hugodk-sch
commited on
Commit
•
7a1426b
1
Parent(s):
605ee93
Model save
Browse files- README.md +14 -16
- adapter_model.safetensors +1 -1
- all_results.json +2 -15
- train_results.json +2 -2
- trainer_state.json +415 -415
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:
|
24 |
-
- Rewards/chosen: 0.
|
25 |
-
- Rewards/rejected: 0.
|
26 |
-
- Rewards/accuracies: 0.
|
27 |
-
- Rewards/margins:
|
28 |
-
- Logps/rejected: -
|
29 |
-
- Logps/chosen: -32.
|
30 |
-
- Logits/rejected: 98.
|
31 |
-
- Logits/chosen:
|
32 |
|
33 |
## Model description
|
34 |
|
@@ -63,9 +61,9 @@ The following hyperparameters were used during training:
|
|
63 |
|
64 |
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
65 |
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
66 |
-
|
|
67 |
-
|
|
68 |
-
|
|
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: 7.6485
|
22 |
+
- Rewards/chosen: 0.0015
|
23 |
+
- Rewards/rejected: -0.0116
|
24 |
+
- Rewards/accuracies: 0.5390
|
25 |
+
- Rewards/margins: 0.0131
|
26 |
+
- Logps/rejected: -36.0246
|
27 |
+
- Logps/chosen: -32.4356
|
28 |
+
- Logits/rejected: 98.9867
|
29 |
+
- Logits/chosen: 99.0023
|
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 |
+
| 6.7218 | 0.26 | 100 | 7.5705 | -0.0028 | -0.0151 | 0.5083 | 0.0123 | -36.0421 | -32.4573 | 98.6549 | 98.6709 |
|
65 |
+
| 6.2997 | 0.52 | 200 | 7.1457 | -0.0201 | -0.0563 | 0.5403 | 0.0361 | -36.2478 | -32.5438 | 98.9065 | 98.9208 |
|
66 |
+
| 6.0651 | 0.78 | 300 | 7.6485 | 0.0015 | -0.0116 | 0.5390 | 0.0131 | -36.0246 | -32.4356 | 98.9867 | 99.0023 |
|
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 sha256:8e50138c59173f57650f9d0ad3b14bfcb03b10ed6b1305895e7aa0ced25a3fe7
|
3 |
size 671150064
|
all_results.json
CHANGED
@@ -1,20 +1,7 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"
|
4 |
-
"
|
5 |
-
"eval_logps/chosen": -32.3518180847168,
|
6 |
-
"eval_logps/rejected": -35.857688903808594,
|
7 |
-
"eval_loss": 2.3100497722625732,
|
8 |
-
"eval_rewards/accuracies": 0.5215947031974792,
|
9 |
-
"eval_rewards/chosen": 0.045678745955228806,
|
10 |
-
"eval_rewards/margins": -0.00873997900635004,
|
11 |
-
"eval_rewards/rejected": 0.05441872030496597,
|
12 |
-
"eval_runtime": 103.7492,
|
13 |
-
"eval_samples": 343,
|
14 |
-
"eval_samples_per_second": 3.306,
|
15 |
-
"eval_steps_per_second": 0.414,
|
16 |
-
"train_loss": 2.572123973400562,
|
17 |
-
"train_runtime": 2556.3668,
|
18 |
"train_samples": 3079,
|
19 |
"train_samples_per_second": 1.204,
|
20 |
"train_steps_per_second": 0.151
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"train_loss": 5.985443219271573,
|
4 |
+
"train_runtime": 2557.1257,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.204,
|
7 |
"train_steps_per_second": 0.151
|
train_results.json
CHANGED
@@ -1,7 +1,7 @@
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
-
"train_loss":
|
4 |
-
"train_runtime":
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.204,
|
7 |
"train_steps_per_second": 0.151
|
|
|
1 |
{
|
2 |
"epoch": 1.0,
|
3 |
+
"train_loss": 5.985443219271573,
|
4 |
+
"train_runtime": 2557.1257,
|
5 |
"train_samples": 3079,
|
6 |
"train_samples_per_second": 1.204,
|
7 |
"train_steps_per_second": 0.151
|
trainer_state.json
CHANGED
@@ -10,13 +10,13 @@
|
|
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,
|
17 |
"logps/chosen": -29.073104858398438,
|
18 |
"logps/rejected": -26.25731658935547,
|
19 |
-
"loss":
|
20 |
"rewards/accuracies": 0.0,
|
21 |
"rewards/chosen": 0.0,
|
22 |
"rewards/margins": 0.0,
|
@@ -25,628 +25,628 @@
|
|
25 |
},
|
26 |
{
|
27 |
"epoch": 0.03,
|
28 |
-
"grad_norm":
|
29 |
"learning_rate": 1.282051282051282e-06,
|
30 |
-
"logits/chosen": 81.
|
31 |
-
"logits/rejected": 80.
|
32 |
-
"logps/chosen": -34.
|
33 |
-
"logps/rejected": -
|
34 |
-
"loss":
|
35 |
-
"rewards/accuracies": 0.
|
36 |
-
"rewards/chosen": 0.
|
37 |
-
"rewards/margins": 0.
|
38 |
-
"rewards/rejected": -0.
|
39 |
"step": 10
|
40 |
},
|
41 |
{
|
42 |
"epoch": 0.05,
|
43 |
-
"grad_norm":
|
44 |
"learning_rate": 2.564102564102564e-06,
|
45 |
-
"logits/chosen": 80.
|
46 |
-
"logits/rejected": 80.
|
47 |
-
"logps/chosen": -33.
|
48 |
-
"logps/rejected": -30.
|
49 |
-
"loss":
|
50 |
-
"rewards/accuracies": 0.
|
51 |
-
"rewards/chosen": 0.
|
52 |
-
"rewards/margins": 0.
|
53 |
-
"rewards/rejected": 0.
|
54 |
"step": 20
|
55 |
},
|
56 |
{
|
57 |
"epoch": 0.08,
|
58 |
-
"grad_norm":
|
59 |
"learning_rate": 3.846153846153847e-06,
|
60 |
-
"logits/chosen": 82.
|
61 |
-
"logits/rejected": 82.
|
62 |
-
"logps/chosen": -33.
|
63 |
-
"logps/rejected": -31.
|
64 |
-
"loss":
|
65 |
-
"rewards/accuracies": 0.
|
66 |
-
"rewards/chosen": 0.
|
67 |
-
"rewards/margins":
|
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": 80.
|
76 |
-
"logits/rejected": 80.
|
77 |
-
"logps/chosen": -
|
78 |
-
"logps/rejected": -33.
|
79 |
-
"loss":
|
80 |
"rewards/accuracies": 0.5375000238418579,
|
81 |
-
"rewards/chosen": 0.
|
82 |
-
"rewards/margins": 0.
|
83 |
-
"rewards/rejected":
|
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": -
|
93 |
-
"logps/rejected": -31.
|
94 |
-
"loss":
|
95 |
-
"rewards/accuracies": 0.
|
96 |
-
"rewards/chosen":
|
97 |
-
"rewards/margins":
|
98 |
-
"rewards/rejected":
|
99 |
"step": 50
|
100 |
},
|
101 |
{
|
102 |
"epoch": 0.16,
|
103 |
-
"grad_norm":
|
104 |
"learning_rate": 4.954691471941119e-06,
|
105 |
-
"logits/chosen": 82.
|
106 |
-
"logits/rejected": 82.
|
107 |
-
"logps/chosen": -31.
|
108 |
-
"logps/rejected": -29.
|
109 |
-
"loss":
|
110 |
-
"rewards/accuracies": 0.
|
111 |
-
"rewards/chosen":
|
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": 83.
|
121 |
-
"logits/rejected": 83.
|
122 |
-
"logps/chosen": -30.
|
123 |
-
"logps/rejected": -
|
124 |
-
"loss":
|
125 |
-
"rewards/accuracies": 0.
|
126 |
-
"rewards/chosen": -0.
|
127 |
-
"rewards/margins":
|
128 |
-
"rewards/rejected": 0.
|
129 |
"step": 70
|
130 |
},
|
131 |
{
|
132 |
"epoch": 0.21,
|
133 |
-
"grad_norm":
|
134 |
"learning_rate": 4.828760511501322e-06,
|
135 |
-
"logits/chosen": 81.
|
136 |
-
"logits/rejected": 81.
|
137 |
-
"logps/chosen": -31.
|
138 |
-
"logps/rejected": -30.
|
139 |
-
"loss":
|
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":
|
155 |
-
"rewards/accuracies": 0.
|
156 |
-
"rewards/chosen": 0.
|
157 |
-
"rewards/margins": 0.
|
158 |
-
"rewards/rejected": -0.
|
159 |
"step": 90
|
160 |
},
|
161 |
{
|
162 |
"epoch": 0.26,
|
163 |
-
"grad_norm":
|
164 |
"learning_rate": 4.626245458345211e-06,
|
165 |
-
"logits/chosen": 83.
|
166 |
-
"logits/rejected": 83.
|
167 |
-
"logps/chosen": -34.
|
168 |
-
"logps/rejected": -31.
|
169 |
-
"loss":
|
170 |
"rewards/accuracies": 0.5625,
|
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":
|
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": 0.413,
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
-
"grad_norm":
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
-
"logits/chosen": 83.
|
197 |
-
"logits/rejected": 83.
|
198 |
-
"logps/chosen": -32.
|
199 |
-
"logps/rejected": -32.
|
200 |
-
"loss":
|
201 |
-
"rewards/accuracies": 0.
|
202 |
-
"rewards/chosen": 0.
|
203 |
-
"rewards/margins": 0.
|
204 |
-
"rewards/rejected": -0.
|
205 |
"step": 110
|
206 |
},
|
207 |
{
|
208 |
"epoch": 0.31,
|
209 |
-
"grad_norm":
|
210 |
"learning_rate": 4.353806263777678e-06,
|
211 |
-
"logits/chosen":
|
212 |
-
"logits/rejected":
|
213 |
-
"logps/chosen": -28.
|
214 |
-
"logps/rejected": -35.
|
215 |
-
"loss":
|
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":
|
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": -27.
|
244 |
-
"logps/rejected": -
|
245 |
-
"loss":
|
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": 81.
|
257 |
-
"logits/rejected": 81.
|
258 |
-
"logps/chosen": -
|
259 |
-
"logps/rejected": -32.
|
260 |
-
"loss":
|
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": 83.
|
272 |
-
"logits/rejected": 83.
|
273 |
-
"logps/chosen": -33.
|
274 |
-
"logps/rejected": -
|
275 |
-
"loss":
|
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": 84.
|
287 |
-
"logits/rejected":
|
288 |
-
"logps/chosen": -30.
|
289 |
-
"logps/rejected": -32.
|
290 |
-
"loss":
|
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": 81.
|
302 |
-
"logits/rejected": 81.
|
303 |
-
"logps/chosen": -30.
|
304 |
-
"logps/rejected": -31.
|
305 |
-
"loss":
|
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": 83.
|
317 |
-
"logits/rejected": 83.
|
318 |
-
"logps/chosen": -30.
|
319 |
-
"logps/rejected": -30.
|
320 |
-
"loss":
|
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": 79.
|
332 |
-
"logits/rejected": 79.
|
333 |
-
"logps/chosen": -34.
|
334 |
-
"logps/rejected": -32.
|
335 |
-
"loss":
|
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":
|
349 |
-
"eval_rewards/accuracies": 0.
|
350 |
-
"eval_rewards/chosen": -0.
|
351 |
-
"eval_rewards/margins": 0.
|
352 |
-
"eval_rewards/rejected": -0.
|
353 |
-
"eval_runtime": 104.
|
354 |
-
"eval_samples_per_second": 3.
|
355 |
"eval_steps_per_second": 0.413,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
-
"grad_norm":
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
-
"logits/chosen": 81.
|
363 |
-
"logits/rejected": 81.
|
364 |
-
"logps/chosen": -33.
|
365 |
-
"logps/rejected": -35.
|
366 |
-
"loss":
|
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": -30.
|
381 |
-
"loss":
|
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": 81.
|
393 |
-
"logits/rejected": 81.
|
394 |
-
"logps/chosen": -32.
|
395 |
-
"logps/rejected": -
|
396 |
-
"loss":
|
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": 83.
|
408 |
-
"logits/rejected":
|
409 |
-
"logps/chosen": -
|
410 |
-
"logps/rejected": -31.
|
411 |
-
"loss":
|
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": 82.
|
423 |
-
"logits/rejected": 82.
|
424 |
-
"logps/chosen": -
|
425 |
-
"logps/rejected": -29.
|
426 |
-
"loss":
|
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": 79.
|
438 |
-
"logits/rejected":
|
439 |
-
"logps/chosen": -30.
|
440 |
-
"logps/rejected": -36.
|
441 |
-
"loss":
|
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": 79.
|
453 |
-
"logits/rejected": 79.
|
454 |
-
"logps/chosen": -30.
|
455 |
-
"logps/rejected": -31.
|
456 |
-
"loss":
|
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": 81.
|
468 |
-
"logits/rejected": 81.
|
469 |
-
"logps/chosen": -31.
|
470 |
-
"logps/rejected": -29.
|
471 |
-
"loss":
|
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":
|
483 |
-
"logits/rejected":
|
484 |
-
"logps/chosen": -33.
|
485 |
-
"logps/rejected": -
|
486 |
-
"loss":
|
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": 77.
|
498 |
-
"logits/rejected": 77.
|
499 |
-
"logps/chosen": -32.
|
500 |
-
"logps/rejected": -
|
501 |
-
"loss":
|
502 |
-
"rewards/accuracies": 0.
|
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": 99.
|
511 |
-
"eval_logits/rejected": 98.
|
512 |
-
"eval_logps/chosen": -32.
|
513 |
-
"eval_logps/rejected": -
|
514 |
-
"eval_loss":
|
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": 84.
|
529 |
-
"logits/rejected": 84.
|
530 |
-
"logps/chosen": -
|
531 |
-
"logps/rejected": -32.
|
532 |
-
"loss":
|
533 |
-
"rewards/accuracies": 0.
|
534 |
-
"rewards/chosen": 0.
|
535 |
-
"rewards/margins": 0.
|
536 |
-
"rewards/rejected": 0.
|
537 |
"step": 310
|
538 |
},
|
539 |
{
|
540 |
"epoch": 0.83,
|
541 |
-
"grad_norm":
|
542 |
"learning_rate": 4.229036944380913e-07,
|
543 |
-
"logits/chosen":
|
544 |
-
"logits/rejected":
|
545 |
-
"logps/chosen": -30.
|
546 |
-
"logps/rejected": -28.
|
547 |
-
"loss":
|
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": 79.
|
559 |
-
"logits/rejected": 79.
|
560 |
-
"logps/chosen": -29.
|
561 |
-
"logps/rejected": -32.
|
562 |
-
"loss":
|
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": 83.
|
574 |
-
"logits/rejected": 83.
|
575 |
-
"logps/chosen": -32.
|
576 |
-
"logps/rejected": -33.
|
577 |
-
"loss":
|
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": 82.
|
589 |
-
"logits/rejected": 82.
|
590 |
-
"logps/chosen": -32.
|
591 |
-
"logps/rejected": -33.
|
592 |
-
"loss":
|
593 |
-
"rewards/accuracies": 0.
|
594 |
-
"rewards/chosen": 0.
|
595 |
-
"rewards/margins": 0.
|
596 |
-
"rewards/rejected": 0.
|
597 |
"step": 350
|
598 |
},
|
599 |
{
|
600 |
"epoch": 0.94,
|
601 |
-
"grad_norm":
|
602 |
"learning_rate": 6.41315865106129e-08,
|
603 |
-
"logits/chosen":
|
604 |
-
"logits/rejected":
|
605 |
-
"logps/chosen": -28.
|
606 |
-
"logps/rejected": -31.
|
607 |
-
"loss":
|
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": 83.
|
619 |
-
"logits/rejected": 83.
|
620 |
-
"logps/chosen": -
|
621 |
-
"logps/rejected": -35.
|
622 |
-
"loss":
|
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": 77.
|
634 |
-
"logits/rejected": 77.
|
635 |
-
"logps/chosen": -
|
636 |
-
"logps/rejected": -28.
|
637 |
-
"loss":
|
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":
|
649 |
-
"train_runtime":
|
650 |
"train_samples_per_second": 1.204,
|
651 |
"train_steps_per_second": 0.151
|
652 |
}
|
|
|
10 |
"log_history": [
|
11 |
{
|
12 |
"epoch": 0.0,
|
13 |
+
"grad_norm": 656.0,
|
14 |
"learning_rate": 1.282051282051282e-07,
|
15 |
"logits/chosen": 88.18099975585938,
|
16 |
"logits/rejected": 88.25153350830078,
|
17 |
"logps/chosen": -29.073104858398438,
|
18 |
"logps/rejected": -26.25731658935547,
|
19 |
+
"loss": 6.25,
|
20 |
"rewards/accuracies": 0.0,
|
21 |
"rewards/chosen": 0.0,
|
22 |
"rewards/margins": 0.0,
|
|
|
25 |
},
|
26 |
{
|
27 |
"epoch": 0.03,
|
28 |
+
"grad_norm": 536.0,
|
29 |
"learning_rate": 1.282051282051282e-06,
|
30 |
+
"logits/chosen": 81.07795715332031,
|
31 |
+
"logits/rejected": 80.78132629394531,
|
32 |
+
"logps/chosen": -34.14677047729492,
|
33 |
+
"logps/rejected": -33.015296936035156,
|
34 |
+
"loss": 6.1804,
|
35 |
+
"rewards/accuracies": 0.5277777910232544,
|
36 |
+
"rewards/chosen": 0.0189919825643301,
|
37 |
+
"rewards/margins": 0.030369587242603302,
|
38 |
+
"rewards/rejected": -0.011377605609595776,
|
39 |
"step": 10
|
40 |
},
|
41 |
{
|
42 |
"epoch": 0.05,
|
43 |
+
"grad_norm": 592.0,
|
44 |
"learning_rate": 2.564102564102564e-06,
|
45 |
+
"logits/chosen": 80.67132568359375,
|
46 |
+
"logits/rejected": 80.56291198730469,
|
47 |
+
"logps/chosen": -33.51841354370117,
|
48 |
+
"logps/rejected": -30.76877212524414,
|
49 |
+
"loss": 6.6549,
|
50 |
+
"rewards/accuracies": 0.574999988079071,
|
51 |
+
"rewards/chosen": 0.034299030900001526,
|
52 |
+
"rewards/margins": 0.03255577012896538,
|
53 |
+
"rewards/rejected": 0.0017432533204555511,
|
54 |
"step": 20
|
55 |
},
|
56 |
{
|
57 |
"epoch": 0.08,
|
58 |
+
"grad_norm": 448.0,
|
59 |
"learning_rate": 3.846153846153847e-06,
|
60 |
+
"logits/chosen": 82.46919250488281,
|
61 |
+
"logits/rejected": 82.5018539428711,
|
62 |
+
"logps/chosen": -33.83555603027344,
|
63 |
+
"logps/rejected": -31.259449005126953,
|
64 |
+
"loss": 7.4075,
|
65 |
+
"rewards/accuracies": 0.44999998807907104,
|
66 |
+
"rewards/chosen": 0.03545312210917473,
|
67 |
+
"rewards/margins": 0.0025764047168195248,
|
68 |
+
"rewards/rejected": 0.03287670761346817,
|
69 |
"step": 30
|
70 |
},
|
71 |
{
|
72 |
"epoch": 0.1,
|
73 |
+
"grad_norm": 612.0,
|
74 |
"learning_rate": 4.999896948438434e-06,
|
75 |
+
"logits/chosen": 80.91734313964844,
|
76 |
+
"logits/rejected": 80.9122314453125,
|
77 |
+
"logps/chosen": -32.870399475097656,
|
78 |
+
"logps/rejected": -33.192291259765625,
|
79 |
+
"loss": 6.9708,
|
80 |
"rewards/accuracies": 0.5375000238418579,
|
81 |
+
"rewards/chosen": 0.06073303893208504,
|
82 |
+
"rewards/margins": 0.025485849007964134,
|
83 |
+
"rewards/rejected": 0.03524719178676605,
|
84 |
"step": 40
|
85 |
},
|
86 |
{
|
87 |
"epoch": 0.13,
|
88 |
+
"grad_norm": 528.0,
|
89 |
"learning_rate": 4.987541037542187e-06,
|
90 |
+
"logits/chosen": 78.42672729492188,
|
91 |
+
"logits/rejected": 78.44120788574219,
|
92 |
+
"logps/chosen": -30.90542984008789,
|
93 |
+
"logps/rejected": -31.027896881103516,
|
94 |
+
"loss": 6.7099,
|
95 |
+
"rewards/accuracies": 0.550000011920929,
|
96 |
+
"rewards/chosen": 0.059963710606098175,
|
97 |
+
"rewards/margins": 0.05088571831583977,
|
98 |
+
"rewards/rejected": 0.009077982977032661,
|
99 |
"step": 50
|
100 |
},
|
101 |
{
|
102 |
"epoch": 0.16,
|
103 |
+
"grad_norm": 498.0,
|
104 |
"learning_rate": 4.954691471941119e-06,
|
105 |
+
"logits/chosen": 82.86418151855469,
|
106 |
+
"logits/rejected": 82.91536712646484,
|
107 |
+
"logps/chosen": -31.078943252563477,
|
108 |
+
"logps/rejected": -29.606725692749023,
|
109 |
+
"loss": 7.0829,
|
110 |
+
"rewards/accuracies": 0.4749999940395355,
|
111 |
+
"rewards/chosen": 0.0192436333745718,
|
112 |
+
"rewards/margins": 0.022484425455331802,
|
113 |
+
"rewards/rejected": -0.0032407897524535656,
|
114 |
"step": 60
|
115 |
},
|
116 |
{
|
117 |
"epoch": 0.18,
|
118 |
+
"grad_norm": 960.0,
|
119 |
"learning_rate": 4.901618883413549e-06,
|
120 |
+
"logits/chosen": 83.54774475097656,
|
121 |
+
"logits/rejected": 83.57620239257812,
|
122 |
+
"logps/chosen": -30.6588077545166,
|
123 |
+
"logps/rejected": -33.22206115722656,
|
124 |
+
"loss": 6.976,
|
125 |
+
"rewards/accuracies": 0.5874999761581421,
|
126 |
+
"rewards/chosen": -0.0059564015828073025,
|
127 |
+
"rewards/margins": 0.0309152714908123,
|
128 |
+
"rewards/rejected": -0.03687167167663574,
|
129 |
"step": 70
|
130 |
},
|
131 |
{
|
132 |
"epoch": 0.21,
|
133 |
+
"grad_norm": 556.0,
|
134 |
"learning_rate": 4.828760511501322e-06,
|
135 |
+
"logits/chosen": 81.12471771240234,
|
136 |
+
"logits/rejected": 81.11027526855469,
|
137 |
+
"logps/chosen": -31.388225555419922,
|
138 |
+
"logps/rejected": -30.88912010192871,
|
139 |
+
"loss": 6.8226,
|
140 |
+
"rewards/accuracies": 0.5375000238418579,
|
141 |
+
"rewards/chosen": 0.023343011736869812,
|
142 |
+
"rewards/margins": 0.049637455493211746,
|
143 |
+
"rewards/rejected": -0.026294440031051636,
|
144 |
"step": 80
|
145 |
},
|
146 |
{
|
147 |
"epoch": 0.23,
|
148 |
+
"grad_norm": 680.0,
|
149 |
"learning_rate": 4.7367166013034295e-06,
|
150 |
+
"logits/chosen": 78.06856536865234,
|
151 |
+
"logits/rejected": 78.03953552246094,
|
152 |
+
"logps/chosen": -32.45366668701172,
|
153 |
+
"logps/rejected": -31.23443603515625,
|
154 |
+
"loss": 6.1663,
|
155 |
+
"rewards/accuracies": 0.6499999761581421,
|
156 |
+
"rewards/chosen": 0.03690256550908089,
|
157 |
+
"rewards/margins": 0.09349562227725983,
|
158 |
+
"rewards/rejected": -0.056593067944049835,
|
159 |
"step": 90
|
160 |
},
|
161 |
{
|
162 |
"epoch": 0.26,
|
163 |
+
"grad_norm": 396.0,
|
164 |
"learning_rate": 4.626245458345211e-06,
|
165 |
+
"logits/chosen": 83.3121566772461,
|
166 |
+
"logits/rejected": 83.32379913330078,
|
167 |
+
"logps/chosen": -34.064453125,
|
168 |
+
"logps/rejected": -31.79021644592285,
|
169 |
+
"loss": 6.7218,
|
170 |
"rewards/accuracies": 0.5625,
|
171 |
+
"rewards/chosen": 0.04810718446969986,
|
172 |
+
"rewards/margins": 0.05604150891304016,
|
173 |
+
"rewards/rejected": -0.00793431792408228,
|
174 |
"step": 100
|
175 |
},
|
176 |
{
|
177 |
"epoch": 0.26,
|
178 |
+
"eval_logits/chosen": 98.67085266113281,
|
179 |
+
"eval_logits/rejected": 98.65489959716797,
|
180 |
+
"eval_logps/chosen": -32.457313537597656,
|
181 |
+
"eval_logps/rejected": -36.04214859008789,
|
182 |
+
"eval_loss": 7.570494651794434,
|
183 |
+
"eval_rewards/accuracies": 0.5083056688308716,
|
184 |
+
"eval_rewards/chosen": -0.002827130490913987,
|
185 |
+
"eval_rewards/margins": 0.0122976154088974,
|
186 |
+
"eval_rewards/rejected": -0.015124746598303318,
|
187 |
+
"eval_runtime": 104.1006,
|
188 |
+
"eval_samples_per_second": 3.295,
|
189 |
"eval_steps_per_second": 0.413,
|
190 |
"step": 100
|
191 |
},
|
192 |
{
|
193 |
"epoch": 0.29,
|
194 |
+
"grad_norm": 752.0,
|
195 |
"learning_rate": 4.498257201263691e-06,
|
196 |
+
"logits/chosen": 83.45378112792969,
|
197 |
+
"logits/rejected": 83.33932495117188,
|
198 |
+
"logps/chosen": -32.50124740600586,
|
199 |
+
"logps/rejected": -32.954078674316406,
|
200 |
+
"loss": 5.6892,
|
201 |
+
"rewards/accuracies": 0.699999988079071,
|
202 |
+
"rewards/chosen": 0.104311004281044,
|
203 |
+
"rewards/margins": 0.16394877433776855,
|
204 |
+
"rewards/rejected": -0.05963778495788574,
|
205 |
"step": 110
|
206 |
},
|
207 |
{
|
208 |
"epoch": 0.31,
|
209 |
+
"grad_norm": 620.0,
|
210 |
"learning_rate": 4.353806263777678e-06,
|
211 |
+
"logits/chosen": 83.64091491699219,
|
212 |
+
"logits/rejected": 83.74860382080078,
|
213 |
+
"logps/chosen": -28.32021141052246,
|
214 |
+
"logps/rejected": -35.66899871826172,
|
215 |
+
"loss": 5.3716,
|
216 |
+
"rewards/accuracies": 0.737500011920929,
|
217 |
+
"rewards/chosen": 0.12187393009662628,
|
218 |
+
"rewards/margins": 0.16384465992450714,
|
219 |
+
"rewards/rejected": -0.04197072237730026,
|
220 |
"step": 120
|
221 |
},
|
222 |
{
|
223 |
"epoch": 0.34,
|
224 |
+
"grad_norm": 484.0,
|
225 |
"learning_rate": 4.1940827077152755e-06,
|
226 |
+
"logits/chosen": 80.95816040039062,
|
227 |
+
"logits/rejected": 80.97350311279297,
|
228 |
+
"logps/chosen": -30.554763793945312,
|
229 |
+
"logps/rejected": -32.3652458190918,
|
230 |
+
"loss": 6.2467,
|
231 |
+
"rewards/accuracies": 0.637499988079071,
|
232 |
+
"rewards/chosen": 0.07051900774240494,
|
233 |
+
"rewards/margins": 0.15724298357963562,
|
234 |
+
"rewards/rejected": -0.08672399073839188,
|
235 |
"step": 130
|
236 |
},
|
237 |
{
|
238 |
"epoch": 0.36,
|
239 |
+
"grad_norm": 428.0,
|
240 |
"learning_rate": 4.0204024186666215e-06,
|
241 |
+
"logits/chosen": 82.39835357666016,
|
242 |
+
"logits/rejected": 82.39920806884766,
|
243 |
+
"logps/chosen": -27.08688735961914,
|
244 |
+
"logps/rejected": -33.00067901611328,
|
245 |
+
"loss": 5.3964,
|
246 |
+
"rewards/accuracies": 0.6875,
|
247 |
+
"rewards/chosen": 0.07208801805973053,
|
248 |
+
"rewards/margins": 0.18754826486110687,
|
249 |
+
"rewards/rejected": -0.11546025425195694,
|
250 |
"step": 140
|
251 |
},
|
252 |
{
|
253 |
"epoch": 0.39,
|
254 |
+
"grad_norm": 482.0,
|
255 |
"learning_rate": 3.834196265035119e-06,
|
256 |
+
"logits/chosen": 81.19883728027344,
|
257 |
+
"logits/rejected": 81.17581939697266,
|
258 |
+
"logps/chosen": -28.9885311126709,
|
259 |
+
"logps/rejected": -32.907562255859375,
|
260 |
+
"loss": 5.3313,
|
261 |
+
"rewards/accuracies": 0.7250000238418579,
|
262 |
+
"rewards/chosen": 0.08011577278375626,
|
263 |
+
"rewards/margins": 0.14832790195941925,
|
264 |
+
"rewards/rejected": -0.06821214407682419,
|
265 |
"step": 150
|
266 |
},
|
267 |
{
|
268 |
"epoch": 0.42,
|
269 |
+
"grad_norm": 1040.0,
|
270 |
"learning_rate": 3.636998309800573e-06,
|
271 |
+
"logits/chosen": 83.12278747558594,
|
272 |
+
"logits/rejected": 83.13607025146484,
|
273 |
+
"logps/chosen": -33.499732971191406,
|
274 |
+
"logps/rejected": -30.342330932617188,
|
275 |
+
"loss": 6.478,
|
276 |
+
"rewards/accuracies": 0.699999988079071,
|
277 |
+
"rewards/chosen": 0.1274038851261139,
|
278 |
+
"rewards/margins": 0.2053629606962204,
|
279 |
+
"rewards/rejected": -0.07795912027359009,
|
280 |
"step": 160
|
281 |
},
|
282 |
{
|
283 |
"epoch": 0.44,
|
284 |
+
"grad_norm": 684.0,
|
285 |
"learning_rate": 3.4304331721118078e-06,
|
286 |
+
"logits/chosen": 84.02305603027344,
|
287 |
+
"logits/rejected": 83.95697784423828,
|
288 |
+
"logps/chosen": -30.76401710510254,
|
289 |
+
"logps/rejected": -32.3089599609375,
|
290 |
+
"loss": 5.8957,
|
291 |
+
"rewards/accuracies": 0.6625000238418579,
|
292 |
+
"rewards/chosen": 0.09905388206243515,
|
293 |
+
"rewards/margins": 0.1568334400653839,
|
294 |
+
"rewards/rejected": -0.057779550552368164,
|
295 |
"step": 170
|
296 |
},
|
297 |
{
|
298 |
"epoch": 0.47,
|
299 |
+
"grad_norm": 498.0,
|
300 |
"learning_rate": 3.2162026428305436e-06,
|
301 |
+
"logits/chosen": 81.92304992675781,
|
302 |
+
"logits/rejected": 81.90460205078125,
|
303 |
+
"logps/chosen": -30.485610961914062,
|
304 |
+
"logps/rejected": -31.443450927734375,
|
305 |
+
"loss": 5.0645,
|
306 |
+
"rewards/accuracies": 0.6875,
|
307 |
+
"rewards/chosen": 0.1423078328371048,
|
308 |
+
"rewards/margins": 0.19955609738826752,
|
309 |
+
"rewards/rejected": -0.05724828690290451,
|
310 |
"step": 180
|
311 |
},
|
312 |
{
|
313 |
"epoch": 0.49,
|
314 |
+
"grad_norm": 502.0,
|
315 |
"learning_rate": 2.996071664294641e-06,
|
316 |
+
"logits/chosen": 83.73831176757812,
|
317 |
+
"logits/rejected": 83.71122741699219,
|
318 |
+
"logps/chosen": -30.49334716796875,
|
319 |
+
"logps/rejected": -30.74527931213379,
|
320 |
+
"loss": 7.3479,
|
321 |
+
"rewards/accuracies": 0.625,
|
322 |
+
"rewards/chosen": 0.05461405590176582,
|
323 |
+
"rewards/margins": 0.11006426811218262,
|
324 |
+
"rewards/rejected": -0.055450212210416794,
|
325 |
"step": 190
|
326 |
},
|
327 |
{
|
328 |
"epoch": 0.52,
|
329 |
+
"grad_norm": 294.0,
|
330 |
"learning_rate": 2.7718537898066833e-06,
|
331 |
+
"logits/chosen": 79.29591369628906,
|
332 |
+
"logits/rejected": 79.25688171386719,
|
333 |
+
"logps/chosen": -34.00603485107422,
|
334 |
+
"logps/rejected": -32.55823516845703,
|
335 |
+
"loss": 6.2997,
|
336 |
+
"rewards/accuracies": 0.675000011920929,
|
337 |
+
"rewards/chosen": 0.1493411809206009,
|
338 |
+
"rewards/margins": 0.16230711340904236,
|
339 |
+
"rewards/rejected": -0.012965935282409191,
|
340 |
"step": 200
|
341 |
},
|
342 |
{
|
343 |
"epoch": 0.52,
|
344 |
+
"eval_logits/chosen": 98.92082977294922,
|
345 |
+
"eval_logits/rejected": 98.90645599365234,
|
346 |
+
"eval_logps/chosen": -32.54380416870117,
|
347 |
+
"eval_logps/rejected": -36.24779510498047,
|
348 |
+
"eval_loss": 7.145691394805908,
|
349 |
+
"eval_rewards/accuracies": 0.5402824282646179,
|
350 |
+
"eval_rewards/chosen": -0.02012600749731064,
|
351 |
+
"eval_rewards/margins": 0.036127228289842606,
|
352 |
+
"eval_rewards/rejected": -0.056253232061862946,
|
353 |
+
"eval_runtime": 104.0138,
|
354 |
+
"eval_samples_per_second": 3.298,
|
355 |
"eval_steps_per_second": 0.413,
|
356 |
"step": 200
|
357 |
},
|
358 |
{
|
359 |
"epoch": 0.55,
|
360 |
+
"grad_norm": 1192.0,
|
361 |
"learning_rate": 2.5453962426402006e-06,
|
362 |
+
"logits/chosen": 81.78749084472656,
|
363 |
+
"logits/rejected": 81.68329620361328,
|
364 |
+
"logps/chosen": -33.50835037231445,
|
365 |
+
"logps/rejected": -35.296775817871094,
|
366 |
+
"loss": 5.7897,
|
367 |
+
"rewards/accuracies": 0.6625000238418579,
|
368 |
+
"rewards/chosen": 0.08747398853302002,
|
369 |
+
"rewards/margins": 0.14960959553718567,
|
370 |
+
"rewards/rejected": -0.062135614454746246,
|
371 |
"step": 210
|
372 |
},
|
373 |
{
|
374 |
"epoch": 0.57,
|
375 |
+
"grad_norm": 382.0,
|
376 |
"learning_rate": 2.3185646976551794e-06,
|
377 |
+
"logits/chosen": 83.9058609008789,
|
378 |
+
"logits/rejected": 83.9858627319336,
|
379 |
+
"logps/chosen": -31.16668128967285,
|
380 |
+
"logps/rejected": -30.917465209960938,
|
381 |
+
"loss": 6.2677,
|
382 |
+
"rewards/accuracies": 0.699999988079071,
|
383 |
+
"rewards/chosen": 0.15645496547222137,
|
384 |
+
"rewards/margins": 0.1941571831703186,
|
385 |
+
"rewards/rejected": -0.03770221397280693,
|
386 |
"step": 220
|
387 |
},
|
388 |
{
|
389 |
"epoch": 0.6,
|
390 |
+
"grad_norm": 688.0,
|
391 |
"learning_rate": 2.0932279108998323e-06,
|
392 |
+
"logits/chosen": 81.14820098876953,
|
393 |
+
"logits/rejected": 81.19589233398438,
|
394 |
+
"logps/chosen": -32.21388626098633,
|
395 |
+
"logps/rejected": -34.05015182495117,
|
396 |
+
"loss": 6.1846,
|
397 |
+
"rewards/accuracies": 0.6000000238418579,
|
398 |
+
"rewards/chosen": 0.11837179958820343,
|
399 |
+
"rewards/margins": 0.12597264349460602,
|
400 |
+
"rewards/rejected": -0.007600849959999323,
|
401 |
"step": 230
|
402 |
},
|
403 |
{
|
404 |
"epoch": 0.62,
|
405 |
+
"grad_norm": 764.0,
|
406 |
"learning_rate": 1.8712423238279358e-06,
|
407 |
+
"logits/chosen": 83.51747131347656,
|
408 |
+
"logits/rejected": 83.77994537353516,
|
409 |
+
"logps/chosen": -30.83194351196289,
|
410 |
+
"logps/rejected": -31.766277313232422,
|
411 |
+
"loss": 3.8985,
|
412 |
+
"rewards/accuracies": 0.7875000238418579,
|
413 |
+
"rewards/chosen": 0.17122383415699005,
|
414 |
+
"rewards/margins": 0.21807047724723816,
|
415 |
+
"rewards/rejected": -0.0468466654419899,
|
416 |
"step": 240
|
417 |
},
|
418 |
{
|
419 |
"epoch": 0.65,
|
420 |
+
"grad_norm": 528.0,
|
421 |
"learning_rate": 1.6544367689701824e-06,
|
422 |
+
"logits/chosen": 82.23722076416016,
|
423 |
+
"logits/rejected": 82.29248809814453,
|
424 |
+
"logps/chosen": -26.981786727905273,
|
425 |
+
"logps/rejected": -29.9254093170166,
|
426 |
+
"loss": 5.8549,
|
427 |
+
"rewards/accuracies": 0.675000011920929,
|
428 |
+
"rewards/chosen": 0.122822105884552,
|
429 |
+
"rewards/margins": 0.12775607407093048,
|
430 |
+
"rewards/rejected": -0.0049339523538947105,
|
431 |
"step": 250
|
432 |
},
|
433 |
{
|
434 |
"epoch": 0.68,
|
435 |
+
"grad_norm": 744.0,
|
436 |
"learning_rate": 1.4445974030621963e-06,
|
437 |
+
"logits/chosen": 79.6116943359375,
|
438 |
+
"logits/rejected": 79.72075653076172,
|
439 |
+
"logps/chosen": -30.516536712646484,
|
440 |
+
"logps/rejected": -36.090003967285156,
|
441 |
+
"loss": 4.7707,
|
442 |
+
"rewards/accuracies": 0.6875,
|
443 |
+
"rewards/chosen": 0.20889195799827576,
|
444 |
+
"rewards/margins": 0.20656757056713104,
|
445 |
+
"rewards/rejected": 0.0023243515752255917,
|
446 |
"step": 260
|
447 |
},
|
448 |
{
|
449 |
"epoch": 0.7,
|
450 |
+
"grad_norm": 348.0,
|
451 |
"learning_rate": 1.243452991757889e-06,
|
452 |
+
"logits/chosen": 79.08982849121094,
|
453 |
+
"logits/rejected": 79.111083984375,
|
454 |
+
"logps/chosen": -30.892492294311523,
|
455 |
+
"logps/rejected": -31.574138641357422,
|
456 |
+
"loss": 4.8727,
|
457 |
+
"rewards/accuracies": 0.762499988079071,
|
458 |
+
"rewards/chosen": 0.1772860586643219,
|
459 |
+
"rewards/margins": 0.1910295933485031,
|
460 |
+
"rewards/rejected": -0.013743534684181213,
|
461 |
"step": 270
|
462 |
},
|
463 |
{
|
464 |
"epoch": 0.73,
|
465 |
+
"grad_norm": 700.0,
|
466 |
"learning_rate": 1.0526606671603523e-06,
|
467 |
+
"logits/chosen": 81.58210754394531,
|
468 |
+
"logits/rejected": 81.3693618774414,
|
469 |
+
"logps/chosen": -31.06771469116211,
|
470 |
+
"logps/rejected": -29.591297149658203,
|
471 |
+
"loss": 7.3884,
|
472 |
+
"rewards/accuracies": 0.6875,
|
473 |
+
"rewards/chosen": 0.14843526482582092,
|
474 |
+
"rewards/margins": 0.13329505920410156,
|
475 |
+
"rewards/rejected": 0.015140200033783913,
|
476 |
"step": 280
|
477 |
},
|
478 |
{
|
479 |
"epoch": 0.75,
|
480 |
+
"grad_norm": 428.0,
|
481 |
"learning_rate": 8.737922755071455e-07,
|
482 |
+
"logits/chosen": 81.82354736328125,
|
483 |
+
"logits/rejected": 81.73272705078125,
|
484 |
+
"logps/chosen": -33.02164077758789,
|
485 |
+
"logps/rejected": -32.17236328125,
|
486 |
+
"loss": 4.7078,
|
487 |
+
"rewards/accuracies": 0.6875,
|
488 |
+
"rewards/chosen": 0.21652066707611084,
|
489 |
+
"rewards/margins": 0.24762043356895447,
|
490 |
+
"rewards/rejected": -0.03109976276755333,
|
491 |
"step": 290
|
492 |
},
|
493 |
{
|
494 |
"epoch": 0.78,
|
495 |
+
"grad_norm": 704.0,
|
496 |
"learning_rate": 7.08321427484816e-07,
|
497 |
+
"logits/chosen": 77.42338562011719,
|
498 |
+
"logits/rejected": 77.51200866699219,
|
499 |
+
"logps/chosen": -32.24098205566406,
|
500 |
+
"logps/rejected": -29.014667510986328,
|
501 |
+
"loss": 6.0651,
|
502 |
+
"rewards/accuracies": 0.737500011920929,
|
503 |
+
"rewards/chosen": 0.2329459935426712,
|
504 |
+
"rewards/margins": 0.21547456085681915,
|
505 |
+
"rewards/rejected": 0.017471451312303543,
|
506 |
"step": 300
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.78,
|
510 |
+
"eval_logits/chosen": 99.00227355957031,
|
511 |
+
"eval_logits/rejected": 98.98667907714844,
|
512 |
+
"eval_logps/chosen": -32.43557357788086,
|
513 |
+
"eval_logps/rejected": -36.024574279785156,
|
514 |
+
"eval_loss": 7.648494243621826,
|
515 |
+
"eval_rewards/accuracies": 0.5390365719795227,
|
516 |
+
"eval_rewards/chosen": 0.0015206302050501108,
|
517 |
+
"eval_rewards/margins": 0.01312990766018629,
|
518 |
+
"eval_rewards/rejected": -0.011609279550611973,
|
519 |
+
"eval_runtime": 104.0603,
|
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": 1200.0,
|
527 |
"learning_rate": 5.576113578589035e-07,
|
528 |
+
"logits/chosen": 84.3228759765625,
|
529 |
+
"logits/rejected": 84.34573364257812,
|
530 |
+
"logps/chosen": -29.936237335205078,
|
531 |
+
"logps/rejected": -32.30895233154297,
|
532 |
+
"loss": 5.3743,
|
533 |
+
"rewards/accuracies": 0.75,
|
534 |
+
"rewards/chosen": 0.18545907735824585,
|
535 |
+
"rewards/margins": 0.2072969228029251,
|
536 |
+
"rewards/rejected": -0.02183784916996956,
|
537 |
"step": 310
|
538 |
},
|
539 |
{
|
540 |
"epoch": 0.83,
|
541 |
+
"grad_norm": 406.0,
|
542 |
"learning_rate": 4.229036944380913e-07,
|
543 |
+
"logits/chosen": 81.96125030517578,
|
544 |
+
"logits/rejected": 81.95556640625,
|
545 |
+
"logps/chosen": -30.652542114257812,
|
546 |
+
"logps/rejected": -28.881006240844727,
|
547 |
+
"loss": 5.1557,
|
548 |
+
"rewards/accuracies": 0.675000011920929,
|
549 |
+
"rewards/chosen": 0.18920376896858215,
|
550 |
+
"rewards/margins": 0.18150660395622253,
|
551 |
+
"rewards/rejected": 0.007697162684053183,
|
552 |
"step": 320
|
553 |
},
|
554 |
{
|
555 |
"epoch": 0.86,
|
556 |
+
"grad_norm": 576.0,
|
557 |
"learning_rate": 3.053082288996112e-07,
|
558 |
+
"logits/chosen": 79.27213287353516,
|
559 |
+
"logits/rejected": 79.31598663330078,
|
560 |
+
"logps/chosen": -29.210277557373047,
|
561 |
+
"logps/rejected": -32.73192596435547,
|
562 |
+
"loss": 5.7764,
|
563 |
+
"rewards/accuracies": 0.7124999761581421,
|
564 |
+
"rewards/chosen": 0.22846612334251404,
|
565 |
+
"rewards/margins": 0.2478782832622528,
|
566 |
+
"rewards/rejected": -0.019412150606513023,
|
567 |
"step": 330
|
568 |
},
|
569 |
{
|
570 |
"epoch": 0.88,
|
571 |
+
"grad_norm": 708.0,
|
572 |
"learning_rate": 2.0579377374915805e-07,
|
573 |
+
"logits/chosen": 83.35621643066406,
|
574 |
+
"logits/rejected": 83.36161041259766,
|
575 |
+
"logps/chosen": -32.25800323486328,
|
576 |
+
"logps/rejected": -33.48296356201172,
|
577 |
+
"loss": 5.3324,
|
578 |
+
"rewards/accuracies": 0.675000011920929,
|
579 |
+
"rewards/chosen": 0.19082824885845184,
|
580 |
+
"rewards/margins": 0.213682621717453,
|
581 |
+
"rewards/rejected": -0.022854382172226906,
|
582 |
"step": 340
|
583 |
},
|
584 |
{
|
585 |
"epoch": 0.91,
|
586 |
+
"grad_norm": 322.0,
|
587 |
"learning_rate": 1.2518018074041684e-07,
|
588 |
+
"logits/chosen": 82.49974822998047,
|
589 |
+
"logits/rejected": 82.50485229492188,
|
590 |
+
"logps/chosen": -32.731300354003906,
|
591 |
+
"logps/rejected": -33.191688537597656,
|
592 |
+
"loss": 4.9862,
|
593 |
+
"rewards/accuracies": 0.675000011920929,
|
594 |
+
"rewards/chosen": 0.18248799443244934,
|
595 |
+
"rewards/margins": 0.20246508717536926,
|
596 |
+
"rewards/rejected": -0.019977085292339325,
|
597 |
"step": 350
|
598 |
},
|
599 |
{
|
600 |
"epoch": 0.94,
|
601 |
+
"grad_norm": 520.0,
|
602 |
"learning_rate": 6.41315865106129e-08,
|
603 |
+
"logits/chosen": 83.8789291381836,
|
604 |
+
"logits/rejected": 83.89030456542969,
|
605 |
+
"logps/chosen": -28.570327758789062,
|
606 |
+
"logps/rejected": -31.54669761657715,
|
607 |
+
"loss": 5.4694,
|
608 |
+
"rewards/accuracies": 0.637499988079071,
|
609 |
+
"rewards/chosen": 0.19637688994407654,
|
610 |
+
"rewards/margins": 0.17574702203273773,
|
611 |
+
"rewards/rejected": 0.02062985673546791,
|
612 |
"step": 360
|
613 |
},
|
614 |
{
|
615 |
"epoch": 0.96,
|
616 |
+
"grad_norm": 370.0,
|
617 |
"learning_rate": 2.3150941078050325e-08,
|
618 |
+
"logits/chosen": 83.33052062988281,
|
619 |
+
"logits/rejected": 83.347900390625,
|
620 |
+
"logps/chosen": -31.880367279052734,
|
621 |
+
"logps/rejected": -35.12144470214844,
|
622 |
+
"loss": 7.7502,
|
623 |
+
"rewards/accuracies": 0.6499999761581421,
|
624 |
+
"rewards/chosen": 0.1991589069366455,
|
625 |
+
"rewards/margins": 0.20157304406166077,
|
626 |
+
"rewards/rejected": -0.0024141333997249603,
|
627 |
"step": 370
|
628 |
},
|
629 |
{
|
630 |
"epoch": 0.99,
|
631 |
+
"grad_norm": 552.0,
|
632 |
"learning_rate": 2.575864278703266e-09,
|
633 |
+
"logits/chosen": 77.32600402832031,
|
634 |
+
"logits/rejected": 77.20854187011719,
|
635 |
+
"logps/chosen": -29.861562728881836,
|
636 |
+
"logps/rejected": -28.112417221069336,
|
637 |
+
"loss": 5.5564,
|
638 |
+
"rewards/accuracies": 0.6499999761581421,
|
639 |
+
"rewards/chosen": 0.14291271567344666,
|
640 |
+
"rewards/margins": 0.13328143954277039,
|
641 |
+
"rewards/rejected": 0.009631268680095673,
|
642 |
"step": 380
|
643 |
},
|
644 |
{
|
645 |
"epoch": 1.0,
|
646 |
"step": 385,
|
647 |
"total_flos": 0.0,
|
648 |
+
"train_loss": 5.985443219271573,
|
649 |
+
"train_runtime": 2557.1257,
|
650 |
"train_samples_per_second": 1.204,
|
651 |
"train_steps_per_second": 0.151
|
652 |
}
|