025894165,
|
490 |
+
"rewards/rejected": -0.3333272337913513,
|
491 |
"step": 290
|
492 |
},
|
493 |
{
|
494 |
"epoch": 0.78,
|
495 |
+
"grad_norm": 26.875,
|
496 |
"learning_rate": 7.08321427484816e-07,
|
497 |
+
"logits/chosen": 75.82341003417969,
|
498 |
+
"logits/rejected": 75.91938781738281,
|
499 |
+
"logps/chosen": -32.24137878417969,
|
500 |
+
"logps/rejected": -29.33247947692871,
|
501 |
+
"loss": 0.5511,
|
502 |
+
"rewards/accuracies": 0.6625000238418579,
|
503 |
+
"rewards/chosen": 0.5821641087532043,
|
504 |
+
"rewards/margins": 0.6973918676376343,
|
505 |
+
"rewards/rejected": -0.11522769927978516,
|
506 |
"step": 300
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.78,
|
510 |
+
"eval_logits/chosen": 98.5571517944336,
|
511 |
+
"eval_logits/rejected": 98.53510284423828,
|
512 |
+
"eval_logps/chosen": -32.68053436279297,
|
513 |
+
"eval_logps/rejected": -36.284751892089844,
|
514 |
+
"eval_loss": 0.7289925813674927,
|
515 |
+
"eval_rewards/accuracies": 0.545265793800354,
|
516 |
+
"eval_rewards/chosen": -0.11868056654930115,
|
517 |
+
"eval_rewards/margins": 0.04043150320649147,
|
518 |
+
"eval_rewards/rejected": -0.1591120809316635,
|
519 |
+
"eval_runtime": 104.072,
|
520 |
+
"eval_samples_per_second": 3.296,
|
521 |
"eval_steps_per_second": 0.413,
|
522 |
"step": 300
|
523 |
},
|
524 |
{
|
525 |
"epoch": 0.81,
|
526 |
+
"grad_norm": 21.625,
|
527 |
"learning_rate": 5.576113578589035e-07,
|
528 |
+
"logits/chosen": 82.95035552978516,
|
529 |
+
"logits/rejected": 82.97549438476562,
|
530 |
+
"logps/chosen": -30.2395076751709,
|
531 |
+
"logps/rejected": -32.58390808105469,
|
532 |
+
"loss": 0.5895,
|
533 |
+
"rewards/accuracies": 0.7124999761581421,
|
534 |
+
"rewards/chosen": 0.31201133131980896,
|
535 |
+
"rewards/margins": 0.5040835738182068,
|
536 |
+
"rewards/rejected": -0.19207225739955902,
|
537 |
"step": 310
|
538 |
},
|
539 |
{
|
540 |
"epoch": 0.83,
|
541 |
+
"grad_norm": 18.0,
|
542 |
"learning_rate": 4.229036944380913e-07,
|
543 |
+
"logits/chosen": 80.41062927246094,
|
544 |
+
"logits/rejected": 80.41141510009766,
|
545 |
+
"logps/chosen": -30.43392562866211,
|
546 |
+
"logps/rejected": -29.273609161376953,
|
547 |
+
"loss": 0.4816,
|
548 |
+
"rewards/accuracies": 0.762499988079071,
|
549 |
+
"rewards/chosen": 0.5823162794113159,
|
550 |
+
"rewards/margins": 0.7593742609024048,
|
551 |
+
"rewards/rejected": -0.1770579218864441,
|
552 |
"step": 320
|
553 |
},
|
554 |
{
|
555 |
"epoch": 0.86,
|
556 |
+
"grad_norm": 16.625,
|
557 |
"learning_rate": 3.053082288996112e-07,
|
558 |
+
"logits/chosen": 77.58735656738281,
|
559 |
+
"logits/rejected": 77.6371841430664,
|
560 |
+
"logps/chosen": -29.118694305419922,
|
561 |
+
"logps/rejected": -33.023460388183594,
|
562 |
+
"loss": 0.4851,
|
563 |
+
"rewards/accuracies": 0.762499988079071,
|
564 |
+
"rewards/chosen": 0.6169556975364685,
|
565 |
+
"rewards/margins": 0.8112555742263794,
|
566 |
+
"rewards/rejected": -0.19429990649223328,
|
567 |
"step": 330
|
568 |
},
|
569 |
{
|
570 |
"epoch": 0.88,
|
571 |
+
"grad_norm": 33.5,
|
572 |
"learning_rate": 2.0579377374915805e-07,
|
573 |
+
"logits/chosen": 81.87162780761719,
|
574 |
+
"logits/rejected": 81.90576171875,
|
575 |
+
"logps/chosen": -32.27037048339844,
|
576 |
+
"logps/rejected": -33.92435836791992,
|
577 |
+
"loss": 0.5157,
|
578 |
+
"rewards/accuracies": 0.75,
|
579 |
+
"rewards/chosen": 0.4708884358406067,
|
580 |
+
"rewards/margins": 0.7487217783927917,
|
581 |
+
"rewards/rejected": -0.27783337235450745,
|
582 |
"step": 340
|
583 |
},
|
584 |
{
|
585 |
"epoch": 0.91,
|
586 |
+
"grad_norm": 13.0,
|
587 |
"learning_rate": 1.2518018074041684e-07,
|
588 |
+
"logits/chosen": 80.89490509033203,
|
589 |
+
"logits/rejected": 80.90391540527344,
|
590 |
+
"logps/chosen": -32.372520446777344,
|
591 |
+
"logps/rejected": -33.53419876098633,
|
592 |
+
"loss": 0.4983,
|
593 |
+
"rewards/accuracies": 0.762499988079071,
|
594 |
+
"rewards/chosen": 0.6356090307235718,
|
595 |
+
"rewards/margins": 0.8568087816238403,
|
596 |
+
"rewards/rejected": -0.22119958698749542,
|
597 |
"step": 350
|
598 |
},
|
599 |
{
|
600 |
"epoch": 0.94,
|
601 |
+
"grad_norm": 20.125,
|
602 |
"learning_rate": 6.41315865106129e-08,
|
603 |
+
"logits/chosen": 82.41497802734375,
|
604 |
+
"logits/rejected": 82.44314575195312,
|
605 |
+
"logps/chosen": -28.537517547607422,
|
606 |
+
"logps/rejected": -31.826416015625,
|
607 |
+
"loss": 0.5313,
|
608 |
+
"rewards/accuracies": 0.6875,
|
609 |
+
"rewards/chosen": 0.5073467493057251,
|
610 |
+
"rewards/margins": 0.5956294536590576,
|
611 |
+
"rewards/rejected": -0.08828280866146088,
|
612 |
"step": 360
|
613 |
},
|
614 |
{
|
615 |
"epoch": 0.96,
|
616 |
+
"grad_norm": 21.75,
|
617 |
"learning_rate": 2.3150941078050325e-08,
|
618 |
+
"logits/chosen": 81.85018920898438,
|
619 |
+
"logits/rejected": 81.87150573730469,
|
620 |
+
"logps/chosen": -31.847997665405273,
|
621 |
+
"logps/rejected": -35.508277893066406,
|
622 |
+
"loss": 0.5486,
|
623 |
+
"rewards/accuracies": 0.6875,
|
624 |
+
"rewards/chosen": 0.5140814781188965,
|
625 |
+
"rewards/margins": 0.7135335206985474,
|
626 |
+
"rewards/rejected": -0.19945205748081207,
|
627 |
"step": 370
|
628 |
},
|
629 |
{
|
630 |
"epoch": 0.99,
|
631 |
+
"grad_norm": 25.5,
|
632 |
"learning_rate": 2.575864278703266e-09,
|
633 |
+
"logits/chosen": 75.74685668945312,
|
634 |
+
"logits/rejected": 75.61808013916016,
|
635 |
+
"logps/chosen": -29.746694564819336,
|
636 |
+
"logps/rejected": -28.432201385498047,
|
637 |
+
"loss": 0.5715,
|
638 |
+
"rewards/accuracies": 0.699999988079071,
|
639 |
+
"rewards/chosen": 0.4147160053253174,
|
640 |
+
"rewards/margins": 0.5505291223526001,
|
641 |
+
"rewards/rejected": -0.1358131319284439,
|
642 |
"step": 380
|
643 |
},
|
644 |
{
|
645 |
"epoch": 1.0,
|
646 |
"step": 385,
|
647 |
"total_flos": 0.0,
|
648 |
+
"train_loss": 0.5860598966672823,
|
649 |
+
"train_runtime": 2558.949,
|
650 |
+
"train_samples_per_second": 1.203,
|
651 |
+
"train_steps_per_second": 0.15
|
652 |
}
|
653 |
],
|
654 |
"logging_steps": 10,
|