hugodk-sch
commited on
Commit
•
20b77c6
1
Parent(s):
925fd5f
Model save
Browse files- README.md +13 -16
- adapter_model.safetensors +1 -1
- all_results.json +4 -17
- train_results.json +4 -4
- trainer_state.json +412 -412
README.md
CHANGED
@@ -1,13 +1,10 @@
|
|
1 |
---
|
2 |
library_name: peft
|
3 |
tags:
|
4 |
-
- alignment-handbook
|
5 |
- trl
|
6 |
- dpo
|
7 |
- generated_from_trainer
|
8 |
base_model: NorLLM-AI/NorMistral-7B
|
9 |
-
datasets:
|
10 |
-
- hugodk-sch/aftonposten_title_prefs
|
11 |
model-index:
|
12 |
- name: norllm-ai-normistral-7b-align-scan
|
13 |
results: []
|
@@ -18,17 +15,17 @@ should probably proofread and complete it, then remove this comment. -->
|
|
18 |
|
19 |
# norllm-ai-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: -35.
|
29 |
-
- Logps/chosen: -
|
30 |
-
- Logits/rejected: -2.
|
31 |
-
- Logits/chosen: -2.
|
32 |
|
33 |
## Model description
|
34 |
|
@@ -63,9 +60,9 @@ 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 |
library_name: peft
|
3 |
tags:
|
|
|
4 |
- trl
|
5 |
- dpo
|
6 |
- generated_from_trainer
|
7 |
base_model: NorLLM-AI/NorMistral-7B
|
|
|
|
|
8 |
model-index:
|
9 |
- name: norllm-ai-normistral-7b-align-scan
|
10 |
results: []
|
|
|
15 |
|
16 |
# norllm-ai-normistral-7b-align-scan
|
17 |
|
18 |
+
This model is a fine-tuned version of [NorLLM-AI/NorMistral-7B](https://huggingface.co/NorLLM-AI/NorMistral-7B) on the None dataset.
|
19 |
It achieves the following results on the evaluation set:
|
20 |
+
- Loss: 0.9559
|
21 |
+
- Rewards/chosen: -0.0753
|
22 |
+
- Rewards/rejected: -0.1195
|
23 |
+
- Rewards/accuracies: 0.6005
|
24 |
+
- Rewards/margins: 0.0441
|
25 |
+
- Logps/rejected: -35.8927
|
26 |
+
- Logps/chosen: -32.0356
|
27 |
+
- Logits/rejected: -2.8365
|
28 |
+
- Logits/chosen: -2.8392
|
29 |
|
30 |
## Model description
|
31 |
|
|
|
60 |
|
61 |
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
62 |
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
63 |
+
| 0.9581 | 0.26 | 100 | 0.9819 | 0.0013 | -0.0170 | 0.5955 | 0.0182 | -34.8678 | -31.2699 | -2.8027 | -2.8053 |
|
64 |
+
| 0.8824 | 0.52 | 200 | 0.9602 | -0.0546 | -0.0945 | 0.6092 | 0.0399 | -35.6429 | -31.8280 | -2.8288 | -2.8315 |
|
65 |
+
| 0.8245 | 0.78 | 300 | 0.9559 | -0.0753 | -0.1195 | 0.6005 | 0.0441 | -35.8927 | -32.0356 | -2.8365 | -2.8392 |
|
66 |
|
67 |
|
68 |
### 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 sha256:7a283772062be6a02cdb1fec51f20731bfc277bc3a6038cdf2bb5811ddde0a0f
|
3 |
size 671150064
|
all_results.json
CHANGED
@@ -1,21 +1,8 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"
|
4 |
-
"
|
5 |
-
"eval_logps/chosen": -31.491378784179688,
|
6 |
-
"eval_logps/rejected": -35.14382553100586,
|
7 |
-
"eval_loss": 0.9104546904563904,
|
8 |
-
"eval_rewards/accuracies": 0.5514950156211853,
|
9 |
-
"eval_rewards/chosen": -0.08357102423906326,
|
10 |
-
"eval_rewards/margins": 0.09469697624444962,
|
11 |
-
"eval_rewards/rejected": -0.17826800048351288,
|
12 |
-
"eval_runtime": 112.9595,
|
13 |
-
"eval_samples": 343,
|
14 |
-
"eval_samples_per_second": 3.036,
|
15 |
-
"eval_steps_per_second": 0.381,
|
16 |
-
"train_loss": 0.7203964704042906,
|
17 |
-
"train_runtime": 2720.5927,
|
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": 0.8805358874333369,
|
4 |
+
"train_runtime": 2723.8224,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
"train_samples": 3079,
|
6 |
+
"train_samples_per_second": 1.13,
|
7 |
+
"train_steps_per_second": 0.141
|
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": 0.8805358874333369,
|
4 |
+
"train_runtime": 2723.8224,
|
5 |
"train_samples": 3079,
|
6 |
+
"train_samples_per_second": 1.13,
|
7 |
+
"train_steps_per_second": 0.141
|
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": -2.7358343601226807,
|
16 |
"logits/rejected": -2.7480404376983643,
|
@@ -25,630 +25,630 @@
|
|
25 |
},
|
26 |
{
|
27 |
"epoch": 0.03,
|
28 |
-
"grad_norm":
|
29 |
"learning_rate": 1.282051282051282e-06,
|
30 |
-
"logits/chosen": -3.
|
31 |
-
"logits/rejected": -2.
|
32 |
-
"logps/chosen": -33.
|
33 |
-
"logps/rejected": -31.
|
34 |
-
"loss": 1.
|
35 |
-
"rewards/accuracies": 0.
|
36 |
-
"rewards/chosen": -0.
|
37 |
-
"rewards/margins": -0.
|
38 |
-
"rewards/rejected":
|
39 |
"step": 10
|
40 |
},
|
41 |
{
|
42 |
"epoch": 0.05,
|
43 |
-
"grad_norm":
|
44 |
"learning_rate": 2.564102564102564e-06,
|
45 |
-
"logits/chosen": -2.
|
46 |
-
"logits/rejected": -2.
|
47 |
-
"logps/chosen": -32.
|
48 |
-
"logps/rejected": -28.
|
49 |
-
"loss": 1.
|
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": -3.
|
61 |
-
"logits/rejected": -3.
|
62 |
-
"logps/chosen": -32.
|
63 |
-
"logps/rejected": -30.
|
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": -2.
|
76 |
-
"logits/rejected": -2.
|
77 |
-
"logps/chosen": -31.
|
78 |
-
"logps/rejected": -32.
|
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": -2.
|
91 |
-
"logits/rejected": -2.
|
92 |
-
"logps/chosen": -29.
|
93 |
-
"logps/rejected": -30.
|
94 |
-
"loss": 0.
|
95 |
-
"rewards/accuracies": 0.
|
96 |
-
"rewards/chosen": 0.
|
97 |
-
"rewards/margins": 0.
|
98 |
-
"rewards/rejected":
|
99 |
"step": 50
|
100 |
},
|
101 |
{
|
102 |
"epoch": 0.16,
|
103 |
-
"grad_norm":
|
104 |
"learning_rate": 4.954691471941119e-06,
|
105 |
-
"logits/chosen": -2.
|
106 |
-
"logits/rejected": -2.
|
107 |
-
"logps/chosen": -29.
|
108 |
-
"logps/rejected": -
|
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": -2.
|
121 |
-
"logits/rejected": -2.
|
122 |
-
"logps/chosen": -29.
|
123 |
-
"logps/rejected": -30.
|
124 |
-
"loss": 0.
|
125 |
-
"rewards/accuracies": 0.
|
126 |
-
"rewards/chosen": 0.
|
127 |
-
"rewards/margins": 0.
|
128 |
-
"rewards/rejected": -
|
129 |
"step": 70
|
130 |
},
|
131 |
{
|
132 |
"epoch": 0.21,
|
133 |
-
"grad_norm":
|
134 |
"learning_rate": 4.828760511501322e-06,
|
135 |
-
"logits/chosen": -2.
|
136 |
-
"logits/rejected": -2.
|
137 |
-
"logps/chosen": -29.
|
138 |
-
"logps/rejected": -29.
|
139 |
-
"loss": 0.
|
140 |
"rewards/accuracies": 0.637499988079071,
|
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": -2.
|
151 |
-
"logits/rejected": -2.
|
152 |
-
"logps/chosen": -32.
|
153 |
-
"logps/rejected": -30.
|
154 |
-
"loss": 0.
|
155 |
"rewards/accuracies": 0.637499988079071,
|
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": -
|
166 |
-
"logits/rejected": -
|
167 |
-
"logps/chosen": -31.
|
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": -2.
|
179 |
-
"eval_logits/rejected": -2.
|
180 |
-
"eval_logps/chosen": -31.
|
181 |
-
"eval_logps/rejected": -34.
|
182 |
-
"eval_loss": 0.
|
183 |
-
"eval_rewards/accuracies": 0.
|
184 |
-
"eval_rewards/chosen":
|
185 |
-
"eval_rewards/margins": 0.
|
186 |
-
"eval_rewards/rejected": -0.
|
187 |
-
"eval_runtime": 113.
|
188 |
"eval_samples_per_second": 3.025,
|
189 |
"eval_steps_per_second": 0.379,
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
-
"grad_norm":
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
-
"logits/chosen": -2.
|
197 |
-
"logits/rejected": -2.
|
198 |
-
"logps/chosen": -31.
|
199 |
-
"logps/rejected": -31.
|
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": -3.
|
212 |
-
"logits/rejected": -3.
|
213 |
-
"logps/chosen": -28.
|
214 |
-
"logps/rejected": -34.
|
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": -2.
|
227 |
-
"logits/rejected": -2.
|
228 |
-
"logps/chosen": -28.
|
229 |
-
"logps/rejected": -30.
|
230 |
-
"loss": 0.
|
231 |
"rewards/accuracies": 0.6625000238418579,
|
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": -3.
|
242 |
-
"logits/rejected": -3.
|
243 |
-
"logps/chosen": -27.
|
244 |
-
"logps/rejected": -32.
|
245 |
-
"loss": 0.
|
246 |
-
"rewards/accuracies": 0.
|
247 |
-
"rewards/chosen": 0.
|
248 |
-
"rewards/margins": 0.
|
249 |
-
"rewards/rejected": -0.
|
250 |
"step": 140
|
251 |
},
|
252 |
{
|
253 |
"epoch": 0.39,
|
254 |
-
"grad_norm":
|
255 |
"learning_rate": 3.834196265035119e-06,
|
256 |
-
"logits/chosen": -2.
|
257 |
-
"logits/rejected": -2.
|
258 |
-
"logps/chosen": -27.
|
259 |
-
"logps/rejected": -
|
260 |
-
"loss": 0.
|
261 |
"rewards/accuracies": 0.625,
|
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": -3.
|
272 |
-
"logits/rejected": -3.
|
273 |
-
"logps/chosen": -31.
|
274 |
-
"logps/rejected": -
|
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": -2.
|
287 |
-
"logits/rejected": -2.
|
288 |
-
"logps/chosen": -29.
|
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": -2.
|
302 |
-
"logits/rejected": -2.
|
303 |
-
"logps/chosen": -29.
|
304 |
-
"logps/rejected": -30.
|
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": -2.
|
317 |
-
"logits/rejected": -2.
|
318 |
-
"logps/chosen": -
|
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":
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
-
"logits/chosen": -2.
|
332 |
-
"logits/rejected": -2.
|
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": -2.
|
345 |
-
"eval_logits/rejected": -2.
|
346 |
-
"eval_logps/chosen": -31.
|
347 |
-
"eval_logps/rejected": -35.
|
348 |
-
"eval_loss": 0.
|
349 |
-
"eval_rewards/accuracies": 0.
|
350 |
-
"eval_rewards/chosen": -0.
|
351 |
-
"eval_rewards/margins": 0.
|
352 |
-
"eval_rewards/rejected": -0.
|
353 |
-
"eval_runtime": 113.
|
354 |
-
"eval_samples_per_second": 3.
|
355 |
"eval_steps_per_second": 0.38,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
-
"grad_norm":
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
-
"logits/chosen": -2.
|
363 |
-
"logits/rejected": -2.
|
364 |
-
"logps/chosen": -
|
365 |
-
"logps/rejected": -
|
366 |
-
"loss": 0.
|
367 |
-
"rewards/accuracies": 0.
|
368 |
-
"rewards/chosen": 0.
|
369 |
-
"rewards/margins": 0.
|
370 |
-
"rewards/rejected": -0.
|
371 |
"step": 210
|
372 |
},
|
373 |
{
|
374 |
"epoch": 0.57,
|
375 |
-
"grad_norm":
|
376 |
"learning_rate": 2.3185646976551794e-06,
|
377 |
-
"logits/chosen": -2.
|
378 |
-
"logits/rejected": -2.
|
379 |
-
"logps/chosen": -29.
|
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": -2.
|
393 |
-
"logits/rejected": -2.
|
394 |
-
"logps/chosen": -
|
395 |
-
"logps/rejected": -
|
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": -3.
|
408 |
-
"logits/rejected": -3.
|
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": -2.
|
423 |
-
"logits/rejected": -2.
|
424 |
-
"logps/chosen": -
|
425 |
-
"logps/rejected": -
|
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": -2.
|
438 |
-
"logits/rejected": -2.
|
439 |
-
"logps/chosen": -29.
|
440 |
-
"logps/rejected": -
|
441 |
-
"loss": 0.
|
442 |
-
"rewards/accuracies": 0.
|
443 |
-
"rewards/chosen": 0.
|
444 |
-
"rewards/margins": 0.
|
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": -2.
|
453 |
-
"logits/rejected": -2.
|
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": -2.
|
468 |
-
"logits/rejected": -2.
|
469 |
-
"logps/chosen": -
|
470 |
-
"logps/rejected": -
|
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": -2.
|
483 |
-
"logits/rejected": -2.
|
484 |
-
"logps/chosen": -
|
485 |
-
"logps/rejected": -
|
486 |
-
"loss": 0.
|
487 |
-
"rewards/accuracies": 0.
|
488 |
-
"rewards/chosen": 0.
|
489 |
-
"rewards/margins": 0.
|
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": -2.
|
498 |
-
"logits/rejected": -2.
|
499 |
-
"logps/chosen": -
|
500 |
-
"logps/rejected": -
|
501 |
-
"loss": 0.
|
502 |
"rewards/accuracies": 0.6875,
|
503 |
-
"rewards/chosen": 0.
|
504 |
-
"rewards/margins": 0.
|
505 |
-
"rewards/rejected": -0.
|
506 |
"step": 300
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.78,
|
510 |
-
"eval_logits/chosen": -2.
|
511 |
-
"eval_logits/rejected": -2.
|
512 |
-
"eval_logps/chosen": -
|
513 |
-
"eval_logps/rejected": -35.
|
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": -2.
|
529 |
-
"logits/rejected": -2.
|
530 |
-
"logps/chosen": -
|
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": -3.
|
544 |
-
"logits/rejected": -3.
|
545 |
-
"logps/chosen": -29.
|
546 |
-
"logps/rejected": -
|
547 |
-
"loss": 0.
|
548 |
-
"rewards/accuracies": 0.
|
549 |
-
"rewards/chosen": 0.
|
550 |
-
"rewards/margins": 0.
|
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": -2.
|
559 |
-
"logits/rejected": -2.
|
560 |
-
"logps/chosen": -
|
561 |
-
"logps/rejected": -
|
562 |
-
"loss": 0.
|
563 |
-
"rewards/accuracies": 0.
|
564 |
-
"rewards/chosen": 0.
|
565 |
-
"rewards/margins": 0.
|
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": -3.
|
574 |
-
"logits/rejected": -3.
|
575 |
-
"logps/chosen": -
|
576 |
-
"logps/rejected": -
|
577 |
-
"loss": 0.
|
578 |
-
"rewards/accuracies": 0.
|
579 |
-
"rewards/chosen": 0.
|
580 |
-
"rewards/margins": 0.
|
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": -3.
|
589 |
-
"logits/rejected": -3.
|
590 |
-
"logps/chosen": -
|
591 |
-
"logps/rejected": -
|
592 |
-
"loss": 0.
|
593 |
"rewards/accuracies": 0.762499988079071,
|
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": -2.
|
604 |
-
"logits/rejected": -2.
|
605 |
-
"logps/chosen": -
|
606 |
-
"logps/rejected": -
|
607 |
-
"loss": 0.
|
608 |
-
"rewards/accuracies": 0.
|
609 |
-
"rewards/chosen": 0.
|
610 |
-
"rewards/margins": 0.
|
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": -2.
|
619 |
-
"logits/rejected": -2.
|
620 |
-
"logps/chosen": -
|
621 |
-
"logps/rejected": -
|
622 |
-
"loss": 0.
|
623 |
-
"rewards/accuracies": 0.
|
624 |
-
"rewards/chosen": 0.
|
625 |
-
"rewards/margins": 0.
|
626 |
-
"rewards/rejected": -0.
|
627 |
"step": 370
|
628 |
},
|
629 |
{
|
630 |
"epoch": 0.99,
|
631 |
-
"grad_norm":
|
632 |
"learning_rate": 2.575864278703266e-09,
|
633 |
-
"logits/chosen": -2.
|
634 |
-
"logits/rejected": -2.
|
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":
|
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": 7.90625,
|
14 |
"learning_rate": 1.282051282051282e-07,
|
15 |
"logits/chosen": -2.7358343601226807,
|
16 |
"logits/rejected": -2.7480404376983643,
|
|
|
25 |
},
|
26 |
{
|
27 |
"epoch": 0.03,
|
28 |
+
"grad_norm": 13.4375,
|
29 |
"learning_rate": 1.282051282051282e-06,
|
30 |
+
"logits/chosen": -3.009408473968506,
|
31 |
+
"logits/rejected": -2.9977328777313232,
|
32 |
+
"logps/chosen": -33.20761489868164,
|
33 |
+
"logps/rejected": -31.960758209228516,
|
34 |
+
"loss": 1.003,
|
35 |
+
"rewards/accuracies": 0.375,
|
36 |
+
"rewards/chosen": -0.0026669783983379602,
|
37 |
+
"rewards/margins": -0.002981698838993907,
|
38 |
+
"rewards/rejected": 0.00031472102273255587,
|
39 |
"step": 10
|
40 |
},
|
41 |
{
|
42 |
"epoch": 0.05,
|
43 |
+
"grad_norm": 9.1875,
|
44 |
"learning_rate": 2.564102564102564e-06,
|
45 |
+
"logits/chosen": -2.8990566730499268,
|
46 |
+
"logits/rejected": -2.894094944000244,
|
47 |
+
"logps/chosen": -32.50642013549805,
|
48 |
+
"logps/rejected": -28.963293075561523,
|
49 |
+
"loss": 1.005,
|
50 |
+
"rewards/accuracies": 0.4000000059604645,
|
51 |
+
"rewards/chosen": -0.0031326524913311005,
|
52 |
+
"rewards/margins": -0.00499005988240242,
|
53 |
+
"rewards/rejected": 0.0018574076239019632,
|
54 |
"step": 20
|
55 |
},
|
56 |
{
|
57 |
"epoch": 0.08,
|
58 |
+
"grad_norm": 8.8125,
|
59 |
"learning_rate": 3.846153846153847e-06,
|
60 |
+
"logits/chosen": -3.0970427989959717,
|
61 |
+
"logits/rejected": -3.108668804168701,
|
62 |
+
"logps/chosen": -32.80826187133789,
|
63 |
+
"logps/rejected": -30.164159774780273,
|
64 |
+
"loss": 0.9931,
|
65 |
+
"rewards/accuracies": 0.550000011920929,
|
66 |
+
"rewards/chosen": 0.010042307898402214,
|
67 |
+
"rewards/margins": 0.006853123661130667,
|
68 |
+
"rewards/rejected": 0.0031891819089651108,
|
69 |
"step": 30
|
70 |
},
|
71 |
{
|
72 |
"epoch": 0.1,
|
73 |
+
"grad_norm": 9.0625,
|
74 |
"learning_rate": 4.999896948438434e-06,
|
75 |
+
"logits/chosen": -2.8629202842712402,
|
76 |
+
"logits/rejected": -2.8535947799682617,
|
77 |
+
"logps/chosen": -31.525354385375977,
|
78 |
+
"logps/rejected": -32.399391174316406,
|
79 |
+
"loss": 0.9653,
|
80 |
+
"rewards/accuracies": 0.612500011920929,
|
81 |
+
"rewards/chosen": 0.030497059226036072,
|
82 |
+
"rewards/margins": 0.034690458327531815,
|
83 |
+
"rewards/rejected": -0.004193395376205444,
|
84 |
"step": 40
|
85 |
},
|
86 |
{
|
87 |
"epoch": 0.13,
|
88 |
+
"grad_norm": 8.1875,
|
89 |
"learning_rate": 4.987541037542187e-06,
|
90 |
+
"logits/chosen": -2.87955904006958,
|
91 |
+
"logits/rejected": -2.877352714538574,
|
92 |
+
"logps/chosen": -29.324417114257812,
|
93 |
+
"logps/rejected": -30.05527114868164,
|
94 |
+
"loss": 0.9618,
|
95 |
+
"rewards/accuracies": 0.5625,
|
96 |
+
"rewards/chosen": 0.04157434031367302,
|
97 |
+
"rewards/margins": 0.03820687159895897,
|
98 |
+
"rewards/rejected": 0.0033674687147140503,
|
99 |
"step": 50
|
100 |
},
|
101 |
{
|
102 |
"epoch": 0.16,
|
103 |
+
"grad_norm": 8.0625,
|
104 |
"learning_rate": 4.954691471941119e-06,
|
105 |
+
"logits/chosen": -2.9058096408843994,
|
106 |
+
"logits/rejected": -2.90769624710083,
|
107 |
+
"logps/chosen": -29.792831420898438,
|
108 |
+
"logps/rejected": -27.967548370361328,
|
109 |
+
"loss": 0.9651,
|
110 |
+
"rewards/accuracies": 0.625,
|
111 |
+
"rewards/chosen": 0.03369411081075668,
|
112 |
+
"rewards/margins": 0.03488066419959068,
|
113 |
+
"rewards/rejected": -0.0011865533888339996,
|
114 |
"step": 60
|
115 |
},
|
116 |
{
|
117 |
"epoch": 0.18,
|
118 |
+
"grad_norm": 10.75,
|
119 |
"learning_rate": 4.901618883413549e-06,
|
120 |
+
"logits/chosen": -2.98918080329895,
|
121 |
+
"logits/rejected": -2.9952526092529297,
|
122 |
+
"logps/chosen": -29.09405517578125,
|
123 |
+
"logps/rejected": -30.891315460205078,
|
124 |
+
"loss": 0.9814,
|
125 |
+
"rewards/accuracies": 0.6000000238418579,
|
126 |
+
"rewards/chosen": 0.018691470846533775,
|
127 |
+
"rewards/margins": 0.018642673268914223,
|
128 |
+
"rewards/rejected": 4.879874177277088e-05,
|
129 |
"step": 70
|
130 |
},
|
131 |
{
|
132 |
"epoch": 0.21,
|
133 |
+
"grad_norm": 9.375,
|
134 |
"learning_rate": 4.828760511501322e-06,
|
135 |
+
"logits/chosen": -2.804455518722534,
|
136 |
+
"logits/rejected": -2.8193769454956055,
|
137 |
+
"logps/chosen": -29.27066993713379,
|
138 |
+
"logps/rejected": -29.863460540771484,
|
139 |
+
"loss": 0.9541,
|
140 |
"rewards/accuracies": 0.637499988079071,
|
141 |
+
"rewards/chosen": 0.027005871757864952,
|
142 |
+
"rewards/margins": 0.04592365399003029,
|
143 |
+
"rewards/rejected": -0.018917780369520187,
|
144 |
"step": 80
|
145 |
},
|
146 |
{
|
147 |
"epoch": 0.23,
|
148 |
+
"grad_norm": 9.5,
|
149 |
"learning_rate": 4.7367166013034295e-06,
|
150 |
+
"logits/chosen": -2.8940837383270264,
|
151 |
+
"logits/rejected": -2.8764431476593018,
|
152 |
+
"logps/chosen": -32.690391540527344,
|
153 |
+
"logps/rejected": -30.344482421875,
|
154 |
+
"loss": 0.9531,
|
155 |
"rewards/accuracies": 0.637499988079071,
|
156 |
+
"rewards/chosen": 0.014603940770030022,
|
157 |
+
"rewards/margins": 0.04688751697540283,
|
158 |
+
"rewards/rejected": -0.03228358179330826,
|
159 |
"step": 90
|
160 |
},
|
161 |
{
|
162 |
"epoch": 0.26,
|
163 |
+
"grad_norm": 8.4375,
|
164 |
"learning_rate": 4.626245458345211e-06,
|
165 |
+
"logits/chosen": -2.999018669128418,
|
166 |
+
"logits/rejected": -2.9999501705169678,
|
167 |
+
"logps/chosen": -31.824331283569336,
|
168 |
+
"logps/rejected": -31.003093719482422,
|
169 |
+
"loss": 0.9581,
|
170 |
+
"rewards/accuracies": 0.625,
|
171 |
+
"rewards/chosen": 0.023254599422216415,
|
172 |
+
"rewards/margins": 0.041946638375520706,
|
173 |
+
"rewards/rejected": -0.01869204081594944,
|
174 |
"step": 100
|
175 |
},
|
176 |
{
|
177 |
"epoch": 0.26,
|
178 |
+
"eval_logits/chosen": -2.805298089981079,
|
179 |
+
"eval_logits/rejected": -2.802727460861206,
|
180 |
+
"eval_logps/chosen": -31.269874572753906,
|
181 |
+
"eval_logps/rejected": -34.867759704589844,
|
182 |
+
"eval_loss": 0.9819382429122925,
|
183 |
+
"eval_rewards/accuracies": 0.5955149531364441,
|
184 |
+
"eval_rewards/chosen": 0.0012576259905472398,
|
185 |
+
"eval_rewards/margins": 0.018218038603663445,
|
186 |
+
"eval_rewards/rejected": -0.01696041226387024,
|
187 |
+
"eval_runtime": 113.3967,
|
188 |
"eval_samples_per_second": 3.025,
|
189 |
"eval_steps_per_second": 0.379,
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
+
"grad_norm": 10.4375,
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
+
"logits/chosen": -2.9547231197357178,
|
197 |
+
"logits/rejected": -2.9307026863098145,
|
198 |
+
"logps/chosen": -31.882299423217773,
|
199 |
+
"logps/rejected": -31.55330467224121,
|
200 |
+
"loss": 0.9271,
|
201 |
+
"rewards/accuracies": 0.675000011920929,
|
202 |
+
"rewards/chosen": 0.03344918042421341,
|
203 |
+
"rewards/margins": 0.07289471477270126,
|
204 |
+
"rewards/rejected": -0.039445530623197556,
|
205 |
"step": 110
|
206 |
},
|
207 |
{
|
208 |
"epoch": 0.31,
|
209 |
+
"grad_norm": 8.9375,
|
210 |
"learning_rate": 4.353806263777678e-06,
|
211 |
+
"logits/chosen": -3.0366623401641846,
|
212 |
+
"logits/rejected": -3.0655264854431152,
|
213 |
+
"logps/chosen": -28.671463012695312,
|
214 |
+
"logps/rejected": -34.533443450927734,
|
215 |
+
"loss": 0.9122,
|
216 |
+
"rewards/accuracies": 0.699999988079071,
|
217 |
+
"rewards/chosen": 0.049243707209825516,
|
218 |
+
"rewards/margins": 0.0878329798579216,
|
219 |
+
"rewards/rejected": -0.03858928382396698,
|
220 |
"step": 120
|
221 |
},
|
222 |
{
|
223 |
"epoch": 0.34,
|
224 |
+
"grad_norm": 8.5625,
|
225 |
"learning_rate": 4.1940827077152755e-06,
|
226 |
+
"logits/chosen": -2.741682529449463,
|
227 |
+
"logits/rejected": -2.7387285232543945,
|
228 |
+
"logps/chosen": -28.681751251220703,
|
229 |
+
"logps/rejected": -30.781078338623047,
|
230 |
+
"loss": 0.9069,
|
231 |
"rewards/accuracies": 0.6625000238418579,
|
232 |
+
"rewards/chosen": 0.029292941093444824,
|
233 |
+
"rewards/margins": 0.09402014315128326,
|
234 |
+
"rewards/rejected": -0.06472718715667725,
|
235 |
"step": 130
|
236 |
},
|
237 |
{
|
238 |
"epoch": 0.36,
|
239 |
+
"grad_norm": 8.6875,
|
240 |
"learning_rate": 4.0204024186666215e-06,
|
241 |
+
"logits/chosen": -3.0192549228668213,
|
242 |
+
"logits/rejected": -3.0171239376068115,
|
243 |
+
"logps/chosen": -27.48164939880371,
|
244 |
+
"logps/rejected": -32.18421936035156,
|
245 |
+
"loss": 0.9468,
|
246 |
+
"rewards/accuracies": 0.5625,
|
247 |
+
"rewards/chosen": 0.00045731887803412974,
|
248 |
+
"rewards/margins": 0.0532023087143898,
|
249 |
+
"rewards/rejected": -0.05274498462677002,
|
250 |
"step": 140
|
251 |
},
|
252 |
{
|
253 |
"epoch": 0.39,
|
254 |
+
"grad_norm": 8.8125,
|
255 |
"learning_rate": 3.834196265035119e-06,
|
256 |
+
"logits/chosen": -2.8165481090545654,
|
257 |
+
"logits/rejected": -2.8123271465301514,
|
258 |
+
"logps/chosen": -27.479320526123047,
|
259 |
+
"logps/rejected": -32.02801513671875,
|
260 |
+
"loss": 0.8849,
|
261 |
"rewards/accuracies": 0.625,
|
262 |
+
"rewards/chosen": 0.034618668258190155,
|
263 |
+
"rewards/margins": 0.11505589634180069,
|
264 |
+
"rewards/rejected": -0.08043723553419113,
|
265 |
"step": 150
|
266 |
},
|
267 |
{
|
268 |
"epoch": 0.42,
|
269 |
+
"grad_norm": 10.25,
|
270 |
"learning_rate": 3.636998309800573e-06,
|
271 |
+
"logits/chosen": -3.137382745742798,
|
272 |
+
"logits/rejected": -3.1210262775421143,
|
273 |
+
"logps/chosen": -31.968801498413086,
|
274 |
+
"logps/rejected": -30.1705265045166,
|
275 |
+
"loss": 0.8343,
|
276 |
+
"rewards/accuracies": 0.762499988079071,
|
277 |
+
"rewards/chosen": 0.05229174345731735,
|
278 |
+
"rewards/margins": 0.17136676609516144,
|
279 |
+
"rewards/rejected": -0.11907501518726349,
|
280 |
"step": 160
|
281 |
},
|
282 |
{
|
283 |
"epoch": 0.44,
|
284 |
+
"grad_norm": 9.5625,
|
285 |
"learning_rate": 3.4304331721118078e-06,
|
286 |
+
"logits/chosen": -2.9499218463897705,
|
287 |
+
"logits/rejected": -2.958042621612549,
|
288 |
+
"logps/chosen": -29.790802001953125,
|
289 |
+
"logps/rejected": -32.55553436279297,
|
290 |
+
"loss": 0.8471,
|
291 |
+
"rewards/accuracies": 0.6625000238418579,
|
292 |
+
"rewards/chosen": 0.010381256230175495,
|
293 |
+
"rewards/margins": 0.15598610043525696,
|
294 |
+
"rewards/rejected": -0.14560487866401672,
|
295 |
"step": 170
|
296 |
},
|
297 |
{
|
298 |
"epoch": 0.47,
|
299 |
+
"grad_norm": 10.0625,
|
300 |
"learning_rate": 3.2162026428305436e-06,
|
301 |
+
"logits/chosen": -2.810059070587158,
|
302 |
+
"logits/rejected": -2.8075289726257324,
|
303 |
+
"logps/chosen": -29.805904388427734,
|
304 |
+
"logps/rejected": -30.83272361755371,
|
305 |
+
"loss": 0.9106,
|
306 |
+
"rewards/accuracies": 0.637499988079071,
|
307 |
+
"rewards/chosen": -0.013605661690235138,
|
308 |
+
"rewards/margins": 0.09284953773021698,
|
309 |
+
"rewards/rejected": -0.10645520687103271,
|
310 |
"step": 180
|
311 |
},
|
312 |
{
|
313 |
"epoch": 0.49,
|
314 |
+
"grad_norm": 6.375,
|
315 |
"learning_rate": 2.996071664294641e-06,
|
316 |
+
"logits/chosen": -2.920551300048828,
|
317 |
+
"logits/rejected": -2.917001247406006,
|
318 |
+
"logps/chosen": -30.380672454833984,
|
319 |
+
"logps/rejected": -29.473743438720703,
|
320 |
+
"loss": 0.9132,
|
321 |
+
"rewards/accuracies": 0.5874999761581421,
|
322 |
+
"rewards/chosen": -0.019395861774683,
|
323 |
+
"rewards/margins": 0.09222523123025894,
|
324 |
+
"rewards/rejected": -0.11162110418081284,
|
325 |
"step": 190
|
326 |
},
|
327 |
{
|
328 |
"epoch": 0.52,
|
329 |
+
"grad_norm": 9.75,
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
+
"logits/chosen": -2.9904088973999023,
|
332 |
+
"logits/rejected": -2.9779367446899414,
|
333 |
+
"logps/chosen": -33.54608917236328,
|
334 |
+
"logps/rejected": -31.4923095703125,
|
335 |
+
"loss": 0.8824,
|
336 |
+
"rewards/accuracies": 0.574999988079071,
|
337 |
+
"rewards/chosen": 0.015293078497052193,
|
338 |
+
"rewards/margins": 0.12111523002386093,
|
339 |
+
"rewards/rejected": -0.10582216084003448,
|
340 |
"step": 200
|
341 |
},
|
342 |
{
|
343 |
"epoch": 0.52,
|
344 |
+
"eval_logits/chosen": -2.8315017223358154,
|
345 |
+
"eval_logits/rejected": -2.8287980556488037,
|
346 |
+
"eval_logps/chosen": -31.827983856201172,
|
347 |
+
"eval_logps/rejected": -35.64292526245117,
|
348 |
+
"eval_loss": 0.9602205753326416,
|
349 |
+
"eval_rewards/accuracies": 0.6092192530632019,
|
350 |
+
"eval_rewards/chosen": -0.05455298721790314,
|
351 |
+
"eval_rewards/margins": 0.039924219250679016,
|
352 |
+
"eval_rewards/rejected": -0.09447719901800156,
|
353 |
+
"eval_runtime": 113.1607,
|
354 |
+
"eval_samples_per_second": 3.031,
|
355 |
"eval_steps_per_second": 0.38,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
+
"grad_norm": 12.125,
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
+
"logits/chosen": -2.9306328296661377,
|
363 |
+
"logits/rejected": -2.9321701526641846,
|
364 |
+
"logps/chosen": -33.18594741821289,
|
365 |
+
"logps/rejected": -35.648868560791016,
|
366 |
+
"loss": 0.8442,
|
367 |
+
"rewards/accuracies": 0.6499999761581421,
|
368 |
+
"rewards/chosen": -0.015253487043082714,
|
369 |
+
"rewards/margins": 0.16348664462566376,
|
370 |
+
"rewards/rejected": -0.17874012887477875,
|
371 |
"step": 210
|
372 |
},
|
373 |
{
|
374 |
"epoch": 0.57,
|
375 |
+
"grad_norm": 10.375,
|
376 |
"learning_rate": 2.3185646976551794e-06,
|
377 |
+
"logits/chosen": -2.9161877632141113,
|
378 |
+
"logits/rejected": -2.9325830936431885,
|
379 |
+
"logps/chosen": -29.98288345336914,
|
380 |
+
"logps/rejected": -30.195465087890625,
|
381 |
+
"loss": 0.818,
|
382 |
+
"rewards/accuracies": 0.7124999761581421,
|
383 |
+
"rewards/chosen": 0.019515613093972206,
|
384 |
+
"rewards/margins": 0.18623510003089905,
|
385 |
+
"rewards/rejected": -0.1667194664478302,
|
386 |
"step": 220
|
387 |
},
|
388 |
{
|
389 |
"epoch": 0.6,
|
390 |
+
"grad_norm": 8.9375,
|
391 |
"learning_rate": 2.0932279108998323e-06,
|
392 |
+
"logits/chosen": -2.9633114337921143,
|
393 |
+
"logits/rejected": -2.9677231311798096,
|
394 |
+
"logps/chosen": -31.465539932250977,
|
395 |
+
"logps/rejected": -33.331993103027344,
|
396 |
+
"loss": 0.8836,
|
397 |
+
"rewards/accuracies": 0.6499999761581421,
|
398 |
+
"rewards/chosen": -0.03136688470840454,
|
399 |
+
"rewards/margins": 0.1260170042514801,
|
400 |
+
"rewards/rejected": -0.15738390386104584,
|
401 |
"step": 230
|
402 |
},
|
403 |
{
|
404 |
"epoch": 0.62,
|
405 |
+
"grad_norm": 10.5625,
|
406 |
"learning_rate": 1.8712423238279358e-06,
|
407 |
+
"logits/chosen": -3.0217947959899902,
|
408 |
+
"logits/rejected": -3.029099702835083,
|
409 |
+
"logps/chosen": -31.11104965209961,
|
410 |
+
"logps/rejected": -31.991535186767578,
|
411 |
+
"loss": 0.845,
|
412 |
+
"rewards/accuracies": 0.7124999761581421,
|
413 |
+
"rewards/chosen": -0.020612526684999466,
|
414 |
+
"rewards/margins": 0.15501661598682404,
|
415 |
+
"rewards/rejected": -0.1756291538476944,
|
416 |
"step": 240
|
417 |
},
|
418 |
{
|
419 |
"epoch": 0.65,
|
420 |
+
"grad_norm": 9.25,
|
421 |
"learning_rate": 1.6544367689701824e-06,
|
422 |
+
"logits/chosen": -2.8557331562042236,
|
423 |
+
"logits/rejected": -2.844738721847534,
|
424 |
+
"logps/chosen": -27.252483367919922,
|
425 |
+
"logps/rejected": -30.79751968383789,
|
426 |
+
"loss": 0.9072,
|
427 |
+
"rewards/accuracies": 0.5874999761581421,
|
428 |
+
"rewards/chosen": -0.029826903715729713,
|
429 |
+
"rewards/margins": 0.0967610701918602,
|
430 |
+
"rewards/rejected": -0.12658795714378357,
|
431 |
"step": 250
|
432 |
},
|
433 |
{
|
434 |
"epoch": 0.68,
|
435 |
+
"grad_norm": 10.875,
|
436 |
"learning_rate": 1.4445974030621963e-06,
|
437 |
+
"logits/chosen": -2.8332571983337402,
|
438 |
+
"logits/rejected": -2.854872226715088,
|
439 |
+
"logps/chosen": -29.75779151916504,
|
440 |
+
"logps/rejected": -36.265663146972656,
|
441 |
+
"loss": 0.7843,
|
442 |
+
"rewards/accuracies": 0.762499988079071,
|
443 |
+
"rewards/chosen": 0.007217098027467728,
|
444 |
+
"rewards/margins": 0.21637487411499023,
|
445 |
+
"rewards/rejected": -0.209157794713974,
|
446 |
"step": 260
|
447 |
},
|
448 |
{
|
449 |
"epoch": 0.7,
|
450 |
+
"grad_norm": 11.6875,
|
451 |
"learning_rate": 1.243452991757889e-06,
|
452 |
+
"logits/chosen": -2.975635051727295,
|
453 |
+
"logits/rejected": -2.9822394847869873,
|
454 |
+
"logps/chosen": -31.34609031677246,
|
455 |
+
"logps/rejected": -32.07633972167969,
|
456 |
+
"loss": 0.8354,
|
457 |
+
"rewards/accuracies": 0.762499988079071,
|
458 |
+
"rewards/chosen": -0.05647352337837219,
|
459 |
+
"rewards/margins": 0.16463439166545868,
|
460 |
+
"rewards/rejected": -0.22110792994499207,
|
461 |
"step": 270
|
462 |
},
|
463 |
{
|
464 |
"epoch": 0.73,
|
465 |
+
"grad_norm": 10.9375,
|
466 |
"learning_rate": 1.0526606671603523e-06,
|
467 |
+
"logits/chosen": -2.9911038875579834,
|
468 |
+
"logits/rejected": -2.9766478538513184,
|
469 |
+
"logps/chosen": -31.343215942382812,
|
470 |
+
"logps/rejected": -30.425678253173828,
|
471 |
+
"loss": 0.8864,
|
472 |
+
"rewards/accuracies": 0.6625000238418579,
|
473 |
+
"rewards/chosen": -0.06215282529592514,
|
474 |
+
"rewards/margins": 0.1172800287604332,
|
475 |
+
"rewards/rejected": -0.17943285405635834,
|
476 |
"step": 280
|
477 |
},
|
478 |
{
|
479 |
"epoch": 0.75,
|
480 |
+
"grad_norm": 13.5,
|
481 |
"learning_rate": 8.737922755071455e-07,
|
482 |
+
"logits/chosen": -2.9202213287353516,
|
483 |
+
"logits/rejected": -2.9021944999694824,
|
484 |
+
"logps/chosen": -32.332275390625,
|
485 |
+
"logps/rejected": -32.9135627746582,
|
486 |
+
"loss": 0.7348,
|
487 |
+
"rewards/accuracies": 0.7749999761581421,
|
488 |
+
"rewards/chosen": 0.030402958393096924,
|
489 |
+
"rewards/margins": 0.26942819356918335,
|
490 |
+
"rewards/rejected": -0.23902523517608643,
|
491 |
"step": 290
|
492 |
},
|
493 |
{
|
494 |
"epoch": 0.78,
|
495 |
+
"grad_norm": 9.0625,
|
496 |
"learning_rate": 7.08321427484816e-07,
|
497 |
+
"logits/chosen": -2.9209041595458984,
|
498 |
+
"logits/rejected": -2.9169058799743652,
|
499 |
+
"logps/chosen": -32.4577522277832,
|
500 |
+
"logps/rejected": -29.638641357421875,
|
501 |
+
"loss": 0.8245,
|
502 |
"rewards/accuracies": 0.6875,
|
503 |
+
"rewards/chosen": -0.02350962720811367,
|
504 |
+
"rewards/margins": 0.17999625205993652,
|
505 |
+
"rewards/rejected": -0.20350590348243713,
|
506 |
"step": 300
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.78,
|
510 |
+
"eval_logits/chosen": -2.8392350673675537,
|
511 |
+
"eval_logits/rejected": -2.836524486541748,
|
512 |
+
"eval_logps/chosen": -32.03562545776367,
|
513 |
+
"eval_logps/rejected": -35.89268112182617,
|
514 |
+
"eval_loss": 0.9559255838394165,
|
515 |
+
"eval_rewards/accuracies": 0.6004983186721802,
|
516 |
+
"eval_rewards/chosen": -0.07531756907701492,
|
517 |
+
"eval_rewards/margins": 0.044134579598903656,
|
518 |
+
"eval_rewards/rejected": -0.11945215612649918,
|
519 |
+
"eval_runtime": 112.8661,
|
520 |
+
"eval_samples_per_second": 3.039,
|
521 |
+
"eval_steps_per_second": 0.381,
|
522 |
"step": 300
|
523 |
},
|
524 |
{
|
525 |
"epoch": 0.81,
|
526 |
+
"grad_norm": 10.1875,
|
527 |
"learning_rate": 5.576113578589035e-07,
|
528 |
+
"logits/chosen": -2.80441951751709,
|
529 |
+
"logits/rejected": -2.8214964866638184,
|
530 |
+
"logps/chosen": -29.70196533203125,
|
531 |
+
"logps/rejected": -33.18954849243164,
|
532 |
+
"loss": 0.8066,
|
533 |
+
"rewards/accuracies": 0.824999988079071,
|
534 |
+
"rewards/chosen": -0.047254521399736404,
|
535 |
+
"rewards/margins": 0.19849951565265656,
|
536 |
+
"rewards/rejected": -0.24575403332710266,
|
537 |
"step": 310
|
538 |
},
|
539 |
{
|
540 |
"epoch": 0.83,
|
541 |
+
"grad_norm": 9.5,
|
542 |
"learning_rate": 4.229036944380913e-07,
|
543 |
+
"logits/chosen": -3.0470097064971924,
|
544 |
+
"logits/rejected": -3.033452033996582,
|
545 |
+
"logps/chosen": -29.982385635375977,
|
546 |
+
"logps/rejected": -30.24712562561035,
|
547 |
+
"loss": 0.8017,
|
548 |
+
"rewards/accuracies": 0.7250000238418579,
|
549 |
+
"rewards/chosen": 0.022920172661542892,
|
550 |
+
"rewards/margins": 0.21575525403022766,
|
551 |
+
"rewards/rejected": -0.19283509254455566,
|
552 |
"step": 320
|
553 |
},
|
554 |
{
|
555 |
"epoch": 0.86,
|
556 |
+
"grad_norm": 11.875,
|
557 |
"learning_rate": 3.053082288996112e-07,
|
558 |
+
"logits/chosen": -2.96113657951355,
|
559 |
+
"logits/rejected": -2.9435229301452637,
|
560 |
+
"logps/chosen": -28.380762100219727,
|
561 |
+
"logps/rejected": -32.93931579589844,
|
562 |
+
"loss": 0.7579,
|
563 |
+
"rewards/accuracies": 0.7250000238418579,
|
564 |
+
"rewards/chosen": -0.0018691420555114746,
|
565 |
+
"rewards/margins": 0.25289779901504517,
|
566 |
+
"rewards/rejected": -0.25476694107055664,
|
567 |
"step": 330
|
568 |
},
|
569 |
{
|
570 |
"epoch": 0.88,
|
571 |
+
"grad_norm": 10.5625,
|
572 |
"learning_rate": 2.0579377374915805e-07,
|
573 |
+
"logits/chosen": -3.182105541229248,
|
574 |
+
"logits/rejected": -3.1877188682556152,
|
575 |
+
"logps/chosen": -31.98989486694336,
|
576 |
+
"logps/rejected": -35.27183532714844,
|
577 |
+
"loss": 0.7995,
|
578 |
+
"rewards/accuracies": 0.6875,
|
579 |
+
"rewards/chosen": -0.035068999975919724,
|
580 |
+
"rewards/margins": 0.22251394391059875,
|
581 |
+
"rewards/rejected": -0.2575829327106476,
|
582 |
"step": 340
|
583 |
},
|
584 |
{
|
585 |
"epoch": 0.91,
|
586 |
+
"grad_norm": 11.75,
|
587 |
"learning_rate": 1.2518018074041684e-07,
|
588 |
+
"logits/chosen": -3.057163953781128,
|
589 |
+
"logits/rejected": -3.0605666637420654,
|
590 |
+
"logps/chosen": -30.571701049804688,
|
591 |
+
"logps/rejected": -33.298683166503906,
|
592 |
+
"loss": 0.8066,
|
593 |
"rewards/accuracies": 0.762499988079071,
|
594 |
+
"rewards/chosen": 0.023518025875091553,
|
595 |
+
"rewards/margins": 0.20205526053905487,
|
596 |
+
"rewards/rejected": -0.1785372495651245,
|
597 |
"step": 350
|
598 |
},
|
599 |
{
|
600 |
"epoch": 0.94,
|
601 |
+
"grad_norm": 11.0,
|
602 |
"learning_rate": 6.41315865106129e-08,
|
603 |
+
"logits/chosen": -2.8954837322235107,
|
604 |
+
"logits/rejected": -2.8972554206848145,
|
605 |
+
"logps/chosen": -28.211322784423828,
|
606 |
+
"logps/rejected": -31.518001556396484,
|
607 |
+
"loss": 0.7939,
|
608 |
+
"rewards/accuracies": 0.7124999761581421,
|
609 |
+
"rewards/chosen": 0.04384060949087143,
|
610 |
+
"rewards/margins": 0.21254661679267883,
|
611 |
+
"rewards/rejected": -0.1687059849500656,
|
612 |
"step": 360
|
613 |
},
|
614 |
{
|
615 |
"epoch": 0.96,
|
616 |
+
"grad_norm": 10.3125,
|
617 |
"learning_rate": 2.3150941078050325e-08,
|
618 |
+
"logits/chosen": -2.973878860473633,
|
619 |
+
"logits/rejected": -2.9718644618988037,
|
620 |
+
"logps/chosen": -30.932153701782227,
|
621 |
+
"logps/rejected": -33.92184829711914,
|
622 |
+
"loss": 0.8202,
|
623 |
+
"rewards/accuracies": 0.6875,
|
624 |
+
"rewards/chosen": -0.014550352469086647,
|
625 |
+
"rewards/margins": 0.1848299205303192,
|
626 |
+
"rewards/rejected": -0.1993802934885025,
|
627 |
"step": 370
|
628 |
},
|
629 |
{
|
630 |
"epoch": 0.99,
|
631 |
+
"grad_norm": 9.8125,
|
632 |
"learning_rate": 2.575864278703266e-09,
|
633 |
+
"logits/chosen": -2.927494764328003,
|
634 |
+
"logits/rejected": -2.9113755226135254,
|
635 |
+
"logps/chosen": -29.455524444580078,
|
636 |
+
"logps/rejected": -30.044204711914062,
|
637 |
+
"loss": 0.8384,
|
638 |
+
"rewards/accuracies": 0.7250000238418579,
|
639 |
+
"rewards/chosen": -0.05104842036962509,
|
640 |
+
"rewards/margins": 0.16441020369529724,
|
641 |
+
"rewards/rejected": -0.21545863151550293,
|
642 |
"step": 380
|
643 |
},
|
644 |
{
|
645 |
"epoch": 1.0,
|
646 |
"step": 385,
|
647 |
"total_flos": 0.0,
|
648 |
+
"train_loss": 0.8805358874333369,
|
649 |
+
"train_runtime": 2723.8224,
|
650 |
+
"train_samples_per_second": 1.13,
|
651 |
+
"train_steps_per_second": 0.141
|
652 |
}
|
653 |
],
|
654 |
"logging_steps": 10,
|